CN113625242B - Radar signal sorting method based on potential distance graph combined PCA and improved cloud model - Google Patents
Radar signal sorting method based on potential distance graph combined PCA and improved cloud model Download PDFInfo
- Publication number
- CN113625242B CN113625242B CN202110836585.0A CN202110836585A CN113625242B CN 113625242 B CN113625242 B CN 113625242B CN 202110836585 A CN202110836585 A CN 202110836585A CN 113625242 B CN113625242 B CN 113625242B
- Authority
- CN
- China
- Prior art keywords
- data
- potential energy
- radar signal
- sorting
- distance
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 53
- 238000010586 diagram Methods 0.000 claims abstract description 12
- 238000011156 evaluation Methods 0.000 claims abstract description 5
- 238000005381 potential energy Methods 0.000 claims description 69
- 239000011159 matrix material Substances 0.000 claims description 10
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 230000008859 change Effects 0.000 claims description 5
- 230000008030 elimination Effects 0.000 claims description 2
- 238000003379 elimination reaction Methods 0.000 claims description 2
- 238000000513 principal component analysis Methods 0.000 description 20
- 230000000694 effects Effects 0.000 description 5
- 230000003993 interaction Effects 0.000 description 4
- 238000010606 normalization Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000009467 reduction Effects 0.000 description 4
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/418—Theoretical aspects
Abstract
The application provides a radar signal sorting method based on potential distance diagram combined PCA and an improved cloud model. The method for solving the problem of batch increase in multi-mode radar signal classification is mainly researched, namely, the problem of classifying different modulation modes of one radar into multiple radars. The method comprises the following steps: pre-sorting the multi-mode radar signals by using a potential distance graph; the main characteristics after the pre-sorting are extracted through PCA to serve as new characteristics of radar signals; the membership relation between the data clusters is calculated by utilizing an improved cloud model theory, and simple and efficient classification evaluation standards are set to finish the sorting of multimode radar signals. The method can improve the sorting accuracy of the multimode radar signals, solve the problem of batch increment in multimode radar signal sorting, and improve the efficiency of multimode radar signal sorting to a certain extent.
Description
Technical Field
The application relates to a radar signal sorting method based on potential distance diagram combined PCA and an improved cloud model, which is a method for sorting different modulation modes of a radar into different radars, namely, the problem of batch increase, and belongs to the field of radar signal sorting.
Background
Multimode radar has become a primary radar in modern electronic battlefield, with multiple modes of operation and complex modulated waveforms. The intercepted multimode radar signals are often difficult to sort, so that different working modes of one radar can be easily sorted into different radars in a short time. This causes great trouble in the identification of subsequent radar signals, thereby directly affecting the effect of electronic warfare. Therefore, how to sort multimode radar signals quickly and accurately has long been a very important problem in the field of electronic warfare.
In response to this problem, a method that has been used in recent years is to extract various features in and between pulses and sort signals based on these features. In terms of sorting using inter-Pulse parameters, the main features include Pulse Deion Word (PDW) and Pulse repetition interval (Pulse Repetition Interval, PRI). The classification method mainly comprises clustering, convolutional neural network, support vector machines (Support Vector Machine, SVM) and the like. The above methods can achieve sorting of multimode radar signals at low signal-to-noise ratios. However, in the sorting process, a judgment process of sorting results is generally lacking. Therefore, a problem of sorting different modulation modes of one radar into a plurality of radars is unavoidable in a short time.
Disclosure of Invention
Aiming at the problem of batch increase of multimode radar signal sorting, the application provides a method based on potential distance graph combined PCA and an improved cloud model. According to the method, firstly, the potential distance graph is utilized to pre-sort radar signals, signal data points are clustered into different data clusters, then PCA is utilized to extract and construct main features, finally, cloud model theory is improved, membership degree relation between the data clusters is calculated, and the problem of batch increment of multi-mode radar signal sorting is solved.
The application aims at realizing the following steps:
step one: constructing a data field, improving a density peak clustering algorithm by using the data field, and solving the problem that the cutting distance r in the clustering algorithm needs manual experience setting;
step two: eliminating interference points by using potential energy values;
step three: selecting a clustering center by using the potential distance graph, and further completing radar signal pre-sorting;
step four: extracting main features by using Principal Component Analysis (PCA), thereby constructing new features F;
step five: based on the new feature F, the membership relationship between the data clusters is analyzed by utilizing an improved cloud model theory, and the problem of batch increase of multi-mode radar signal sorting is solved.
The application also includes such structural features:
1. the first step is as follows:
the functional model for forming the data field is as follows:
wherein ,represents the potential energy value of the ith data in position x; />Representing potential energy values at location x; m is m i Representing the quality of each datum; assuming that each data quality is equal and that the sum of the data quality is 1; sigma represents the influencing factor of the data field.
The potential energy function of the data field is very similar to the gaussian kernel function, except that one more concept of data quality is presented above. The position of the influencing factor sigma in the potential energy function is the position of the truncated distance r in the density peak clustering algorithm. As long as a sigma method is found, the magnitude of r is determined by analyzing the relation between r and sigma, so that the problem that r needs to be determined manually and empirically is solved. In terms of sigma selection, the most widely used method is to obtain the minimum value of the unitary function of the potential entropy about the influence factor sigma, and the data field stability corresponding to the sigma obtained at this time is the highest. The method is specifically as follows:
wherein ,is a normalization factor; />Is the potential energy value at each point.
The best σ can be found by the above equation, and the relationship between σ and r is analyzed below. The field function in the data field has the form exp (- |x-x) i ||/σ) k When k=2, the function is in the form of a gaussian kernel function. According to the "3σ" nature of the gaussian kernel function, the data with values distributed over the range of (μ -3σ, μ+3σ) represents 99.74% of the total data when the function is represented by the following equation. Thus, the range of effective interaction forces of the potential energy function of the data field can be determined to beThus, the specific cutoff distance r is as follows:
after r is determined, r is taken to the position of σ in the functional model forming the data field to obtain the potential energy value for each point.
2. The second step is specifically as follows:
and eliminating the interference points by utilizing the potential energy values of the data points. The potential energy value is much lower than at data densities because there is no data around the interference point. The potential energy value of each data point is ordered from big to small, the potential energy value of the interference point is always ordered at the last, and compared with the potential energy value of the data density place, the potential energy value of the interference point has larger fall. By comparing the change of the ratio k between two data potential energy values, the data points are data interference points at and after the position where k is large, so that the discrete interference points can be removed. The specific definition is as follows:
3. the third step is as follows:
after the interference points are removed and potential energy values of the data points are obtained, the distance attribute of each data point can be defined according to the size of the potential energy values. In the dataset (x) 1 ,x 2 ,…x k ) In which the distance of the data points is defined as the ratio x i Point of high potential energy to x i A minimum value of the distance; if point x i Is the point of greatest potential energy in the data set, and the distance attribute of the point is defined as other points x in the data set j (j+.i) to x i Maximum value of the distance. The specific formula is shown as follows:
wherein ,potential energy values representing radar signal data points; d, d ij Representing the distance of two pulse data points; d, d j Representing the distance attribute of the j-th data.
After defining the distance attribute, in order to eliminate the influence of potential energy and the dimension of the distance attribute, normalization processing is needed first to form a potential distance graph at the same order of magnitude. Potential energy is used as an abscissa and distance is used as an ordinate to construct a potential distance graph, so that a clustering center is automatically selected. In the potential distance graph, points with larger potential energy and distance can be separated from other points to become cluster centers, so that the selection of the cluster centers is completed. After the center point of the radar signal sample data cluster is acquired, the rest data points are divided into the center points closest to the center point by calculating Euclidean distances from other data points to the center point, so that the final pre-sorting is finished.
4. The fourth step is specifically as follows:
after pre-sorting, three of each classThe dimensional characteristic parameters DOA, PW and RF lead-in SPSS are subjected to PCA analysis, and the extracted characteristic values are set to lambda > 1 by default to obtain a component matrix and an interpretation total variance. Since there are only 3 inter-pulse parameter features, there are only 1 or 2 new features extracted after dimension reduction by PCA. After the component matrix and the explained total variance matrix are obtained, the coefficients of the following formula can be calculated to obtain the principal component F 1 and F2 ，F 1 and F2 Are not related to each other. The coefficients of the formula are the data in the component matrix divided by the eigenvalues corresponding to the principal components divided by the square root. The corresponding feature values of the principal components can be derived from the interpreted total variance table, after which the final new feature F is constructed. F represents the vast majority of useful information of the inter-pulse feature parameters. The subsequent calculated amount is reduced after the PCA dimension reduction, and a foundation is laid for the subsequent membership discrimination by utilizing the improved cloud model theory and establishing a unified threshold parameter.
F 1 ＝k 11 x PW +k 12 x RF +k 13 x DOA F 2 ＝k 21 x PW +k 22 x RF +k 23 x DOA
5. The fifth step is specifically as follows:
first, the Expectation (Ex), entropy (En), and super Entropy-like (Similar Hyper Entropy, se) of each data cluster after the pre-sorting is completed are obtained, as shown in the following formula:
wherein /> wherein ,xpq Representing the q-th parameter sample in the p-th class cluster; s is the total number of data clusters; l is the total number of droplets in the p-th cluster (p=1, 2, …, s; q=1, 2, …, L).
After computing (Ex, en, se) of the data cluster, taking En as a mean value and Se as a mean value according to the working principle of the forward cloud generator 2 For variance, after generating random number r in the new parameter feature dimension, the membership degree of a certain sample to a certain data cluster is defined as shown in the following formula. After the membership degree of a single sample to a certain data cluster is obtained, the membership degree of the whole data cluster to the certain data cluster can be calculated. And calculating the membership degree of the data cluster for T times based on the new feature F, and then solving the average value, wherein the specific steps are as follows:
where r=norm (En, se) 2 )
After the membership relationship between the data clusters is obtained by using the improved cloud model, the created classification evaluation criteria are as follows:
(1) Taking digital features Ex, en and Se of each data cluster based on the new feature F;
(2) Setting a membership threshold mu;
(3) If membership means between two classes of data clustersAnd judging that the two types of radar signal data belong to different working modes of the same radar, otherwise, judging that the two types of radar signal data belong to different radars.
Compared with the prior art, the application has the beneficial effects that: the core technical content of the application is as follows: according to the characteristics of multi-mode radar signal emission, the application provides a radar signal pre-sorting method based on a potential distance graph, and the method can improve the accuracy and efficiency of multi-mode radar signal pre-sorting; and extracting the main characteristics after pre-sorting by using Principal Component Analysis (PCA), and performing membership analysis between data clusters based on the extracted main characteristics by using an improved cloud model theory to solve the problem of batch increase of multi-mode radar signal sorting.
According to the radar signal sorting method based on the potential distance diagram combined PCA and the improved cloud model, which is disclosed by the application, the sorting accuracy of multimode radar signals can be effectively improved, the sorting efficiency is improved, and the problem of batch increment of multimode radar signal sorting can be better solved. The method is suitable for the sorting condition of multimode radar signals in a complex electromagnetic environment.
The application mainly researches a solution method for the problem of batch increment in multimode radar signal sorting, and the method comprises the following steps: the problem that the cut-off distance in the density peak clustering algorithm needs to be set manually is solved based on the data field, and then a two-dimensional potential distance graph is constructed to select a clustering center, so that the multi-mode radar signal pre-sorting is completed. Secondly, PCA is used for main feature extraction to reduce the subsequent calculation amount. After the main features are obtained, the membership relationship between the data clusters is analyzed based on the main features by utilizing an improved cloud model theory, so that the problem of batch increase in multimode radar signal classification is solved. The method can improve the accuracy rate of multimode radar signal sorting, has more stable and mature sorting effect compared with other algorithms, and has better sorting effect.
Drawings
FIG. 1 is a schematic block diagram of a radar signal sorting method based on a potential distance map combined PCA and an improved cloud model;
FIG. 2 is a three-dimensional plot of a multimode radar signal;
FIG. 3 is a graph of potential entropy versus influence factor;
FIG. 4 is a schematic diagram of eliminating data interference points;
FIG. 5 is a schematic illustration of the final formed potential distance map selection cluster center;
fig. 6 is a sorting result of the set multimode radar signal.
Detailed Description
The application is described in further detail below with reference to the drawings and the detailed description.
The application discloses a radar signal sorting method based on potential distance diagram combined PCA and an improved cloud model, which specifically comprises the following steps:
(1.1) in order to better sort the multi-mode radar signals, a radar signal pre-sorting algorithm based on a potential distance diagram is provided, so that the sorting accuracy is improved, and the algorithm complexity is reduced;
and (1.2) performing membership analysis on the pre-sorted radar signals by using PCA and an improved cloud model to realize sorting of multimode radar signals.
The method feature (1.1) comprises:
(2.1) assume that there are 4 radars, including two integrated system radars and two single system radars. Both complex radar systems have 3 modes of operation. The total of 4 radars has 8 modes of data clusters. The extracted characteristic dimensions comprise an arrival angle (DOA), a carrier frequency (RF) and a Pulse Width (PW), and firstly, three characteristic parameters of radar signal data points are normalized, wherein the normalization method is as follows:
after normalizing the range of the radar signal to-1 to +1, the optimal influence factor sigma is calculated according to the following formula, namely, the minimum value of potential entropy H about the influence factor sigma is calculated, and the corresponding sigma is the optimal influence factor.
wherein />
After determining the influence factor σ, a determined cutoff distance r may be calculated, resulting in a potential energy value for each data point, where:
at this time, the potential energy value of each data point is as follows:
according to the formula, potential energy values of the data points are obtained, the potential energy value attribute can represent the density of the data points, the potential energy values represent the interaction of the data points in the data field, the influence of the interaction is reflected, and the density of the data points can be well reflected.
(2.2) because in practical engineering applications, the parameters of the intercepted radar pulse are often affected by noise, interference and other factors, it is inevitable that some single interference points are present that deviate from the data cluster. After the potential energy value of the data point is obtained, the interference point can be removed according to the potential energy value. Because there is no data around a single point of interference, the potential energy value is much lower than where the data is denser. After sorting the potential energy values of each data point from large to small, the potential energy value of a single data point is always ranked last, with a larger drop than the potential energy value at the data-intensive site. By comparing the change in the ratio k between two data potential values, the data points are data interference points where k is large and thereafter. The method is specifically as follows:
after the interference points are removed, some difficulties are reduced for subsequent classification, and the multi-mode radar signal classification is facilitated.
(2.3) after the interference points are removed, defining the distance attribute of each data point according to the potential energy value. In particular in a dataset (x 1 ,x 2 ,…x k ) Each of the followingThe distance of data points is defined as all points x that are greater than potential energy i A minimum value of the distance; if point x i Is the point of greatest potential energy in the data set, and the distance attribute of the point is defined as other points x in the data set j (j+.i) to x i The maximum value of the distance is expressed as follows:
after determining the distance attribute of the data point, normalizing the potential energy and the distance attribute value to form a potential distance map at the same order of magnitude. The specific method is as follows:
in the potential distance graph, points with larger potential energy and larger distance can be separated from other points to form a clustering center. But only a larger potential energy or a larger distance cannot become a cluster center.
(2.4) after the center point of the radar signal sample data cluster is acquired, dividing the rest data points to the center point closest to the center point by calculating Euclidean distances from other data points to the center point, and then finishing the pre-sorting of the final radar signals.
The method feature (1.2) comprises:
and (3.1) after a pre-selection result is obtained based on the potential distance graph, three-dimensional characteristic parameters DOA, PW and RF of each class are imported into SPSS for main component analysis, and the extracted characteristic values are set as lambda > 1 by default, so that a component matrix and an interpretation total variance are obtained. Since there are only 3 inter-pulse parameter features, there are only 1 or 2 new features extracted after dimension reduction by PCA. The method is specifically as follows:
F 1 ＝k 11 x PW +k 12 x RF +k 13 x DOA
F 2 ＝k 21 x PW +k 22 x RF +k 23 x DOA
in the above, F 1 and F2 The coefficients in (a) are the square root of the data in the component matrix divided by the eigenvalues corresponding to the principal components, which can be derived from the explained total variance table of the SPSS analysis. F represents most useful information of the characteristic parameters between the original pulses, so that the subsequent class and class can be distinguished by representing the original clustering result.
(3.2) after acquiring the new feature F of the radar signal data cluster based on the pre-selection result, acquiring the Expectation (Ex), entropy (En) and super Entropy-like (Similar Hyper Entropy, se) of each data cluster by the new feature representing most of the effective information of each data cluster, as shown in the following formula:
wherein />
After computing (Ex, en, se) of the data cluster, taking En as a mean value and Se as a mean value according to the working principle of the forward cloud generator 2 For variance, after generating random number r in the new parameter feature dimension, the membership degree of a certain sample to a certain data cluster is defined as shown in the following formula. After the membership degree of a single sample to a certain data cluster is obtained, the membership degree of the whole data cluster to the certain data cluster can be calculated. Membership degree of data cluster based on new feature FThe line T times is calculated, and then the average value is calculated, specifically as follows:
where r=norm (En, se) 2 )
Thus, the membership degree among the data clusters is obtained according to the new feature F extracted by PCAMembership degree->The similarity degree between the data clusters is reflected, the similarity degree between the data clusters of different modes of the same radar is certainly high, and the similarity degree between the data clusters of different radars is certainly poor. According to the improved cloud model theory, the membership relationship between different data clusters of the radar signal can be obtained.
(3.3) membership degree between different data clusters of the obtained radar signalThen, a proper membership threshold value mu is required to be set to distinguish whether the radar signal data clusters belong to the data of the same radar. According to the 3En rule of the cloud model, is located in [ Ex-3E ] n ,Ex+3E n ]The other data belongs to a small probability event. And is located at [ Ex-E ] n ,Ex+E n ]The contribution of the "base element" of the interval is the largest of the total contributions. Because of the strong similarity between different modulation modes of the same radar radiation source, it is located within the "base element" of the data cluster formed by the other modulation mode. Thus, setting x=ex+en as the reference parameter in the new feature dimension, the specific membership threshold setting is as follows:
(3.4) after determining the membership threshold value μ, creating a final classification evaluation criterion as follows:
(1) The digital features Ex, en, se of the respective data clusters are acquired based on the new feature F.
(2) A membership discrimination threshold mu is set.
(3) If it isAnd judging that the two types of radar signal data belong to different working modes of the same radar.
(4) If it isIt is determined that the two types of radar signal data belong to different radars.
In order to more clearly illustrate the applied method, the embodiment of the application performs flow illustration and effect display through simulation experiments, but does not limit the scope of the embodiment of the application. The experimental conditions are as follows: the 4-part radar comprises two-part integrated system radars and two-part single system radars. Both complex radar systems have 3 modes of operation. During the simulation process, a certain number of interference points are added in order to simulate the influence of random noise, interference and other external factors.
Fig. 1 is a general schematic block diagram of the method of the present application, comprising:
s110 extracts DOA, PW, RF of the radar signal, and a three-dimensional distribution diagram of the radar signal parameter characteristics is shown in fig. 2. The field function pattern for forming the data field is determined as follows:
wherein ,representing potential energy value of ith data in position x；/>Representing potential energy values at location x; m is m i Representing the quality of each datum; assuming that each data quality is equal and that the sum of the data quality is 1; sigma represents the influencing factor of the data field.
In the aspect of the selection of the influencing factors, the most widely applied method at present is to obtain the minimum value of the unitary function of the potential entropy about the influencing factors sigma, and the data field stability of the corresponding sigma is highest when the potential entropy H is minimum. The method is specifically as follows:
wherein ,is a normalization factor; />Is the potential energy value at each point.
S112, after determining σ, analyzes the relationship between σ and r. The field function in the data field has the form exp (- |x-x) i ||/σ) k When k=2, the function is in the form of a gaussian kernel function. According to the "3σ" nature of the gaussian kernel function, the data with values distributed over the range of (μ -3σ, μ+3σ) represents 99.74% of the total data when the function is represented by the following equation. The range of effective interaction forces of the potential energy function of the data field can be judged to beThus, the specific cutoff distance r is as follows:
after r is determined, r is taken to the position of σ in the functional model forming the data field to obtain the potential energy value for each point. The potential energy properties of the final data point are shown as follows:
wherein />
In this way potential energy values for the data points are obtained.
S120, eliminating the interference points according to the potential energy value after the potential energy value of the data point is obtained. Because there is no data around a single point of interference, the potential energy value is much lower than where the data is denser. After sorting the potential energy values of each data point from large to small, the potential energy value of a single data point is always ranked last, with a larger drop than the potential energy value at the data-intensive site. By comparing the change in the ratio k between two data potential values, the data points are data interference points where k is large and thereafter. The method is specifically as follows:
s130, acquiring the distance attribute of the data point according to the following formula according to the potential energy value of the acquired data point:
wherein ,potential energy values representing radar signal data points; d, d ij Representing the distance of two pulse data points; d, d j Representing the distance attribute of the j-th data.
Thus, the distance attribute of the data point is obtained, and the potential energy and the distance of the data point are normalized according to the following formula.
In the potential distance graph, points with larger potential energy and larger distance can be separated from other points to form a clustering center. But only a larger potential energy or a larger distance cannot become a cluster center.
S140, three-dimensional characteristic parameters DOA, PW and RF of each class are imported into SPSS for main component analysis, and a default setting lambda > 1 is adopted for extracting characteristic values to obtain a component matrix and an interpretation total variance. Since there are only 3 inter-pulse parameter features, there are only 1 or 2 new features extracted after dimension reduction by PCA. The method is specifically as follows:
F 1 ＝k 11 x PW +k 12 x RF +k 13 x DOA
F 2 ＝k 21 x PW +k 22 x RF +k 23 x DOA
in the above, F 1 and F2 The coefficients in (a) are the square root of the data in the component matrix divided by the eigenvalues corresponding to the principal components, which can be derived from the explained total variance table of the SPSS analysis. This way the last new feature F is obtained.
S141 obtains expectations (Ex), entropies (Entropy, en), and super-entropies (Similar Hyper Entropy, se) of the respective data clusters based on the new feature F, as shown in the following formula:
wherein /> wherein ,xpq Representing the q-th parameter sample in the p-th class cluster; s is the total number of data clusters; l is the total number of droplets in the p-th cluster (p=1, 2, …, s; q=1, 2, …, L).
S142 after calculating (Ex, en, se) of the data cluster, taking En as the mean value and Se as the mean value according to the working principle of the forward cloud generator 2 For variance, after generating random number r in the new parameter feature dimension, the membership degree of a certain sample to a certain data cluster is defined as shown in the following formula. After the membership degree of a single sample to a certain data cluster is obtained, the membership degree of the whole data cluster to the certain data cluster can be calculated. And calculating the membership degree of the data cluster for T times based on the new feature F, and then solving the average value, wherein the specific steps are as follows:
where r=norm (En, se) 2 )
S143, after the membership degree among the data clusters is obtained, creating a final classification evaluation standard as follows:
(1) The digital features Ex, en, se of the respective data clusters are acquired based on the new feature F.
(2) A membership discrimination threshold mu is set.
(3) If it isThen it is determined that the two types of radar signal data belong to the sameDifferent modes of operation of a radar.
(4) If it isIt is determined that the two types of radar signal data belong to different radars.
Wherein, the membership discrimination threshold μ is as follows:
fig. 2 is a graph of a set radar signal sample distribution. Fig. 3 is a graph showing the change of potential entropy with influence factor, and it can be seen from the graph: at the minimum potential entropy, the corresponding influence factor σ is 0.06, at which point the calculated cut-off distance r=0.127. Fig. 4 is an exemplary diagram of eliminating interference points, and the set interference point elimination limit is 1.5. Fig. 5 is a resulting potential distance map from which it can be seen that the radar signal is divided into 8 data clusters. Fig. 6 is a final multimode radar signal sorting result. The method provided by the application has good sorting effect on the multimode radar, so that the method can ensure that the correct sorting of multimode radar signals can be realized in a more complex environment, and the problem of batch increase of multimode radar signals is solved.
In summary, the application mainly researches a solution to the problem of 'batch increment' in multimode radar signal classification, namely, the problem of classifying different modulation modes of a radar into multiple radars. The method of the embodiment can provide a radar signal sorting method based on potential distance diagram combined PCA and improved cloud model according to the distribution characteristics of multimode radar signals. According to the method, the pre-sorting of radar signals is completed by utilizing the potential distance graph, main features of the radar signals are extracted by utilizing PCA, membership degrees among pre-sorted data clusters are calculated based on an improved cloud model theory, and then multi-mode radar signal sorting is completed. The method can improve the sorting accuracy of the multimode radar signals, solve the problem of batch increase in multimode radar signal sorting, and improve the efficiency of multimode radar signal sorting to a certain extent.
It will be appreciated by those skilled in the art that, in the foregoing method according to the present application, the sequence number of each step does not mean that the execution sequence of each step should be determined by the function and the internal logic, and should not limit the implementation process of the embodiment of the present application.
Finally, it should be noted that the above embodiments are only intended to describe the technical solution of the present application and not to limit the technical method, the present application extends to other modifications, variations, applications and embodiments in application, and therefore all such modifications, variations, applications, embodiments are considered to be within the spirit and scope of the teachings of the present application.
Claims (2)
1. A radar signal sorting method based on potential distance diagram combined PCA and improved cloud model comprises the following steps:
step one: constructing a data field, and acquiring potential energy values of data points to express the density of each signal sample point by combining a density peak clustering algorithm;
extracting an arrival angle DOA, a pulse width PW and a carrier frequency RF of a radar signal to construct a three-dimensional data field; the potential energy function to form the data field is:
wherein ： represents the potential energy value of the ith data in position x; />Representing potential energy values at location x; m is m i Representing the quality of each datum; assuming that each data quality is equal and that the sum of the data quality is 1; r is acquisition ofThe cutoff distance of potential energy values of data points, sigma represents an influence factor of a data field;
step two: eliminating interference points by using potential energy values;
sorting potential energy values of each data point from large to small, comparing the change of the ratio k between the two data potential energy values, determining interference points, and finishing the elimination of the data interference points; the method is specifically as follows:
step three: determining the distance attribute of the data points, constructing a potential distance graph, selecting a clustering center and then completing the pre-selection of radar signals;
acquiring the distance attribute of the data point according to the following formula according to the potential energy value of the acquired data point:
wherein ,potential energy values representing radar signal data points; d, d ij Representing the distance of two pulse data points; d, d j A distance attribute representing the j-th data;
the potential energy and distance of the data point are normalized as shown in the following formula:
step four: extracting main features by utilizing main component analysis, thereby constructing new features F;
three-dimensional characteristic parameters DOA, PW and RF of each class are imported into SPSS for main component analysis, and a default setting lambda > 1 is adopted for extracting characteristic values to obtain a component matrix and an interpretation total variance; because the number of parameter features among the pulses is only 3, the number of new features extracted after the PCA is used for reducing the dimension is only 1 or 2; the method is specifically as follows:
F 1 ＝k 11 x PW +k 12 x RF +k 13 x DOA
F 2 ＝k 21 x PW +k 22 x RF +k 23 x DOA
in the above, F 1 and F2 The coefficients in (a) are the square root of the division of the data in the component matrix by the eigenvalues corresponding to the principal components, and the eigenvalues corresponding to the principal components can be obtained from the explained total variance table analyzed by the SPSS;
step five: and calculating the membership relationship between the data clusters based on the new characteristic F by utilizing an improved cloud model theory, and solving the problem of batch increase of multi-mode radar signal sorting.
2. The radar signal sorting method based on potential distance map combined PCA and improved cloud model as claimed in claim 1, wherein the method is characterized by comprising the following steps: the fifth step is specifically as follows:
acquiring expected Ex, entropy En and super entropy Se of each data cluster based on the new feature F, wherein the expected Ex, the entropy En and the super entropy Se are shown in the following formula:
wherein />
wherein ,xpq Representing the q-th parameter sample in the p-th class cluster; s is the total number of data clusters; l is the total number of droplets in the p-th cluster, p=1, 2, …, s; q=1, 2, …, L;
after computing (Ex, en, se) of the data cluster, taking En as a mean value and Se as a mean value according to the working principle of the forward cloud generator 2 After generating a random number r on the characteristic dimension of the new parameter for variance, calculating the membership degree of the whole data cluster to a certain data cluster after obtaining the membership degree of a single sample to the certain data cluster; and calculating the membership degree of the data cluster for T times based on the new feature F, and then solving the average value, wherein the specific steps are as follows:
where r=norm (En, se) 2 )
After obtaining the membership between the data clusters, a final classification evaluation criterion is created as follows:
(1) Acquiring digital features Ex, en and Se of each data cluster based on the new feature F;
(2) Setting a membership degree discrimination threshold mu;
(3) If it isJudging that the two types of radar signal data belong to different working modes of the same radar;
(4) If it isJudging that the two types of radar signal data belong to different radars;
wherein, the membership discrimination threshold μ is as follows:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110836585.0A CN113625242B (en) | 2021-07-23 | 2021-07-23 | Radar signal sorting method based on potential distance graph combined PCA and improved cloud model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110836585.0A CN113625242B (en) | 2021-07-23 | 2021-07-23 | Radar signal sorting method based on potential distance graph combined PCA and improved cloud model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113625242A CN113625242A (en) | 2021-11-09 |
| CN113625242B true CN113625242B (en) | 2023-09-29 |
Family
ID=78380775
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110836585.0A Active CN113625242B (en) | 2021-07-23 | 2021-07-23 | Radar signal sorting method based on potential distance graph combined PCA and improved cloud model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113625242B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7034738B1 (en) * | 2003-12-05 | 2006-04-25 | Itt Manufacturing Enterprises, Inc. | Method of radar pattern recognition by sorting signals into data clusters |
| CN105005029A (en) * | 2015-07-17 | 2015-10-28 | 哈尔滨工程大学 | Multi-mode radar signal sorting method based on data field hierarchical clustering |
| CN107656245A (en) * | 2017-08-22 | 2018-02-02 | 哈尔滨工程大学 | A kind of method being applied to information fusion in radar signal sorting |
| CN109613486A (en) * | 2018-12-03 | 2019-04-12 | 中国人民解放军空军工程大学 | A kind of Radar Signal Sorting Method based on core cluster support vector clustering |
| CN109886098A (en) * | 2019-01-11 | 2019-06-14 | 中国船舶重工集团公司第七二四研究所 | A kind of AESA radar frequency agility mode excavation method across sorting interval |
| CN110531335A (en) * | 2019-09-18 | 2019-12-03 | 哈尔滨工程大学 | A kind of low complex degree similitude clustering signal sorting method based on Union-find Sets |
| CN110780271A (en) * | 2019-10-18 | 2020-02-11 | 西安电子科技大学 | Spatial target multi-mode radar classification method based on convolutional neural network |
| CN112986928A (en) * | 2021-03-11 | 2021-06-18 | 哈尔滨工程大学 | Signal sorting multi-source fusion processing method in complex electromagnetic environment |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101007662B1 (en) * | 2009-05-08 | 2011-01-13 | 국방과학연구소 | Radar Signals Clustering Method using Frequency Modulation Characteristics and Combination Characteristics of Signals, and System for Receiving and Processing Radar Signals using the same |
| US11047970B2 (en) * | 2017-05-05 | 2021-06-29 | Texas Instruments Incorporated | Multi-mode radar systems, signal processing methods and configuration methods using pushing windows |
| WO2020256692A1 (en) * | 2019-06-17 | 2020-12-24 | Google Llc | Mobile device-based radar system for applying different power modes to a multi-mode interface |
-
2021
- 2021-07-23 CN CN202110836585.0A patent/CN113625242B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7034738B1 (en) * | 2003-12-05 | 2006-04-25 | Itt Manufacturing Enterprises, Inc. | Method of radar pattern recognition by sorting signals into data clusters |
| CN105005029A (en) * | 2015-07-17 | 2015-10-28 | 哈尔滨工程大学 | Multi-mode radar signal sorting method based on data field hierarchical clustering |
| CN107656245A (en) * | 2017-08-22 | 2018-02-02 | 哈尔滨工程大学 | A kind of method being applied to information fusion in radar signal sorting |
| CN109613486A (en) * | 2018-12-03 | 2019-04-12 | 中国人民解放军空军工程大学 | A kind of Radar Signal Sorting Method based on core cluster support vector clustering |
| CN109886098A (en) * | 2019-01-11 | 2019-06-14 | 中国船舶重工集团公司第七二四研究所 | A kind of AESA radar frequency agility mode excavation method across sorting interval |
| CN110531335A (en) * | 2019-09-18 | 2019-12-03 | 哈尔滨工程大学 | A kind of low complex degree similitude clustering signal sorting method based on Union-find Sets |
| CN110780271A (en) * | 2019-10-18 | 2020-02-11 | 西安电子科技大学 | Spatial target multi-mode radar classification method based on convolutional neural network |
| CN112986928A (en) * | 2021-03-11 | 2021-06-18 | 哈尔滨工程大学 | Signal sorting multi-source fusion processing method in complex electromagnetic environment |
Non-Patent Citations (4)
| Title |
|---|
| Multi-mode radar signal sorting by means of spatial data mining;Wan J, Nan P L, Guo Q,;Journal of Communications and Networks;第18卷(第5期);全文 * |
| 基于PCA 联合K-Means 聚类的雷达信号分选算法;翁永祥, 赵万磊, 郑志娟;舰船电子对抗;第43卷(第2期);全文 * |
| 基于数据场与云模型的多模雷达信号分选算法;国强;宋文明;南普龙;万建;;哈尔滨工业大学学报(第11期);全文 * |
| 数据质量聚类算法;李延, 王大魁, 耿晶;武 汉大学学报: 信息科学版;第44卷(第1期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113625242A (en) | 2021-11-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111580500B (en) | Evaluation method for safety of automatic driving automobile | |
| Ibrahim et al. | Cluster representation of the structural description of images for effective classification | |
| CN111062425B (en) | Unbalanced data set processing method based on C-K-SMOTE algorithm | |
| CN109871855B (en) | Self-adaptive deep multi-core learning method | |
| CN110232371B (en) | High-precision HRRP radar multi-target identification method based on small samples | |
| CN113344019A (en) | K-means algorithm for improving decision value selection initial clustering center | |
| CN110751121A (en) | Unsupervised radar signal sorting method based on clustering and SOFM | |
| CN113537316B (en) | Vehicle detection method based on 4D millimeter wave radar point cloud | |
| CN105809113A (en) | Three-dimensional human face identification method and data processing apparatus using the same | |
| CN109709527A (en) | The Gauss wave crest method of Gauss Decomposition in a kind of Full wave shape laser-measured height echo signal | |
| CN106610977B (en) | Data clustering method and device | |
| CN112508963A (en) | SAR image segmentation method based on fuzzy C-means clustering | |
| CN113484875A (en) | Laser radar point cloud target hierarchical identification method based on mixed Gaussian ordering | |
| CN114626435A (en) | High-accuracy rolling bearing intelligent fault feature selection method | |
| CN113625242B (en) | Radar signal sorting method based on potential distance graph combined PCA and improved cloud model | |
| Wen et al. | An improved de-interleaving algorithm of radar pulses based on SOFM with self-adaptive network topology | |
| CN110348478B (en) | Method for extracting trees in outdoor point cloud scene based on shape classification and combination | |
| CN111914930A (en) | Density peak value clustering method based on self-adaptive micro-cluster fusion | |
| Su et al. | Unsupervised k-means combined with SOFM structure adaptive radar signal sorting algorithm | |
| CN111611293A (en) | Outlier data mining method based on feature weighting and MapReduce | |
| CN110826599B (en) | Sparse representation sample distribution boundary retention feature extraction method | |
| CN113341379A (en) | Radar signal sorting method based on adaptive threshold and iterative control | |
| Zhang et al. | A new outlier detection algorithm based on fast density peak clustering outlier factor. | |
| CN110956177A (en) | Hybrid verification code identification method and system | |
| CN110766087A (en) | Method for improving data clustering quality of k-means based on dispersion maximization method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |