Radar unknown interference clustering method based on adaptive DBSCAN algorithm

Abstract

The invention discloses a radar unknown interference clustering method based on an adaptive DBSCAN algorithm, and the method improves a conventional DBSCAN algorithm, updates the maximum radius corresponding to a core point based on a preset minimum point number and the distance between the core point and a corresponding data point when the maximum radius is determined, obtains the updated maximum radius, and carries out the clustering of the unknown interference of a radar. Therefore, adaptive change of the maximum radius corresponding to each core point is realized, and updating of each core point is realized by updating the maximum radius. Therefore, the clustering time is shortened, the clustering efficiency is improved, and meanwhile, the clustering effect is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of radar signal processing, and in particular relates to a radar unknown interference clustering method based on an adaptive DBSCAN algorithm. Background Art

[0002] Radar is widely used in detection, communication, early warning, navigation and other scenarios, and has the characteristics of all-day and all-weather. However, with the increasing complexity of the electromagnetic environment, various active interference signals emerge in an endless stream, which seriously reduces the detection performance of radar. Therefore, effective anti-interference measures need to be taken to ensure the normal operation of radar. Interference identification, as the primary link of anti-interference, is of great significance in providing guidance for the effective implementation of anti-interference measures by accurately sensing the interference type.

[0003] The current mainstream radar active jammer identification methods are basically closed set identification for known jammers. Closed set identification means that in the case of radar jammers, only known targets are identified, and no identification judgment is made for unknown targets. However, the development of radar jammer technology has made radar jammer styles change with each passing day, and traditional closed set identification can no longer adapt to the existing actual situation, so the research on open set identification methods for radar active jammers is gradually gaining attention. In open set identification, after the unknown jammer is identified, the unknown jammer data is chaotic for the radar, and the radar cannot determine which of the unknown jammer signals are meaningful. Therefore, it is necessary to cluster the jammer signals obtained, and classify the jammer signals with similar features into one category, so as to facilitate the subsequent acquisition and recognition of new jammers in the environment. Clustering algorithms can be used to achieve clustering of sample points.

[0004] The traditional DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm clusters data points by inputting two parameters: maximum radius and minimum number of points. This requires the DBSCAN algorithm to optimize parameter selection through a series of methods during use, making the precise clustering of unknown interference time-consuming and inefficient. In addition, since the minimum number of points and maximum radius remain unchanged during the clustering iteration process, this limits the clustering effect of the traditional DBSCAN algorithm on complex data with uneven density. Summary of the invention

[0005] In order to solve the above problems existing in the prior art, the present invention provides a radar unknown interference clustering method based on an adaptive DBSCAN algorithm. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0006] The present invention provides a radar unknown interference clustering method based on an adaptive DBSCAN algorithm, which defines ε-domain, core point, direct density reachability, density reachability, density connection, noise point, boundary point and unvisited point respectively; the method comprises:

[0007] Preprocessing the obtained interference signal to obtain a time-frequency diagram and an RD diagram of the interference signal;

[0008] Using a radar active interference identification network model based on a multi-layer channel attention mechanism, feature extraction is performed on the time-frequency graph and the RD graph respectively to obtain a time-frequency feature graph and an RD feature graph;

[0009] Performing dimensionality reduction processing on the time-frequency feature graph and the RD feature graph to obtain feature parameters of several dimensions as data points;

[0010] Processing all data points based on the preset maximum radius and the preset minimum number of points to obtain a corresponding core point set;

[0011] Randomly select any core point in the core point set, and obtain an initial maximum radius based on the distance between the core point and the remaining data points and a preset minimum number of points;

[0012] The first core point in the core point set is updated according to the initial maximum radius; for each core point in the core point set except the first core point, each core point is taken as the current core point in turn, and the maximum radius of the cluster to which the current core point belongs is obtained based on the direct density reachable point corresponding to the current core point; the adjustment parameter is obtained based on the unvisited point corresponding to the current core point; the updated maximum radius of the current core point is obtained based on the maximum radius in the cluster and the adjustment parameter; the current core point is updated according to the updated maximum radius, so as to obtain an updated core point set;

[0013] Based on the updated core point set, a clustering result of each unknown interference in the interference signal is obtained.

[0014] In one embodiment of the present invention, ε-domain, core point, direct density reachable, density reachable, density connected, noise point, boundary point and unvisited point are defined respectively, including:

[0015] The ε-domain is defined as follows: for a data point p, a circle is set with p as the center and a preset maximum radius as the radius. The set of data points contained in the circle is called the ε-domain;

[0016] Definition of the core point: In the ε-domain of a data point, if the number of elements contained is greater than or equal to the preset minimum number of points, then the data point is called a core point;

[0017] Definition of direct density reachability: If data point x2 is contained in the ε-domain of data point x1, and data point x1 is a core point, then x2 is said to be directly density reachable from x1;

[0018] The density can be defined as follows: If there are data points x1, x2, ..., x N-1 , x N , and x i+1 By x i Direct density can be reached, then x N It is reachable by x1 density; where i∈[1,N], N≥2;

[0019] Definition of density connection: If for data point x1 and data point x2, there exists a core point x3 such that both x1 and x2 are density-reachable from the core point x3, then data point x1 and data point x2 are said to be density-connected;

[0020] Definition of the noise point: a point that does not belong to any cluster is a noise point;

[0021] Defining the boundary points: data points within the maximum radius of a core point but not core points are boundary points;

[0022] Definition of the unvisited point: a data point that is not defined as a core point, a boundary point, or a noise point is an unvisited point.

[0023] In one embodiment of the present invention, the step of preprocessing the obtained interference signal to obtain a time-frequency diagram and an RD diagram of the interference signal includes:

[0024] Performing short-time Fourier transform on the obtained interference signal within a pulse repetition period to obtain a time-frequency diagram of the interference signal;

[0025] Under the condition of pulse accumulation, moving target detection is performed on the interference signal to obtain an RD diagram of the interference signal.

[0026] In one embodiment of the present invention, a radar active interference identification network model based on a multi-layer channel attention mechanism is used to extract features from the time-frequency graph and the RD graph to obtain a time-frequency feature graph and an RD feature graph, including:

[0027] The time-frequency graph and RD graph are processed in sequence by the corresponding convolutional neural network module, attention mechanism module and multi-layer channel attention feature extraction module in the radar active interference identification network model, so as to obtain the time-frequency feature graph and RD feature graph.

[0028] In one embodiment of the present invention, any core point in the core point set is randomly selected, and an initial maximum radius is obtained based on the distance between the core point and the remaining data points and a preset minimum number of points, including:

[0029] Randomly select any core point in the core point set to obtain the distance between the core point and the remaining data points;

[0030] According to the distance between the core point and the rest of the data points, the mean is obtained. initialize and standard deviation std initialize ;

[0031] Using the first formula, the mean initialize and standard deviation std initialize Processing is performed to obtain the initial maximum radius; the first formula is as follows:

[0032] epsilon1=mean initialize +2*std initialize ;

[0033] Among them, epsilon1 is less than 0.1 times the first target distance, the first target distance is the maximum distance between the core point and the remaining data points, and the data points within the initial maximum radius corresponding to the core point are all unvisited points.

[0034] In one embodiment of the present invention, updating the first core point in the core point set according to the initial maximum radius includes:

[0035] The ε-domain corresponding to the first core point in the core point set is determined according to the initial maximum radius, thereby completing the update of the first core point in the core point set.

[0036] In one embodiment of the present invention, based on the current core point and the directly density reachable point corresponding to it, obtaining the maximum radius within the cluster to which the current core point belongs includes:

[0037] Accumulate the distance between the current core point and its corresponding directly reachable point to form a vector distances;

[0038] According to the vector distances, the mean of the vector distances is obtained. dis , standard deviation std dis and moment skewness ske dis ;

[0039] The second formula is used to calculate the mean dis , standard deviation std dis and moment skewness ske dis Processing is performed to obtain the maximum radius within the cluster; the second formula is as follows:

[0040]

[0041] In one embodiment of the present invention, obtaining adjustment parameters based on the current core point and the unvisited points corresponding thereto includes:

[0042] Based on the distance between the current core point and the second target twice as large as its surrounding distance, the mean is obtained. inVisit and standard deviation std inVisit ; The second target distance is the distance between the preset minimum number of unvisited points;

[0043] The third formula is used to calculate the mean inVisit and standard deviation std inVisit Processing is performed to obtain the adjustment parameter; the third formula is as follows:

[0044] b=mean inVisit +2*std inVisit .

[0045] In one embodiment of the present invention, obtaining the updated maximum radius of the current core point based on the maximum radius within the cluster and the adjustment parameter includes:

[0046] The fourth formula is used to process the maximum radius within the cluster and the adjustment parameter to obtain the updated maximum radius; the fourth formula is as follows:

[0047]

[0048] Among them, a represents the maximum radius within the cluster, and b represents the adjustment parameter.

[0049] In one embodiment of the present invention, based on the updated core point set, a clustering result of each unknown interference in the interference signal is obtained, including:

[0050] Select any core point in the updated core point set, and form a cluster for all density reachable points corresponding to the core point; form a cluster for all density reachable points corresponding to the unvisited points in the updated core point set; until all data points in the updated core point set are visited, based on all clusters, obtain the clustering results of each unknown interference in the interference signal.

[0051] Beneficial effects of the present invention:

[0052] In the solution provided by the present invention, the traditional DBSCAN algorithm is improved. When determining the maximum radius, the maximum radius corresponding to the core point is updated based on the preset minimum number of points and the distance between the core point and the data point corresponding to the core point, and an updated maximum radius is obtained, thereby realizing adaptive changes in the maximum radius corresponding to each core point, and then using the updated maximum radius to realize the update of each core point; thereby reducing the clustering time, improving the clustering efficiency, and enhancing the clustering effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram of the steps of a radar unknown interference clustering method based on an adaptive DBSCAN algorithm provided by an embodiment of the present invention;

[0054] Figure 2 A schematic diagram of the structure of an interference signal provided by an embodiment of the present invention;

[0055] Figure 3 A schematic diagram of core points and noise points provided by an embodiment of the present invention;

[0056] Figure 4a-4f A schematic diagram of clustering results of a radar unknown interference clustering method based on an adaptive DBSCAN algorithm provided in an embodiment of the present invention;

[0057] Figure 5a-5d A comparison chart of the results of the radar unknown interference clustering method based on the adaptive DBSCAN algorithm provided by an embodiment of the present invention and the traditional DBSCAN algorithm. DETAILED DESCRIPTION

[0058] The following will be combined with the accompanying drawings of the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] Radar interference open set identification enables the radar to determine whether the received interference signal is a known interference signal. However, for the radar at this moment, the unknown interference data is chaotic, and the radar cannot determine which of the unknown interference signals are meaningful signals. Therefore, it is necessary to cluster the obtained interference signals and classify interference signals with similar characteristics into one category to facilitate the subsequent acquisition and recognition of new interference in the environment.

[0060] In order to achieve the purpose of clustering various types of interference in interference signals, an embodiment of the present invention provides a radar unknown interference clustering method based on an adaptive DBSCAN algorithm.

[0061] The radar unknown interference clustering method based on the adaptive DBSCAN algorithm provided by the embodiment of the present invention defines ε-area, core point, direct density reachable, density reachable, density connected, noise point, boundary point and unvisited point respectively; the radar unknown interference clustering method, such as Figure 1 As shown, the following steps may be included:

[0062] S1, preprocessing the obtained interference signal to obtain a time-frequency diagram and an RD diagram of the interference signal;

[0063] S2, using the radar active interference identification network model based on the multi-layer channel attention mechanism, extracting features from the time-frequency graph and RD graph to obtain the time-frequency feature graph and RD feature graph;

[0064] S3, performing dimensionality reduction processing on the time-frequency feature map and the RD feature map to obtain feature parameters of several dimensions as data points;

[0065] S4, processing all data points based on a preset maximum radius and a preset minimum number of points to obtain a corresponding core point set;

[0066] S5, randomly selecting any core point in the core point set, and obtaining an initial maximum radius based on the distance between the core point and the remaining data points and a preset minimum number of points;

[0067] S6, updating the first core point in the core point set according to the initial maximum radius; for each core point in the core point set except the first core point, taking it as the current core point in turn, obtaining the maximum radius of the cluster to which the current core point belongs based on the direct density reachable points corresponding to the current core point; obtaining the adjustment parameter based on the current core point and the unvisited points corresponding to the current core point; obtaining the updated maximum radius of the current core point based on the maximum radius in the cluster and the adjustment parameter; updating the current core point according to the updated maximum radius, thereby obtaining an updated core point set;

[0068] S7, based on the updated core point set, obtain the clustering results of each unknown interference in the interference signal.

[0069] The radar unknown interference clustering method provided in the embodiment of the present invention confirms the maximum radius based on the preset minimum number of points and the distance between each core point and its corresponding data point, thereby realizing adaptive change of the maximum radius corresponding to each core point.

[0070] For ease of understanding, each step of the radar unknown interference clustering method provided in an embodiment of the present invention is introduced in detail below.

[0071] Define ε-areas, core points, directly density-reachable, density-reachable, density-connected, noise points, boundary points, and unvisited points, respectively, which can include:

[0072] Definition of ε-domain: For a data point p, a circle is set with p as the center and the maximum radius as the radius. The set of data points contained in the circle is called the ε-domain;

[0073] Definition of core point: In the ε-domain of a data point, if the number of elements contained is greater than or equal to the preset minimum number of points, then the data point is called a core point;

[0074] Definition of direct density reachability: If data point x2 is contained in the ε-domain of data point x1, and data point x1 is a core point, then x2 is said to be directly density reachable from x1;

[0075] The density is defined as: if there are data points x1, x2, ..., x N-1 , x N , and x i+1 By x i Direct density can be reached, then x N It is reachable by x1 density; where i∈[1,N], N≥2;

[0076] Definition of density connection: If for data point x1 and data point x2, there exists a core point x3 such that both x1 and x2 are density-reachable from core point x3, then data point x1 and data point x2 are said to be density-connected;

[0077] Definition of noise points: Points that do not belong to any cluster are noise points;

[0078] Define boundary points: Data points that are within the maximum radius of a core point but are not core points are boundary points;

[0079] Definition of unvisited points: Data points that are not defined as core points, boundary points, or noise points are unvisited points.

[0080] For S1, this may include:

[0081] S11, obtaining an interference signal;

[0082] For a schematic diagram of the structure of the interference signal in the embodiment of the present invention, please refer to Figure 2 .from Figure 2 It can be seen that the interference signal may include:

[0083] Noise amplitude modulation interference, noise frequency modulation interference, noise phase modulation interference, noise convolution interference, noise sweep frequency interference, distance deception interference, speed deception interference, distance wave gate interference, speed wave gate interference, dense false target interference, intermittent sampling forwarding interference, slice forwarding interference, spectrum dispersion interference, smart noise interference. Among the above-mentioned interference signals, the first 5 interference signals can be classified as suppression interference signals, and the last 9 interference signals can be classified as deception interference signals. Among them, noise phase modulation interference, noise sweep frequency interference, speed deception interference, distance wave gate interference, spectrum dispersion interference and smart noise interference can be unknown interference, and noise amplitude modulation interference, noise frequency modulation interference, noise convolution interference, distance deception interference, speed wave gate interference, dense false target interference, intermittent sampling forwarding interference and slice forwarding interference can be known interference.

[0084] S12, performing short-time Fourier transform on the interference signal within a pulse repetition period to obtain a time-frequency diagram of the interference signal;

[0085] S13, performing moving target detection on the interference signal under the condition of pulse accumulation to obtain an RD graph of the interference signal.

[0086] Specifically, the acquired interference signal is subjected to short-time Fourier transform within a pulse repetition period to obtain a time-frequency diagram of the interference signal within a pulse repetition period. At the same time, under the condition of pulse accumulation, the interference signal is subjected to moving target detection to obtain the RD (Range-Dopple) diagram of the interference signal.

[0087] For S2, this may include:

[0088] The time-frequency graph and RD graph are processed by the corresponding convolutional neural network module, attention mechanism module and multi-layer channel attention in the radar active interference identification network model respectively, so as to obtain the time-frequency feature graph and RD feature graph.

[0089] It can be understood that the radar active interference identification network model includes a time-frequency network branch and an RD network branch, and both the time-frequency network branch and the RD network branch include: a sequentially connected convolutional neural network module, an attention mechanism module, a multi-layer channel attention feature extraction module, and a fully connected module. The time-frequency network branch and the RD network branch in the radar active interference identification network model are used to process the time-frequency graph and the RD graph respectively to realize the feature extraction of the radar active interference signal and obtain the corresponding time-frequency feature graph and RD feature graph.

[0090] For S3, this can include:

[0091] Before the fully connected layer of the convolutional neural network, the high-dimensional time-frequency feature map and RD feature map obtained by feature extraction need to be reduced in dimension to obtain the corresponding feature parameters. The proportion of the corresponding feature parameters after dimensionality reduction is from high to low. Among the feature parameters after dimensionality reduction, the feature parameters of several dimensions that account for 95% are taken as data points.

[0092] Specifically, the dimensionality reduction method may include PCA dimensionality reduction.

[0093] For S4, this may include:

[0094] All data points are processed based on a preset maximum radius and a preset minimum number of points to obtain corresponding core points, and all core points are taken as a core point set.

[0095] It can be understood that the preset maximum radius can represent the radius for measuring data similarity during the clustering process. If the distance between two data is less than the preset maximum radius, the two data can be considered similar. A smaller preset maximum radius will produce a smaller cluster, and vice versa. A larger cluster. The preset minimum number of points can mean that if there are data points greater than or equal to the preset minimum number of points within a preset maximum radius, then these data points are considered to belong to the same cluster. The preset minimum number of points is equivalent to a threshold for measuring whether a cluster exists. If the threshold is large, the generated cluster will be more robust. However, if the preset minimum number of points is too large, it will also result in the inability to generate a cluster. For a schematic diagram of the core points and noise points provided in an embodiment of the present invention, please refer to Figure 3 .from Figure 3 It can be seen from the figure that the points that do not belong to any cluster are called unvisited points, with each core point as the center and the preset maximum radius as the radius. In the embodiment of the present invention, the maximum radius can be represented by epsilon, and the minimum number of points can be represented by MinPts.

[0096] For S5, this may include:

[0097] S51, randomly selecting any core point in the core point set, and obtaining the distance between the core point and the remaining data points;

[0098] S52, according to the distance between the core point and the rest of the data points, the mean is obtained. initialize and standard deviation std initialize ;

[0099] S53, using the first formula to calculate the mean initialize and standard deviation std initialize Processing is performed to obtain the initial maximum radius; the first formula is as follows:

[0100] epsilon1=mean initialize +2*std initialize ;

[0101] Among them, epsilon1 is less than 0.1 times the first target distance, the first target distance is the maximum distance between the core point and the remaining data points, and the data points within the initial maximum radius corresponding to the core point are all unvisited points.

[0102] Considering that the maximum radius needs to enclose at least a minimum number of data points near the current core point in the corresponding circle, an initial maximum radius is obtained by randomly selecting any core point and processing it through step S5 to ensure that the circle corresponding to the initial maximum radius with the current core point as the center can include data points greater than or equal to the minimum number of points.

[0103] For S6, this may include:

[0104] S61, updating the first core point in the core point set according to the initial maximum radius, may include:

[0105] The ε-domain corresponding to the first core point in the core point set is determined according to the initial maximum radius, thereby completing the update of the first core point in the core point set.

[0106] S62, for each core point in the core point set except the first core point, based on the direct density reachable points corresponding to the current core point, obtaining the maximum radius within the cluster to which the current core point belongs, may include:

[0107] Accumulate the distance between the current core point and its corresponding directly reachable point to form a vector distances;

[0108] Get the mean of vector distances according to vector distances dis , standard deviation std dis and moment skewness ske dis ;

[0109] Use the second formula to calculate the mean dis , standard deviation std dis and moment skewness ske dis Processing is performed to obtain the maximum radius within the cluster; the second formula is as follows:

[0110]

[0111] It can be understood that the moment skewness ske of the vector distances dis It reflects the distance distribution between the core point in the cluster to which the current core point belongs and its directly reachable density point. If the moment skewness is negative, it means that its distance distribution is left-skewed, with a shorter left tail, and the radius only needs a small expansion to circle most of the samples that should belong to this cluster; if the moment skewness is positive, a larger radius needs to be expanded to circle most of the samples that should belong to this cluster. Therefore, you can choose to use the moment skewness ske of the vector distants dis The coefficient of 3 in the second formula is to increase the slope of the Sigmiod function near 0, so as to sensitively control the change of the radius. Through step S62, the maximum radius of the cluster corresponding to each core point in the core point set except the first core point can be obtained.

[0112] S63, obtaining adjustment parameters based on the current core point and the unvisited points corresponding thereto, may include:

[0113] Based on the distance between the current core point and the second target twice as large as its surrounding distance, the mean is obtained. inVisit and standard deviation std inVisit ; The second target distance is the distance between the preset minimum number of unvisited points;

[0114] Use the third formula to calculate the mean inVisit and standard deviation std inVisit Processing is performed to obtain adjustment parameters; the third formula is as follows:

[0115] b=mean inVisit +2*std inVisit .

[0116] Step S63 can obtain the adjustment parameters corresponding to each core point in the core point set except the first core point. Based on each core point and its corresponding unvisited point, the adjustment parameters are obtained, which can be used to subsequently update the maximum radius corresponding to each core point.

[0117] S64, obtaining an updated maximum radius of the current core point based on the maximum radius within the cluster and the adjustment parameter, may include:

[0118] The fourth formula is used to process the maximum radius within the cluster and the adjustment parameter to obtain the updated maximum radius; the fourth formula is as follows:

[0119]

[0120] Among them, a represents the maximum radius within the cluster, and b represents the adjustment parameter.

[0121] The maximum radius corresponding to each core point is updated by using the maximum radius within the cluster and the adjustment parameter, so as to facilitate the subsequent update of the core point set.

[0122] S65, updating the current core point according to the updated maximum radius, thereby obtaining an updated core point set.

[0123] It can be understood that the embodiment of the present invention realizes adaptive change of the maximum radius parameter during the search process by accumulating the distance between the core points in the cluster to which the core point currently being searched belongs and its directly density reachable points, and by the distance between the core point currently being searched and the surrounding unvisited points, so as to complete the update of the maximum radius.

[0124] For S7, this may include:

[0125] Select any core point in the updated core point set, and form a cluster for all density reachable points corresponding to the core point; form a cluster for all density reachable points corresponding to the unvisited points in the updated core point set; until all data points in the updated core point set are visited, based on all clusters, obtain the clustering results of each unknown interference in the interference signal.

[0126] Specifically, in the process of S7, starting from any core point, all points that are accessible through density from the core point are found to form a cluster. The formed cluster will form a connected area. Then continue to repeat this process for unvisited points until all data points have been visited. When all points have been visited, the connected area constructed according to density accessibility forms the final clustering result as the clustering result of each unknown interference in the interference signal.

[0127] It can be understood that the specific algorithm flow of the radar unknown interference clustering method based on the adaptive DBSCAN algorithm proposed in the embodiment of the present invention is as follows:

[0128] Input: unclustered data X (data of unknown radar interference type), minimum number of points MinPts.

[0129] Output: data type label vector Y.

[0130] Initialization: isVisit = zeros(1,N), isNoise = zeros(1,N), Y = zeros(1,N), C = 0, N is the number of data, D = {|x n -x m ||x n ∈X,x m ∈X,n≠m},m,n=1,2,3,...,N.

[0131] 1) Randomly select an unaccessed data x i , distances = [], if there is no unvisited data, the algorithm stops;

[0132] 2) Select 2*MinPts distances x i The nearest data, whose distance is D initialize ;

[0133] 3) Calculate D initialize The mean and variance of initialize and std initialize ;

[0134] 4) epsilon = mean initialize +2*std initialize ;

[0135] 5) epsilon < 0.1*max(D), otherwise select a new data;

[0136] 6) Search for x i The data set within the maximum radius epsilon is represented by X new , these data need to be unaccessed, otherwise select a new data;

[0137] 7) If X new If the number of elements is greater than the minimum number of points MinPts, then mark x i The position corresponding to isVisit is 1, C=C+1, and the corresponding position in Y is marked x i With X new The elements in are of type C, X new The elements and x i The clustering can be represented as a vector D new , then distances=[distants,D new ]; otherwise, the data is considered to be noise, the corresponding positions of isNoise and isVisit are marked as 1, and 1 is returned);

[0138] 8) Traverse the other unvisited points of type C and denote them as x j , mark the position corresponding to isVisit as 1, and calculate the mean of distants dis , variance std dis and moment skewness ske dis ,but

[0139]

[0140] 9) Search for x j 2*MinPts data that have not been accessed nearby, from x j The distance is represented by D unVisit , calculate D unVisit The mean unVisit and variance std unVisit , b = mean isVisit +2*std isVisit ;

[0141] Then the updated maximum radius is:

[0142]

[0143] 10) Search for x j The data set within the maximum radius epsilon is represented by X new ;

[0144] 11) If X newIf the number of elements in is greater than the minimum number of points MinPts, then mark the corresponding position in Y with x j With X new The elements in are of type C, X new The elements and x j The clustering can be represented as a vector D new , then dis=[dis,D new ]; until all data of type C are accessed, return to 1).

[0145] In order to facilitate understanding of the beneficial effects of the radar unknown interference clustering method based on the adaptive DBSCAN algorithm proposed in the embodiment of the present invention, a specific simulation experiment is carried out on the method below to verify the method.

[0146] Setting up simulation data

[0147] The input parameters used can be noise amplitude modulation interference, noise frequency modulation interference, noise phase modulation interference, noise convolution interference, noise sweep interference, distance deception interference, speed deception interference, distance wave gate interference, speed wave gate interference, dense false target interference, intermittent sampling forwarding interference, slice forwarding interference, spectrum dispersion interference, and smart noise interference; the radar active interference recognition network model based on the multi-layer channel attention mechanism is used to extract the feature mapping of 6 unknown interferences to obtain the final feature map, whose dimension is 512 dimensions, and the convolutional neural network used to extract features is designed for 8 known interferences. After PCA dimensionality reduction, the features with a proportion greater than 95% are obtained, and the 15-dimensional features after dimensionality reduction are obtained. The input data is the 15-dimensional features obtained by dimensionality reduction of the six interferences.

[0148] Simulation content

[0149] The radar unknown interference clustering method based on the adaptive DBSCAN algorithm proposed in the embodiment of the present invention is used to adaptively cluster the above unknown interference, and the corresponding clustering result is shown in FIG. Figure 4a-4f As shown. Figure 4a-4f It can be seen that Figure 4a and Figure 4b All of them are clustering results within the range of JNR (Jamming to Noise Ratio) = 11 to 20 dB. It can be observed that the traditional DBSCAN algorithm clusters the data into 7 clusters, while the adaptive DBSCAN algorithm clusters the data into 6 clusters, but the overall observation is not much different. Figure 4c and Figure 4dThe clustering results are all within the range of JNR = 6 to 20dB. It can be observed that due to the decrease in the interference-to-noise ratio, the clustering results of the two algorithms are relatively chaotic, and the number of noise points increases significantly. The traditional DBSCAN algorithm clusters the data into 14 clusters, while the adaptive DBSCAN algorithm can still cluster the data into 6 clusters. Since the simulation data has 100 samples for each interference-to-noise ratio, there are 1500 samples of each interference within the range of JNR = 6 to 20dB. Figure 4e and Figure 4f To randomly select 200 samples from 1500 samples for clustering, the corresponding clustering results are Figure 4c and Figure 4d The corresponding results are similar, which verifies that the clustering effect has little correlation with the amount of data, but has a greater correlation with JNR.

[0150] In order to verify the effectiveness of the adaptive DBSCAN algorithm compared with the traditional DBSCAN algorithm, the following three groups of data are set, where the "*" data is the first group of data, in which the data points within the class are close, the "x" data is the second group of data, and the "+" data is the third group of data. Each group of data contains 200 sample points, and the data point spacing in the third group of data is larger than that in the first group of data. The comparison diagram of the radar unknown interference clustering method based on the adaptive DBSCAN algorithm provided by the embodiment of the present invention and the traditional DBSCAN algorithm is shown in FIG. Figure 5a-5d , the specific data diagram is as follows Figure 5a As shown in the figure, observe the clustering results. When the minimum number of points Minpts is fixed to 5, if a smaller maximum radius epsilon is used, such as Figure 5b As shown in , the first and third groups of data can be clustered normally, but because the intra-class spacing of the second group of data is larger, more clusters will be generated, and most of the data will be judged as noise; if a larger maximum radius epsilon is used, such as Figure 5c As shown in the figure, the second group of data can be clustered normally, but the first and second groups of data have been overlapped and belong to the same cluster. This is because the interval between the first and third groups of data is smaller than the intra-class interval of the second group of data. When the adaptive DBSCAN algorithm is used, there is no need to consider the selection of the maximum radius corresponding to the clustering algorithm. It is only necessary to measure the minimum number of points Minpts. Figure 5d The overall clustering effect is obvious due to the traditional DBSCAN algorithm. In addition, when finding the optimal clustering, the adaptive DBSCAN algorithm only needs to optimize the minimum number of points Minpts as a discrete parameter, which can find the optimal clustering more conveniently and quickly.

[0151] In the solution provided in the embodiment of the present invention, the traditional DBSCAN algorithm is improved. When determining the maximum radius, the maximum radius corresponding to the core point is updated based on the preset minimum number of points and the distance between the core point and the data point corresponding to the core point, and an updated maximum radius is obtained, thereby realizing adaptive changes in the maximum radius corresponding to each core point, and then using the updated maximum radius to realize the update of each core point; thereby reducing the clustering time, improving the clustering efficiency, and enhancing the clustering effect.

[0152] It should be noted that in the description of the present invention, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0153] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A radar unknown interference clustering method based on an adaptive DBSCAN algorithm, which defines ε-domain, core point, direct density reachable, density reachable, density connected, noise point, boundary point and unvisited point respectively; characterized in that, include: Preprocessing the obtained interference signal to obtain a time-frequency diagram and an RD diagram of the interference signal; Using a radar active interference identification network model based on a multi-layer channel attention mechanism, feature extraction is performed on the time-frequency graph and the RD graph respectively to obtain a time-frequency feature graph and an RD feature graph; Performing dimensionality reduction processing on the time-frequency feature graph and the RD feature graph to obtain feature parameters of several dimensions as data points; Processing all data points based on the preset maximum radius and the preset minimum number of points to obtain a corresponding core point set; Randomly select any core point in the core point set, and obtain an initial maximum radius based on the distance between the core point and the remaining data points and a preset minimum number of points; Updating the first core point in the core point set according to the initial maximum radius; For each core point except the first core point in the core point set, take it as the current core point in turn, and obtain the maximum radius of the cluster to which the current core point belongs based on the direct density reachable points corresponding to the current core point; Based on the current core point and the unvisited point corresponding to it, an adjustment parameter is obtained; based on the maximum radius in the cluster and the adjustment parameter, an updated maximum radius of the current core point is obtained; The current core point is updated according to the updated maximum radius, thereby obtaining an updated core point set; Based on the updated core point set, a clustering result of each unknown interference in the interference signal is obtained.

2. According to claim 1, a radar unknown interference clustering method based on an adaptive DBSCAN algorithm is characterized in that: The ε-domain, core point, direct density reachable, density reachable, density connected, noise point, boundary point and unvisited point are defined respectively, including: The ε-domain is defined as follows: for a data point p, a circle is set with p as the center and a preset maximum radius as the radius. The set of data points contained in the circle is called the ε-domain; Definition of the core point: In the ε-domain of a data point, if the number of elements contained is greater than or equal to the preset minimum number of points, then the data point is called a core point; Definition of direct density reachability: If data point x2 is contained in the ε-domain of data point x1, and data point x1 is a core point, then x2 is said to be directly density reachable from x1; The density can be defined as follows: If there are data points x1, x2, ..., x N-1 , x N , and x i+1 By x i Direct density can be reached, then x N It is reachable by x1 density; where i∈[1,N], N≥2; Definition of density connection: If for data point x1 and data point x2, there exists a core point x3 such that both x1 and x2 are density-reachable from the core point x3, then data point x1 and data point x2 are said to be density-connected; Definition of the noise point: a point that does not belong to any cluster is a noise point; Defining the boundary points: data points within the maximum radius of a core point but not core points are boundary points; Definition of the unvisited point: a data point that is not defined as a core point, a boundary point, or a noise point is an unvisited point.

3. According to claim 1, a radar unknown interference clustering method based on an adaptive DBSCAN algorithm is characterized in that: The step of preprocessing the obtained interference signal to obtain a time-frequency diagram and an RD diagram of the interference signal includes: Performing short-time Fourier transform on the obtained interference signal within a pulse repetition period to obtain a time-frequency diagram of the interference signal; Under the condition of pulse accumulation, moving target detection is performed on the interference signal to obtain an RD diagram of the interference signal.

4. According to claim 1, a radar unknown interference clustering method based on an adaptive DBSCAN algorithm is characterized in that: The step of extracting features from the time-frequency graph and the RD graph by using the radar active interference identification network model based on the multi-layer channel attention mechanism to obtain the time-frequency feature graph and the RD feature graph includes: The time-frequency graph and RD graph are processed in sequence by the corresponding convolutional neural network module, attention mechanism module and multi-layer channel attention feature extraction module in the radar active interference identification network model, so as to obtain the time-frequency feature graph and RD feature graph.

5. The radar unknown interference clustering method based on the adaptive DBSCAN algorithm according to claim 1 is characterized in that: The randomly selecting any core point in the core point set and obtaining an initial maximum radius based on the distance between the core point and the remaining data points and a preset minimum number of points includes: Randomly select any core point in the core point set to obtain the distance between the core point and the remaining data points; According to the distance between the core point and the rest of the data points, the mean is obtained. initialize and standard deviation std initialize ; Using the first formula, the mean initialize and standard deviation std initialize Processing is performed to obtain the initial maximum radius; the first formula is as follows: epsilon1=mean initialize +2*std initialize ; Among them, epsilon1 is less than 0.1 times the first target distance, the first target distance is the maximum distance between the core point and the remaining data points, and the data points within the initial maximum radius corresponding to the core point are all unvisited points.

6. The radar unknown interference clustering method based on the adaptive DBSCAN algorithm according to claim 1 is characterized in that: Updating the first core point in the core point set according to the initial maximum radius includes: The ε-domain corresponding to the first core point in the core point set is determined according to the initial maximum radius, thereby completing the update of the first core point in the core point set.

7. The radar unknown interference clustering method based on the adaptive DBSCAN algorithm according to claim 1 is characterized in that: The step of obtaining the maximum radius of the cluster to which the current core point belongs based on the current core point and its corresponding directly density reachable point includes: Accumulate the distance between the current core point and its corresponding directly reachable point to form a vector distances; According to the vector distances, the mean of the vector distances is obtained. dis , standard deviation std dis and moment skewness ske dis ; The second formula is used to calculate the mean dis , standard deviation std dis and moment skewness ske dis Processing is performed to obtain the maximum radius within the cluster; the second formula is as follows:

8. The radar unknown interference clustering method based on the adaptive DBSCAN algorithm according to claim 1 is characterized in that: The obtaining of adjustment parameters based on the current core point and the unvisited point corresponding thereto includes: Based on the distance between the current core point and the second target twice as large as its surrounding distance, the mean is obtained. inVisit and standard deviation std inVisit ; The second target distance is the distance between the preset minimum number of unvisited points; The third formula is used to calculate the mean inVisit and standard deviation std inVisit Processing is performed to obtain the adjustment parameter; the third formula is as follows: b=mean inVisit +2*std inVisit 。 9. The radar unknown interference clustering method based on the adaptive DBSCAN algorithm according to claim 1 is characterized in that: The updated maximum radius of the current core point is obtained based on the maximum radius within the cluster and the adjustment parameter, including: The fourth formula is used to process the maximum radius within the cluster and the adjustment parameter to obtain the updated maximum radius; the fourth formula is as follows: Among them, a represents the maximum radius within the cluster, and b represents the adjustment parameter.

10. The radar unknown interference clustering method based on the adaptive DBSCAN algorithm according to claim 1 is characterized in that: The step of obtaining a clustering result of each unknown interference in the interference signal based on the updated core point set includes: Select any core point in the updated core point set, and form a cluster for all density reachable points corresponding to the core point; form a cluster for all density reachable points corresponding to the unvisited points in the updated core point set; until all data points in the updated core point set are visited, based on all clusters, obtain the clustering results of each unknown interference in the interference signal.

Images