Radar signal spectral clustering sorting method based on SOM anchor point extraction and graph fusion

Abstract

The invention relates to the technical field of signal processing, data characterization and classification, and discloses a radar signal spectral clustering sorting method based on SOM anchor point extraction and graph fusion, which comprises the following steps of: configuring a radar signal sorting cluster number and radar pulse parameters, obtaining a normalized radar pulse data set, and storing the normalized radar pulse data set; taking each normalized radar pulse in the radar pulse data set as a node, and constructing a KNN graph of the radar pulses; extracting anchor points of the normalized radar pulse data set based on SOM, calculating the similarity between the extracted SOM anchor points and all nodes to obtain a similarity matrix, and constructing an anchor map adjacent matrix based on the similarity matrix, namely a self-adaptive anchor map; performing weighted fusion on the KNN graph and the self-adaptive anchor graph to obtain a fusion graph; and performing spectral clustering sorting based on the radar signal sorting cluster number and the fusion image to obtain a sorting result. According to the method, the limitation that only distance information and density information can be utilized in a classic radar sorting clustering algorithm is broken through, and the sorting performance under the condition of complex radar pulse distribution is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal processing, data representation and classification, and in particular relates to a radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion. Background Art

[0002] Radar signal sorting (RSS) is one of the key technologies of the reconnaissance system. Its purpose is to separate the pulse sequence emitted by each radar from the received random aliased pulse stream. The specific process of RSS is: deinterleave and dealias the received radar pulse sequence, separate the pulse sequence from the same radiation source, and obtain the radiation source information. The acquired radiation source information can be used to assist in the assessment of the threat level of the radiation source, and the radiation source parameter information can also be sent to the jamming equipment to achieve electronic interference with the radiation source. However, with the emergence of multi-function radars with multiple working modes, radar signals with complex distribution forms have brought challenges to radar sorting in complex electromagnetic environments.

[0003] Radar signal sorting algorithms are implemented based on radar pulse description words (PDW), including direction of arrival (DOA), time of arrival (TOA), pulse width (PW), radio frequency (RF), and pulse amplitude (PA). Existing radar signal sorting algorithms can be mainly divided into two categories: PRI-based sorting methods and multi-parameter clustering sorting methods. Early single-parameter radar sorting algorithms were mainly based on pulse repetition interval (PRI) and its transformation parameters, including cumulative difference integrated function (CDIF), sequence difference histogram (SDIF), pulse repetition interval (PRI) transformation method, etc. This algorithm is highly dependent on the measurement accuracy of TOA parameters. The sorting accuracy is affected by pulse loss and false pulses caused by environmental noise and hardware equipment, and its robustness is poor. Therefore, unsupervised clustering algorithms based on multiple reliable parameters have gradually become mainstream, and clustering methods commonly used for radar signal sorting include: hierarchical clustering, fuzzy clustering, support vector clustering, spatial density clustering, K-Means clustering, etc. However, such methods only rely on the distance information and density information of the original features, and have limited ability to mine the characteristic information of pulses. In the case of overlapping parameters, the clustering effect is not good, and a breakthrough needs to be sought from a new perspective.

[0004] In recent years, with the rapid development of graph representation learning, spectral clustering, an unsupervised clustering algorithm based on graphs, has performed well in face recognition, data analysis, and image clustering. The algorithm aims to explore local graph structure data, and can optimally divide data samples of any shape to obtain good clustering accuracy. Graph-based clustering methods must first build a graph model. The most commonly used similarity graphs in the literature are k-nearest neighbors, ε neighborhood, and fully connected graphs. The main idea of ​​existing methods is to use parameters such as k, ε, and σ to represent the local features of the data set.

[0005] A recent study showed that clustering results depend on the choice of similarity graph and its parameters. However, for datasets with arbitrary cluster shapes, varying densities, and unbalanced clusters, how to set appropriate parameters and improve clustering performance becomes a challenging task. Summary of the invention

[0006] The purpose of the present invention is to propose a spectral clustering sorting method based on SOM anchor point extraction and graph fusion to address the technical status quo that the existing radar pulse sequence clustering sorting has limited ability to mine pulse feature information and poor clustering effect under parameter overlapping. Firstly, the K Nearest Neighbors (KNN) algorithm and self-organizing maps (SOM) network are used to construct an adaptive topological graph of the radar pulse sequence, and then the spectral clustering algorithm is used to deinterleave the pulse sequence to improve the spectral clustering performance.

[0007] To achieve the above objectives, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion, comprising the following steps:

[0009] S1. Configure the number of radar signal sorting clusters and radar pulse parameters, and construct a radar pulse data set based on the radar pulse parameters;

[0010] S2. normalizing the radar pulse parameters in the radar pulse data set to obtain a normalized radar pulse data set;

[0011] S3. Using each normalized radar pulse parameter in the radar pulse data set as a node, construct a KNN graph of the radar pulse;

[0012] S4. extracting anchor points of the normalized radar pulse data set based on the SOM, and constructing an adaptive anchor graph based on multiple anchor points; specifically comprising the following sub-steps:

[0013] S40. Assign a random value between [0, 1] to the weight vector of each neuron in the SOM output layer, and configure the number of iterations;

[0014] S41. Taking the normalized radar pulse data set as input, iteratively train the SOM until the number of iterations is completed, and the neuron weight vector finally obtained is the extracted SOM anchor point;

[0015] S42. Calculate the similarity between the SOM anchor point and all nodes to obtain a similarity matrix;

[0016] S43. Constructing an anchor graph adjacency matrix based on the similarity matrix, namely, an adaptive anchor graph;

[0017] S5. weightedly fuse the KNN graph and the adaptive anchor graph to obtain a fused graph;

[0018] S6. Perform spectral clustering sorting based on the radar signal sorting cluster number and fusion graph to obtain the sorting result.

[0019] As a possible implementation, the radar pulse parameters are: direction of arrival, carrier frequency and pulse width.

[0020] As a possible implementation method, S3 specifically includes the following sub-steps:

[0021] S30. Set the number of radar pulse parameters to N, and use the normalized radar pulse parameters as nodes;

[0022] S31. Take each node as the central node, traverse the remaining N-1 nodes, measure the similarity between nodes based on the Euclidean distance, find the K nodes closest to the central node, and construct a KNN graph adjacency matrix;

[0023] S32. The matrix values ​​of the corresponding positions of each node and its K adjacent nodes in the KNN graph adjacency matrix are set to 1 to obtain a directed graph adjacency matrix;

[0024] S33. Symmetrize the adjacency matrix of the directed graph and construct an undirected graph, which is the KNN graph of the radar pulse.

[0025] As a possible implementation manner, S41 specifically includes the following sub-steps:

[0026] S410. Configure the learning rate and neighborhood radius, and for the tth iteration, input the normalized radar pulse data set to the SOM, and calculate the Euclidean distance between each radar pulse parameter in the radar pulse data set and each weight vector;

[0027] S411. Find the neuron with the smallest Euclidean distance, which is the winning neuron. With the winning neuron as the center, the weight vectors of the neurons within the center neighborhood radius are modified based on the learning rate;

[0028] S412. If t is less than the number of iterations, return to S410 and continue the next iteration until t is equal to the number of iterations; if t is equal to the number of iterations, the SOM neuron weight vector obtained at this time is the extracted anchor point.

[0029] As a possible implementation method, the initial learning rate is greater than 0 and less than 1, and the learning rate of each iteration decreases with the number of iterations.

[0030] As a possible implementation method, the KNN graph adjacency matrix is ​​used to improve the stability of the anchor graph adjacency matrix. Weights are assigned to the KNN graph adjacency matrix, and the fusion graph is the sum of the KNN graph adjacency matrix multiplied by the weights and the anchor graph adjacency matrix.

[0031] As a possible implementation method, S42 is specifically as follows:

[0032] The Gaussian kernel function is used to calculate the similarity between the SOM anchor point and all nodes as follows:

[0033]

[0034] The similarity matrix is ​​constructed as follows:

[0035]

[0036] In the formula, x i represents the ith pulse node, c j represents the jth anchor point, M represents the number of anchor points, and w ij represents the similarity between the i-th graph node and the j-th anchor point, and σ represents the parameter controlling the width of the neighborhood.

[0037] As a possible implementation manner, S6 specifically includes the following sub-steps:

[0038] S60. Construct a fusion graph adjacency matrix based on the KNN graph adjacency matrix and the anchor graph adjacency matrix;

[0039] S61. Construct a degree matrix and a Laplacian matrix based on the fusion graph adjacency matrix, and calculate a normalized Laplacian matrix;

[0040] S62. Calculate the eigenvectors corresponding to the c smallest eigenvalues ​​of the normalized Laplace matrix and use them as elements to construct a feature matrix, where c is the number of sorted clusters;

[0041] S63. Each row of the feature matrix represents a sample. K-means clustering is performed on all samples to obtain the final cluster, which is the sorting result.

[0042] As a possible implementation method, taking the winning neuron as the center, the weight vectors of neurons within the central neighborhood radius are modified based on the learning rate, specifically:

[0043]

[0044] Among them, j * is the winning neuron, j is a neuron, j≠j * , X is the signal parameter set, W is the weight vector, t is the current iteration number, η(t) is the learning rate of the tth iteration, N g (t) is the neighborhood radius of the tth iteration.

[0045] In a second aspect, the present invention provides a radar signal spectrum clustering and sorting system based on SOM anchor point extraction and graph fusion, comprising:

[0046] A KNN graph construction unit is configured to configure the number of radar signal sorting clusters and radar pulse parameters, construct a radar pulse data set based on the radar pulse parameters, normalize the radar pulse parameters in the radar pulse data set, obtain the normalized radar pulse data set, and use each normalized radar pulse parameter in the radar pulse data set as a node to construct a KNN graph of the radar pulse;

[0047] The SOM anchor point extraction unit assigns a random value between [0, 1] to the weight vector of each neuron in the SOM output layer and configures the number of iterations; the normalized radar pulse data set is used as input, and the SOM is iteratively trained until the number of iterations is completed. The neuron weight vector finally obtained is the extracted SOM anchor point;

[0048] The adaptive anchor graph construction unit calculates the similarity between the SOM anchor point and all nodes to obtain a similarity matrix; based on the similarity matrix, an anchor graph adjacency matrix is ​​constructed, which is the adaptive anchor graph;

[0049] The fusion graph construction unit uses the KNN graph adjacency matrix to improve the stability of the anchor graph adjacency matrix, assigns weights to the KNN graph adjacency matrix, and the fusion graph is the sum of the KNN graph adjacency matrix multiplied by the weights and the anchor graph adjacency matrix;

[0050] The spectral clustering sorting unit constructs a fusion graph adjacency matrix based on the KNN graph adjacency matrix and the anchor graph adjacency matrix, constructs a degree matrix and a Laplace matrix based on the fusion graph adjacency matrix, calculates the standardized Laplace matrix, calculates the eigenvectors corresponding to the c smallest eigenvalues ​​of the standardized Laplace matrix, and uses them as elements to construct a feature matrix, where c is the number of sorting clusters, and each row of the feature matrix represents a sample. K-means clustering is performed on all samples to obtain the final cluster cluster, which is the sorting result.

[0051] Compared with the prior art, the beneficial effects of the present invention are:

[0052] 1. The radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion provided by the present invention makes full use of the function of SOM to capture the spatial topological structure of input samples, and obtains more evenly distributed and representative anchor points. Compared with the randomly selected anchor points in the prior art, it can better reflect the graph structure characteristics of the input samples.

[0053] 2. The radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion provided by the present invention introduces a clustering method based on graph theory for radar signal sorting, and takes the topological structure of the pulse as one of the features to break through the limitation that the classic radar sorting clustering algorithm can only use distance information and density information, thereby improving the sorting performance under complex radar pulse distribution conditions.

[0054] 3. The radar signal spectral clustering and sorting method based on SOM anchor point extraction and graph fusion provided by the present invention fuses the anchor point graph and the KNN graph to construct a fusion graph, and uses the KNN graph to correct the anchor point graph, thereby further improving the clustering performance of spectral clustering. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0056] Figure 1 It is a flow chart of a radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion in an embodiment of the present invention;

[0057] Figure 2 A flow chart of a method for constructing an adaptive anchor graph after extracting anchor points based on SOM in an embodiment of the present invention;

[0058] Figure 3 This is a diagram showing the sorting results of 4 types of radars using the K-Means-based sorting method in an embodiment of the present invention;

[0059] Figure 4 This is a diagram showing the sorting results of 4 types of radars using the DBSCAN-based sorting method in an embodiment of the present invention;

[0060] Figure 5 This is a diagram showing the sorting results of 4 types of radars using the SVM-based sorting method in an embodiment of the present invention;

[0061] Figure 6 This is a diagram of the sorting results of 4 types of radars using the sorting method based on the method of the present invention in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, words such as "first" and "second" are used to distinguish the same items or similar items with basically the same functions and effects. For example, the first threshold and the second threshold are only used to distinguish different thresholds, and their order is not limited. Those skilled in the art can understand that words such as "first" and "second" do not limit the quantity and execution order, and words such as "first" and "second" do not necessarily limit them to be different.

[0063] It should be noted that, in the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0064] In the present invention, "at least one" means one or more, and "plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. The following at least one item (items) or similar expressions thereof refer to any combination of these items, including any combination of single items (items) or plural items (items). For example, at least one item (items) of a, b or c can mean: a, b, c, the combination of a and b, the combination of a and c, the combination of b and c, or the combination of a, b and c, where a, b, c can be single or multiple.

[0065] With the emergence of multifunctional radars with multiple working modes, radar signals with complex distribution forms have brought challenges to radar signal sorting in complex electromagnetic environments. Aiming at the technical status that the existing radar signal spectrum clustering sorting algorithm has limited ability to mine pulse feature information and poor clustering effect under parameter overlap, the present invention proposes a radar signal spectrum clustering sorting method based on SOM anchor point extraction and graph fusion. Firstly, the K Nearest Neighbors (KNN) algorithm and Self-Organizing Maps (SOM) are used to construct an adaptive topological graph of radar pulse sequences, and then the spectrum clustering algorithm is used to deinterleave the pulse sequence to improve the spectrum clustering performance.

[0066] In the first aspect, the radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion provided by the present invention is as follows: Figure 1 , including the following steps:

[0067] S1. Configure the number of radar signal sorting clusters and radar pulse parameters, and construct a radar pulse data set based on the radar pulse parameters;

[0068] As a possible implementation, the radar pulse parameters are: direction of arrival, carrier frequency and pulse width.

[0069] S2. normalizing the radar pulse parameters in the radar pulse data set to obtain a normalized radar pulse data set;

[0070] As an example, the radar pulse parameters in the radar pulse data set are normalized, specifically: each radar pulse parameter is normalized to [0, 1];

[0071] S3. Using each normalized radar pulse parameter in the radar pulse data set as a node, construct a KNN graph of the radar pulse;

[0072] As a possible implementation method, S3 specifically includes the following sub-steps:

[0073] S30. Set the number of radar pulse parameters to N, and use the normalized radar pulse parameters as nodes;

[0074] S31. Take each node as the central node, traverse the remaining N-1 nodes, measure the similarity between nodes based on the Euclidean distance, find the K nodes closest to the central node, and construct a KNN graph adjacency matrix;

[0075] S32. The matrix values ​​of the corresponding positions of each node and its K adjacent nodes in the KNN graph adjacency matrix are set to 1 to obtain a directed graph adjacency matrix;

[0076] S33. Symmetrize the adjacency matrix of the directed graph and construct an undirected graph, which is the KNN graph of the radar pulse.

[0077] S4. Extract anchor points of the normalized radar pulse dataset based on SOM, and construct an adaptive anchor graph based on the extracted multiple anchor points; see Figure 2 , specifically including the following sub-steps:

[0078] S40. Assign a random value between [0, 1] to the weight vector of each neuron in the SOM output layer, and configure the number of iterations;

[0079] S41. Taking the normalized radar pulse data set as input, iteratively train the SOM until the number of iterations is completed, and the neuron weight vector finally obtained is the extracted SOM anchor point;

[0080] As a possible implementation manner, S41 specifically includes the following sub-steps:

[0081] S410. For the tth iteration, the normalized radar pulse data set is input to the SOM, and the Euclidean distance between each radar pulse parameter in the radar pulse data set and each weight vector is calculated;

[0082] S411. Find the neuron with the smallest Euclidean distance, which is the winning neuron. With the winning neuron as the center, the weight vectors of the neurons within the center neighborhood radius are modified based on the learning rate;

[0083] As a possible implementation method, the weight vectors of neurons within the central neighborhood radius are corrected with the winning neuron as the center, specifically:

[0084]

[0085] Among them, j * is the winning neuron, j is a neuron, j≠j* , X is the signal parameter set, W is the weight vector, t is the current iteration number, η(t) is the learning rate of the tth iteration, N g (t) is the neighborhood radius of the tth iteration.

[0086] S412. Update the size N of the winning neighborhood radius g (t) and learning rate η(t), the domain radius and learning rate function select exponential functions, specifically:

[0087] N g (t) = INT[N g (0)·exp(-t / (5·T))]

[0088] η(t)=η(0)·exp(-t / (5·T))

[0089] Where INT[·] is the rounding function and T is the total number of iterations.

[0090] Let t = t + 1, return to S410, when t = T, the SOM network stops running. The SOM neuron weight vector obtained at this time is the extracted anchor point.

[0091] It should be noted that in the first iteration, the weight vector of the neurons within the initial radius of the central neighborhood is corrected based on the initial learning rate. After each correction, the neighborhood radius and learning rate are updated to enter the next iteration until the configured number of iterations is completed.

[0092] As a possible implementation method, the initial learning rate is greater than 0 and less than 1, and the learning rate of each iteration decreases with the number of iterations.

[0093] S42. Calculate the similarity between the SOM anchor point and all nodes to obtain a similarity matrix;

[0094] As a possible implementation method, S42 is specifically as follows:

[0095] The Gaussian kernel function is used to calculate the similarity between the SOM anchor point and all nodes as follows:

[0096]

[0097] The similarity matrix is ​​constructed as follows:

[0098]

[0099] In the formula, x i represents the i-th graph node, c j represents the jth anchor point, M represents the number of anchor points, and w ijrepresents the similarity between the i-th graph node and the j-th anchor point, and σ represents the parameter controlling the width of the neighborhood.

[0100] S43. Constructing an anchor graph adjacency matrix based on the similarity matrix, namely, an adaptive anchor graph;

[0101] S5. weightedly fuse the KNN graph and the adaptive anchor graph to obtain a fused graph;

[0102] As a possible implementation method, the KNN graph adjacency matrix is ​​used to improve the stability of the anchor graph adjacency matrix. Weights are assigned to the KNN graph adjacency matrix, and the fusion graph is the sum of the KNN graph adjacency matrix multiplied by the weights and the anchor graph adjacency matrix.

[0103] S6. Perform spectral clustering sorting based on the radar signal sorting cluster number and fusion graph to obtain the sorting result.

[0104] As a possible implementation manner, S6 specifically includes the following sub-steps:

[0105] S60. Construct a fusion graph adjacency matrix based on the KNN graph adjacency matrix and the anchor graph adjacency matrix;

[0106] S61. Construct a degree matrix and a Laplacian matrix based on the fusion graph adjacency matrix, and calculate a normalized Laplacian matrix;

[0107] S62. Calculate the eigenvectors corresponding to the c smallest eigenvalues ​​of the normalized Laplace matrix and use them as elements to construct a feature matrix, where c is the number of sorted clusters;

[0108] S63. Each row of the feature matrix represents a sample. K-means clustering is performed on all samples to obtain the final cluster, which is the sorting result.

[0109] In a second aspect, the present invention provides a radar signal spectrum clustering and sorting system based on SOM anchor point extraction and graph fusion, comprising:

[0110] A KNN graph construction unit is configured to configure the number of radar signal sorting clusters and radar pulse parameters, construct a radar pulse data set based on the radar pulse parameters, normalize the radar pulse parameters in the radar pulse data set, obtain the normalized radar pulse data set, and use each normalized radar pulse parameter in the radar pulse data set as a node to construct a KNN graph of the radar pulse;

[0111] The SOM anchor point extraction unit assigns a random value between [0, 1] to the weight vector of each neuron in the SOM output layer and configures the number of iterations; the normalized radar pulse data set is used as input, and the SOM is iteratively trained until the number of iterations is completed. The neuron weight vector finally obtained is the extracted SOM anchor point;

[0112] The adaptive anchor graph construction unit calculates the similarity between the SOM anchor point and all nodes to obtain a similarity matrix; based on the similarity matrix, an anchor graph adjacency matrix is ​​constructed, which is the adaptive anchor graph;

[0113] The fusion graph construction unit uses the KNN graph adjacency matrix to improve the stability of the anchor graph adjacency matrix, assigns weights to the KNN graph adjacency matrix, and the fusion graph is the sum of the KNN graph adjacency matrix multiplied by the weights and the anchor graph adjacency matrix;

[0114] The spectral clustering sorting unit constructs a fusion graph adjacency matrix based on the KNN graph adjacency matrix and the anchor graph adjacency matrix, constructs a degree matrix and a Laplace matrix based on the fusion graph adjacency matrix, calculates the standardized Laplace matrix, calculates the eigenvectors corresponding to the c smallest eigenvalues ​​of the standardized Laplace matrix, and uses them as elements to construct a feature matrix, where c is the number of sorting clusters, and each row of the feature matrix represents a sample. K-means clustering is performed on all samples to obtain the final cluster cluster, which is the sorting result.

[0115] To facilitate understanding of the technical solution of the present application, further description is given below in conjunction with specific embodiments.

[0116] Taking four types of multi-function radars as an example, each type of radar has one to three working modes. The simulation signal set of the multi-function radar is configured. The signal set contains three PDW parameters: direction of arrival (DOA), carrier frequency (RF) and pulse width (PW). The DOA, RF and PW corresponding to each type of radar have multiple modulation types, such as: conventional, slip, staggered, jitter, etc. The same parameter types overlap to varying degrees between different radars or working modes. See Table 1. For example, for the three working modes of Radar 2, the DOA variation range is consistent, and the overlap rates of PW and RF parameters are 40% and 70%, respectively. For Radar 2 and Radar 3, the overlap rates of RF and DOA parameters are 64.29% and 33.33%, respectively.

[0117] Table 1 Multi-function radar simulation data set

[0118]

[0119] First, construct the KNN graph of the radar pulse sequence. The specific steps are as follows:

[0120] S100. Normalize each radar pulse parameter in the radar pulse data set to [0, 1]. The radar pulse data set can be expressed as: N and d represent the number of pulses and feature dimension respectively. The normalization process is defined by the following formula;

[0121]

[0122] in, represents the i-th pulse xi The j-th dimension feature of .

[0123] S101. Take the normalized radar pulse parameters as nodes, take each node as a graph node, traverse the remaining N-1 nodes, measure the similarity between nodes based on the Euclidean distance, find the K most adjacent nodes, and construct the KNN graph adjacency matrix A k , where the Euclidean distance is defined by the following formula:

[0124]

[0125] Among them, x i1 and x i2 Represent nodes i1 and i2 respectively;

[0126] S102. Define that a graph node is associated with its K nearest neighbors; that is, the KNN graph adjacency matrix A k The corresponding element in the is set to 1. For example, if node i1 is associated with node i2, then let a i1i2 =1, thus we get a directed graph, symmetric KNN graph adjacency matrix A k , let the element a in the symmetrical position i1 =a i2 , construct an undirected graph. After the above construction process, the radar pulse data set is converted into an undirected graph structure data, that is, G = (X, A).

[0127] The specific implementation steps of extracting anchor points based on SOM and constructing an adaptive anchor map based on anchor points are as follows:

[0128] S103. Assume that there are M neurons randomly distributed in the output layer space of the SOM, assign a random value between [0, 1] to the weight vector of each neuron in the output layer, and determine the weight matrix W = [w 1 , w 2 , ..., w N ] T , i = 1, 2, ..., N, establish the initial radius N(0) of the neighborhood radius N(t), determine the initial learning rate η(0), 0<η(0)<1 and the number of iterations T, where t represents the tth iteration;

[0129] S104. Input the normalized radar pulse data set X, and calculate the Euclidean distance d between each pulse parameter and each weight vector in the radar pulse data set = || w i -X|| 2 , i = 1, 2, ..., N;

[0130] S105. Find the neuron with the smallest Euclidean distance d and determine it as the winning neuron j * , Taking the winning neuron as the center, the weight vectors of the neurons within the center neighborhood radius are modified:

[0131]

[0132] Among them, j * is the winning neuron, j is a neuron, j≠j * , X is the signal parameter set, W is the weight vector, t is the current iteration number, η(t) is the learning rate of the tth iteration, N g (t) is the neighborhood radius of the tth iteration;

[0133] S106. Update the size N of the winning neighborhood radius g (t) and learning rate η(t), the domain radius and learning rate function select exponential functions, specifically:

[0134] N g (t) = INT[N g (0)·exp(-t / (5·T))]

[0135] η(t)=η(0)·exp(-t / (5·T))

[0136] Let t = t + 1, return to step S104 and continue training until t = T, and the training ends;

[0137] S107. After the training is completed, the weight vectors of M neurons are finally obtained and used as anchor points, that is, the extraction of M anchor points is completed.

[0138] This method makes full use of the ability of SOM to capture the spatial topological structure of input samples and can obtain more evenly distributed and representative anchor points.

[0139] The adaptive anchor map is constructed based on M anchor points. The specific implementation steps are as follows:

[0140] S108. Use the Gaussian kernel function to calculate the similarity between the SOM anchor point and all nodes. The formula is as follows:

[0141]

[0142] The similarity matrix is ​​constructed as follows:

[0143]

[0144] In the formula, x i represents the ith pulse node, c j represents the jth anchor point, M represents the number of anchor points, and w ij represents the similarity between the i-th graph node and the j-th anchor point, and σ represents the parameter controlling the width of the neighborhood.

[0145] S109. Constructing anchor graph adjacency matrix based on similarity matrix: A a =WΛ -z W T , where the diagonal matrix Λ is defined as

[0146] S110. Using KNN graph adjacency matrix A k Improved anchor graph adjacency matrix A a The stability of A k The weight is α, and the KNN graph is weightedly fused with the adaptive anchor graph to obtain the fusion graph as follows:

[0147] A=αA k +A a

[0148] In the specific implementation, based on experience, the weights are preset in the weight interval [0, 1], and the sorting results are obtained by experiment. Then, the preset weights are increased by a step size of 0.1, and the weights obtained each time are experimented. The sorting results obtained by the experiment are compared, and the weight with the best result is selected as the given A. k The weight α.

[0149] S111.A is the adjacency matrix of the fused graph, and the degree matrix D is constructed as follows:

[0150]

[0151] Degree matrix D∈R N×N is a diagonal matrix, d ii Represents the i-th element of the matrix;

[0152] S112. Construct the Laplace matrix as follows:

[0153] L=DA

[0154] S113. Calculate the normalized Laplace matrix, which is defined as follows:

[0155]

[0156] S114. Calculation The eigenvectors corresponding to the c smallest eigenvalues ​​of the eigenvalues ​​are composed of the characteristic matrix Q = [q 1 ,q 2 , ..., q c ]∈R N×k , where c is the number of sorted clusters;

[0157] S115. Each row of the feature matrix Q represents a sample, and a K-means clustering algorithm is performed on the N samples to obtain clusters.

[0158] The final result of radar signal sorting is given by the result of spectral clustering, which includes the cluster label corresponding to each pulse, that is, the radar category information of each pulse.

[0159] The sorting results obtained by the present invention are compared with some existing classical and other advanced clustering methods, including K-Means, DBSCAN, PSO-based K-means (i.e., ABC-K, PSO-K), SOFM, and KNN graph-based spectral clustering (KNN-SC). The comparison results are shown in Table 2:

[0160] Table 2 Experimental results of different methods

[0161]

[0162] As can be seen from the table, the present invention has achieved the best performance, with almost the highest sorting accuracy for four types of radar signals, and the average sorting accuracy reaches 93.01%. The sorting rates for radar 1, radar 2, radar 3 and radar 4 are 88.80%, 87.09%, 100% and 99.93% respectively. Obviously, K-Means, DBSCAN and SOFM perform poorly in terms of accuracy, with only 66.01%, 69.85% and 66.10%. These three types of methods only rely on the distance and density information of the original data, which are suitable for situations where the density and distance distributions of different data classes are significantly different. However, our data simulates four multifunctional radars with overlapping parameters, making the simulation closer to the actual scene of multifunctional radar signal sorting in a complex electromagnetic environment. Compared with KNN-SC, the average sorting accuracy of the present invention is improved by 21.08%, and the sorting accuracy for radar 2 and radar 4 is significantly improved, reaching 60.56% and 28.93%.

[0163] See also Figure 3 , the K-Means-based sorting method classifies radar 2 with two working modes as two radiation sources, resulting in "increased batches"; and when the parameters of the two radars overlap, they are completely sorted as one radar. Figure 4 , density clustering DNSCAN finally sorts 4 types of radars into only 2 types of radars; see Figure 5 , for the SVM algorithm, there is still the problem that radar 1 is sorted as radar 4 and radar 2 is sorted as radar 3; see Figure 6 , the present invention sorts the four types of radars almost correctly.

[0164] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art may understand and implement other variations of the disclosed embodiments by viewing the drawings, the disclosure, and the drawings, etc. In the specification, the word "comprising" does not exclude other components or steps, and "one" or "an" does not exclude multiple situations. A single processor or other unit can implement several functions listed in the specification. Certain measures are recorded in different embodiments, but this does not mean that these measures cannot be combined to produce good results.

[0165] Although the present invention has been described in conjunction with specific features and embodiments thereof, it is apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the present invention. Accordingly, this specification and the accompanying drawings are merely exemplary illustrations of the present invention and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present invention. Obviously, those skilled in the art may make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, the present invention is intended to include such modifications and variations if they fall within the scope of the present invention and its equivalents.

Claims

1. A radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion, characterized in that: The steps include: S1. Configure the number of radar signal sorting clusters and radar pulse parameters, and construct a radar pulse data set based on the radar pulse parameters; S2. normalizing the radar pulse parameters in the radar pulse data set to obtain a normalized radar pulse data set; S3. Using each normalized radar pulse parameter in the radar pulse data set as a node, constructing a KNN graph of the radar pulse; S4. extracting anchor points of the normalized radar pulse data set based on the SOM, and constructing an adaptive anchor graph based on multiple anchor points; specifically comprising the following sub-steps: S40. Assign a random value between [0, 1] to the weight vector of each neuron in the SOM output layer, and configure the number of iterations; S41. Taking the normalized radar pulse data set as input, iteratively training the SOM until the number of iterations is completed, and the neuron weight vector finally obtained is the extracted SOM anchor point; S42. Calculate the similarity between the SOM anchor point and all nodes to obtain a similarity matrix; S43. Constructing an anchor graph adjacency matrix based on the similarity matrix, that is, an adaptive anchor graph; S5. weighted fusion of the KNN graph and the adaptive anchor graph to obtain a fusion graph; S6. Perform spectral clustering sorting based on the radar signal sorting cluster number and the fusion graph to obtain a sorting result.

2. The radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion according to claim 1 is characterized in that: The radar pulse parameters are: arrival direction, carrier frequency and pulse width.

3. The radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion according to claim 1 is characterized in that: The S3 specifically includes the following sub-steps: S30. Setting the number of radar pulse parameters to N, with the normalized radar pulse parameters as nodes; S31. Take each node as the central node, traverse the remaining N-1 nodes, measure the similarity between nodes based on the Euclidean distance, find the K nodes closest to the central node, and construct a KNN graph adjacency matrix; S32. The matrix values ​​of the corresponding positions of each node and its K adjacent nodes in the KNN graph adjacency matrix are set to 1 to obtain a directed graph adjacency matrix; S33. Symmetrize the directed graph adjacency matrix and construct an undirected graph, which is the KNN graph of the radar pulse.

4. The radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion according to claim 1 is characterized in that: The S41 specifically includes the following sub-steps: S410. Configure the learning rate and the neighborhood radius, and for the tth iteration, input the normalized radar pulse data set to the SOM, and calculate the Euclidean distance between each radar pulse parameter in the radar pulse data set and each weight vector; S411. Find the neuron with the smallest Euclidean distance, which is the winning neuron. With the winning neuron as the center, the weight vectors of the neurons within the center neighborhood radius are modified based on the learning rate; S412. If t is less than the number of iterations, return to S410 and continue with the next iteration until t is equal to the number of iterations; If t is equal to the number of iterations, the SOM neuron weight vector obtained at this time is the extracted anchor point.

5. The radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion according to claim 4 is characterized in that: The initial learning rate is greater than 0 and less than 1, and the learning rate of each iteration decreases with the number of iterations.

6. The radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion according to claim 3 is characterized in that: The KNN graph adjacency matrix is ​​used to improve the stability of the anchor graph adjacency matrix, and weights are assigned to the KNN graph adjacency matrix. The fusion graph is the sum of the KNN graph adjacency matrix multiplied by the weights and the anchor graph adjacency matrix.

7. The radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion according to claim 1 is characterized in that: The S42 is specifically: The Gaussian kernel function is used to calculate the similarity between the SOM anchor point and all nodes as follows: The similarity matrix is ​​constructed as follows: In the formula, x i represents the ith pulse node, c j represents the jth anchor point, M represents the number of anchor points, and w ij represents the similarity between the i-th graph node and the j-th anchor point, and σ represents the parameter controlling the width of the neighborhood.

8. The radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion according to claim 3 is characterized in that: The S6 specifically includes the following sub-steps: S60. Constructing a fusion graph adjacency matrix based on the KNN graph adjacency matrix and the anchor graph adjacency matrix; S61. Constructing a degree matrix and a Laplace matrix based on the fusion graph adjacency matrix, and calculating a standardized Laplace matrix; S62. Calculate the eigenvectors corresponding to the c smallest eigenvalues ​​of the standardized Laplace matrix, and use them as elements to construct a feature matrix, where c is the number of sorting clusters; S63. Each row of the feature matrix represents a sample, and K-means clustering is performed on all samples to obtain the final cluster, which is the sorting result.

9. The radar signal spectrum clustering and sorting method based on SOM anchor point extraction and graph fusion according to claim 4 is characterized in that: Taking the winning neuron as the center, the weight vectors of the neurons within the center neighborhood radius are modified, specifically: Among them, j * is the winning neuron, j is a neuron, j≠j * , X is the signal parameter set, W is the weight vector, t is the current iteration number, η(t) is the learning rate of the tth iteration, N g (t) is the neighborhood radius of the tth iteration.

10. A radar signal spectrum clustering and sorting system based on SOM anchor point extraction and graph fusion, characterized in that: include: a KNN graph construction unit, configuring the number of radar signal sorting clusters and radar pulse parameters, constructing a radar pulse data set based on the radar pulse parameters, normalizing the radar pulse parameters in the radar pulse data set, obtaining the normalized radar pulse data set, and constructing a KNN graph of the radar pulse using each normalized radar pulse parameter in the radar pulse data set as a node; The SOM anchor point extraction unit assigns a random value between [0, 1] to the weight vector of each neuron in the SOM output layer and configures the number of iterations; Taking the normalized radar pulse data set as input, iteratively training the SOM until the number of iterations is completed, and the neuron weight vector finally obtained is the extracted SOM anchor point; An adaptive anchor graph construction unit calculates the similarity between the SOM anchor point and all nodes to obtain a similarity matrix; an anchor graph adjacency matrix is ​​constructed based on the similarity matrix, which is an adaptive anchor graph; A fusion graph construction unit, using the KNN graph adjacency matrix to improve the stability of the anchor graph adjacency matrix, assigning weights to the KNN graph adjacency matrix, wherein the fusion graph is the sum of the KNN graph adjacency matrix multiplied by the weights and the anchor graph adjacency matrix; The spectral clustering sorting unit constructs a fusion graph adjacency matrix based on the KNN graph adjacency matrix and the anchor graph adjacency matrix, constructs a degree matrix and a Laplace matrix based on the fusion graph adjacency matrix, calculates a standardized Laplace matrix, calculates the eigenvectors corresponding to the c minimum eigenvalues ​​of the standardized Laplace matrix, and uses them as elements to construct a feature matrix, wherein c is the number of sorting clusters, each row of the feature matrix represents a sample, and performs K-means clustering on all samples to obtain a final cluster cluster, which is the sorting result.

Images