CN109447163B - Radar signal data-oriented moving object detection method - Google Patents

Radar signal data-oriented moving object detection method Download PDF

Info

Publication number
CN109447163B
CN109447163B CN201811296013.2A CN201811296013A CN109447163B CN 109447163 B CN109447163 B CN 109447163B CN 201811296013 A CN201811296013 A CN 201811296013A CN 109447163 B CN109447163 B CN 109447163B
Authority
CN
China
Prior art keywords
clustering
signal
data
value
direction angle
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
Application number
CN201811296013.2A
Other languages
Chinese (zh)
Other versions
CN109447163A (en
Inventor
赵颖
张蓉
罗晓波
周芳芳
赵韦鑫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Central South University
Original Assignee
Central South University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Central South University filed Critical Central South University
Priority to CN201811296013.2A priority Critical patent/CN109447163B/en
Publication of CN109447163A publication Critical patent/CN109447163A/en
Application granted granted Critical
Publication of CN109447163B publication Critical patent/CN109447163B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques

Abstract

The invention discloses a radar signal data-oriented moving object detection method, which comprises the following steps: 1) calculating the information entropy of each dimension according to an information entropy principle, and selecting the dimension with larger information entropy for subsequent clustering; 2) calculating clustering parameters of the DBSCAn by a k-distance method; 3) clustering the signal data based on the selected clustering features and the calculated clustering parameters; 4) anomaly detection is carried out on the direction angle time sequence information of the signals based on an ARIMA method, and a candidate anomaly point set is established for each cluster; 5) adjusting elements in the cluster or adjusting direction angle data of the signal according to the candidate abnormal point set detected in the step 4). The method detects and corrects the moving objects (airplanes and ships) hidden in the radar signals from two aspects of signal data and direction data according to the clustering and anomaly detection theory, and lays a solid foundation for the analysis of the behavior mode of the subsequent moving objects.

Description

Radar signal data-oriented moving object detection method
Technical Field
The invention belongs to the technical field of computer information processing, and relates to a radar signal data-oriented moving object detection method.
Background
Generally, a signal analyzer needs to sort a large amount of signal data, classify signals belonging to the same signal source, and analyze the behavior characteristics of the signal source based on the classification. In order to correctly analyze the behavior characteristics of the signal source, the signal sorting result needs to have higher accuracy, otherwise, the subsequent analysis tasks are all based on an error. Due to the technical development in recent years, many signal acquisition systems can automatically identify and classify some signals, but these systems can only classify signals with very close signal attributes and identify the signal source by a unique row record in a signal table, and if the attribute value of the signal fluctuates greatly, the signal acquisition system can identify the signal as the signal sent by another signal source. Therefore, one record in the signal table cannot completely represent one signal source, and the scattered signal sources need to be classified by means of a clustering method.
Because the signal attribute values are different more, the acquisition system can identify the signal as the signals sent by a plurality of signal sources, so that the signals can only be roughly filtered and the number of the signals can be reduced by simply clustering the signals through the signal attributes, and the accuracy of the clustering result can not be effectively ensured. To improve the accuracy of the clustering result, we have recourse to the direction data of the signal. The clustering result is corrected by detecting whether the direction angle change of the signal in one cluster is coherent or not, so that the accuracy of signal source detection is improved. Where a source may be an aircraft or a ship, the present invention is directed to accurate detection of these moving objects by radar signal data, providing a solid foundation for subsequent analysis by analysts.
Disclosure of Invention
The invention solves the technical problem that the radar signal data-oriented moving object detection method is provided for overcoming the defects of the prior art, and detects and corrects the moving objects (airplanes and ships) hidden in radar signals from two aspects of signal data and direction data according to the clustering and anomaly detection theory, thereby laying a solid foundation for the analysis of the behavior mode of the subsequent moving objects.
A moving object detection method facing radar signal data comprises the following steps:
step 1): selecting clustering characteristics from radar signal data: firstly, calculating the occurrence frequency of each dimension in radar signal data points by using a histogram method, then calculating the information entropy of each dimension according to the information entropy principle, and selecting the dimension with larger information entropy as a clustering feature to perform subsequent clustering;
step 2): calculating clustering parameters of DBScan: on the basis of the clustering characteristics selected in the step 1), calculating the distance between each signal data point and other signal data points and performing ascending sorting on the calculation results to obtain ascending data sets with the same number as the signal data points, then selecting values in fixed sorting positions in each ascending data set based on a k-distance principle and performing ascending sorting to obtain a k-distance value sequence, and simultaneously taking the serial numbers of the fixed sorting positions as min _ samples parameters of the DBSCAn clustering parameters; then drawing a k-distance curve graph according to the k-distance value sequence, and taking the y value at the position with the maximum curve slope change in the graph as the Eps value of the DBSCAn clustering parameter;
step 3): clustering: clustering the signal data points based on the selected clustering characteristics and the calculated clustering parameters;
step 4): and (3) detecting candidate abnormal points: searching direction angle information recorded by all signal data points in each cluster in radar signal data according to the clustering result obtained in the step 3), performing ascending arrangement according to the attribute of the timestamp, simulating a direction angle time sequence based on an ARIMA method, and finally performing anomaly point detection through residual errors to establish a candidate anomaly point set for each cluster;
step 5): evaluating candidate outliers: adjusting elements in the cluster or adjusting direction angle data of the signal data points according to the candidate abnormal point set detected in the step 4).
The method for detecting the moving object facing to the radar signal data comprises the following steps in step 1):
step 1.1): calculating the occurrence frequency of each dimension in radar signal data points by using a histogram method;
step 1.2): and calculating the entropy value of each dimensionality according to an information entropy calculation formula, and selecting the dimensionality with the largest entropy value as a subsequent clustering dimensionality.
In the method for detecting moving objects facing radar signal data, in step 1.1), when the frequency of occurrence of each dimension is calculated, dimension samples x with discrete attributes are used1,x2,…,xi,…,xn-1,xnWherein n represents the number of data of the attribute, and the number of data in the data set is consistent, and the set T = { x } of the sample elements is counted1,x2,…,xi,…,xm-1,xmAnd for each element in the set T, the number of times c (x) it appears in the sample is calculatedi) Finally, the approximate probability of an element in the set, i.e. the frequency with which it appears in the sample, is calculated by the formula
Figure BDA0001851198250000031
For a dimension sample of continuous attributes, firstly discretizing the value range of the attributes into m disjoint intervals Rm＝(xi,xi+1](i is 0, …, m-1), and placing the signal data point in the corresponding interval by comparing the size of the attribute value, and further, counting the number of data c (R) in each intervalm) And calculating the probability of each interval
Figure BDA0001851198250000032
In the method for detecting a moving object oriented to radar signal data, the dimension of the continuous attribute includes: frequency, pulse period, pulse width, pulse period maximum, pulse period minimum, maximum intra-pulse width, and minimum intra-pulse width, the dimensions of the discrete attributes including: bandwidth type, frequency type, pulse period type, pulse width type.
In the method for detecting a moving object oriented to radar signal data, in step 1.2), when an entropy value h (x) of each dimension is calculated, the dimension with discrete attributes is adopted
Figure BDA0001851198250000033
Performing calculations for dimensions of continuous attributes
Figure BDA0001851198250000041
And (6) performing calculation.
The radar signal data-oriented moving object detection method comprises the following specific steps in the step 2):
step 2.1): calculating the distance between each data point and the rest of the data points based on the number of the dimensions selected as the clustering characteristics in the step 1)
Figure BDA0001851198250000042
Figure BDA0001851198250000043
Wherein d isi,jRepresents the data point xiAnd xjThe Euclidean distance between the data points, l represents the number of the dimensionalities as the clustering feature, n represents the number of the data points, and then the calculation result D is comparediSequencing in ascending order to obtain a sequenced data set
Figure BDA0001851198250000044
Represents;
step 2.2): data set obtained according to step 2.1)
Figure BDA0001851198250000045
Selecting a 4 th value of each data set based on a k-distance principle, namely selecting 4 k values of the k-distance principle, and performing ascending sorting to obtain a k-distance value sequence;
step 2.3): according to the k-distance value sequence obtained in the step 2.2), taking the value of each element in the k-distance value sequence as a vertical coordinate, taking the position of each element in the sequence as a horizontal coordinate, drawing a k-distance curve graph, and calculating a y value at the maximum change of the slope of the curve in the graph as a required Eps value; secondly, the user can also select the Eps value according to the graph;
step 2.4): the min _ samples parameter setting is consistent with the value of k, i.e., set to 4.
The method for detecting the moving object facing to the radar signal data, wherein the step 3) comprises the following steps:
clustering the data by using a DBScan algorithm according to the dimension selected in the step 1) and the clustering parameters obtained in the step 2); the obtained clustering result is { c1,c2,…,ct,…,ckT is more than or equal to 1 and less than or equal to k.
The method for detecting the moving object facing to the radar signal data, wherein the step 4) comprises the following steps:
step 4.1): taking the clustering result of the step 3) as { c1,c2,…,ct,…,ckT is more than or equal to 1 and less than or equal to k, and direction angle record information of all signals in each cluster is inquired in a direction table; to be provided with
Figure BDA0001851198250000051
Representing a cluster ctWherein the direction angles of all signals are recorded and arranged in ascending order according to the time stamp attributes of the corresponding records, wherein fiRepresenting the direction angle of a signal, nctRepresenting a cluster ctThe number of all signal direction angle records in;
further, for the cluster ctRespectively counting the direction angle information of each signal, and converting the direction angle information into FctThe direction angle value in (1) is classified according to the signal to which the direction angle value belongs, and the classification result is
Figure BDA0001851198250000052
Figure BDA0001851198250000053
Wherein m isctRepresenting the central signal of the clusterThe number of numbers, i.e., the number of elements;
Figure BDA0001851198250000054
a sequence of direction angle data representing a signal. And with CFRepresenting the number of direction angle records of a signal for use in step 5);
step 4.2): and (3) carrying out anomaly detection on the direction angle time sequence data of each cluster based on an ARIMA method:
the method comprises the steps of simulating a direction angle time sequence by using an ARIMA method, firstly determining three parameters of p, d and q, wherein d represents the difference times when the time sequence is changed into a stable sequence, determining whether the time sequence is stable by ADF (automatic frequency analysis) test to obtain a parameter d, firstly determining a possible value range by an ACF (anisotropic conductive film) diagram and a PACF (picture archiving and communication) diagram for the parameter p and the parameter q, then selecting the p and the q which enable the value of AIC and BIC to be minimum as parameters, and adopting the formula of AIC as follows: AIC ═ 2ln (l) +2k, the formula of BIC is: BIC ═ 2ln (L) + ln (n) × k, where L is the maximum likelihood of the model when (p, d, q) takes a certain set of values, nn is the number of data, and kk is the number of variables of the model; the direction angle time sequence data of a certain cluster is set as f1,f2,…,fn} analog sequence by ARIMA (p, d, q) of { f'1,f′2,…,f′nIs the residual sequence is { e }1,e2,…,enIn which en＝fn-f′nFinally, the mean value of the residual sequence is calculated
Figure BDA0001851198250000055
Sum variance
Figure BDA0001851198250000056
If | eiMu | is greater than 3 sigma, then the corresponding f is considerediIs an anomaly point;
step 4.3): via step 4.2) a cluster c is obtainedtSet of outliers of
Figure BDA0001851198250000057
Wherein n istIndicating the number of candidate outliers in the cluster.
The method for detecting the moving object facing to the radar signal data, wherein the step 5) comprises the following steps:
step 5.1): based on the result obtained in the step 4.3), clustering c according to the signal to which the abnormal point belongstCandidate outlier set N oftClassifying; the classification result is expressed as
Figure BDA0001851198250000061
Wherein m istRepresenting the number of signals to which all candidate outliers belong, a subset X of candidate outliers1＝{f1,f2,…,fi,…,fx1-1,fx1An anomaly point f iniBelonging to a single signal, x1 being represented at NtThe number of abnormal points belonging to the same signal; with combined use of CnfNumber of outliers representing a signal;
step 5.2): comparing the clusters c according to the statistical results of 4.1) and 5.1)tDirection angle recording number C of the same signalFAnd number of outliers Cnf. If, if
Figure BDA0001851198250000062
The signal is not considered to belong to the cluster, and the signal is deleted from the cluster; if it is not
Figure BDA0001851198250000063
Considering part of the direction angle of the signal as noise, and removing the noise from the signal;
step 5.3): step 5 is executed in a loop until all clusters are traversed.
The method has the technical effects that the moving objects (airplanes and ships) hidden in the radar signals are detected and corrected from the two aspects of signal data and direction data according to the clustering and anomaly detection theory, and a solid foundation is laid for the analysis of the behavior mode of the subsequent moving objects.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a radar signal clustering view;
fig. 3 is an EPS curve view.
Detailed Description
In order to make the objects, design considerations and advantages of the present invention more apparent, the present invention will be described in further detail below with reference to specific examples and the accompanying drawings.
The invention comprises the following steps:
step 1): and selecting clustering characteristics. Because the signal data has more characteristics and some characteristics may be noise data, the clustering method not only does not contribute to the clustering result, but also can reduce the clustering effect, the information entropy of each dimension is calculated according to the information entropy principle, and the dimension with larger information entropy is selected for subsequent clustering;
step 2): the clustering parameters of DBScan are calculated. Calculating clustering parameters of the DBSCAn by a k-distance method;
step 3): and (6) clustering. Clustering the signal data based on the selected clustering features and the calculated clustering parameters;
step 4): and detecting candidate abnormal points. Anomaly detection is carried out on the direction angle time sequence information of the signals based on an ARIMA method, and a candidate anomaly point set is established for each cluster;
step 5): candidate outliers are evaluated. Adjusting elements in the cluster or adjusting direction angle data of the signal according to the candidate abnormal point set detected in the step 4).
The step 1) comprises the following steps:
step 1.1): the radar signal data oriented by the invention consists of two tables, wherein one table is a signal table which records the basic attribute value of a logic signal, namely the frequency, pulse period and other attributes of the signal sent by the same signal source (such as an airplane or a ship); the other one is a direction table which records the direction angle change information of a logic signal (such as an airplane or a ship) in the signal table, and the signal table and the table are in one-to-many relationship.
Due to the error of the signal receiving system, it is possible to use the same signal source S1The emitted signals are identified as a plurality of signal sources S1,S1′,S″1…, and storing the signals in a signal table as a plurality of records with different signal ids (different from the signal source S)1,S′1,S″1…) are identical. In order to help signal analysts have a correct knowledge of the signals of the same signal source, we need to identify these signals which belong to essentially the same signal source. It should be noted here that although the system does not correctly identify the signal source (S)1,S′1,S″1…) into a category, but the direction information of each signal source in the direction table is generated by the movement of the signal source, and the condition that the direction information of other signal sources is identified incorrectly and stored under the signal source does not exist.
In order to help signal analysts correctly know the behavior characteristics of the same signal source, the signals in the signal table need to be clustered, i.e., the signals belonging to the same signal source are classified into one category. Since the number of features recorded in the signal table is large, and some of the features may be noise data, not only does not contribute to the clustering result, but also may reduce the clustering effect, we need to select those features containing much information for clustering. Here, it is assumed that the data set has n pieces of data.
Step 1.2): and calculating the probability distribution of each dimension in the radar direction data by using a histogram method. Here we need to distinguish between discrete and continuous attributes. For discrete attributes, assume its sample is x1,x2,…,xi,…,xn-1,xnWhere n indicates that the number of data for the attribute is consistent with the number of data in the data set. The set T ═ x of the sample elements is counted1,x2,…,xi,…,xm-1,xmAnd for each element in the set T, calculating the number c (x) of its reappearance in the samplei) Finally, the approximate probability of an element in the set, i.e. the frequency with which it appears in the sample, is calculated by the formula
Figure BDA0001851198250000081
For continuous attributes, we need to discretize the value range of the attribute,divided into m disjoint intervals Rm＝(xi,xi+1](i-0, …, m-1) and places the data point in the corresponding bin by comparing the size of the attribute value. Further, the number c (R) of data in each section is countedm) And calculating the probability of each interval
Figure BDA0001851198250000082
Step 1.3): calculating formula according to information entropy
Figure BDA0001851198250000083
Or
Figure BDA0001851198250000084
And calculating the entropy value of each dimensionality, and selecting the dimension with the largest entropy value as a subsequent clustering dimension. Meanwhile, the visual interactive system also supports user-defined clustering dimensions.
The step 2) comprises the following steps:
step 2.1): calculating the distance between each data point and the rest of the data points based on the subsequent clustering dimensions selected in the step 1)
Figure BDA0001851198250000085
Wherein d isi,jRepresents the data point xiAnd xjThe euclidean distance between the data points is n, the number of the data points is n, and the number of the dimensions serving as the clustering features is l, for example, in step 1.3), the 3 dimensions with the largest entropy value are taken, and l is 3, that is, the respective euclidean distances are calculated for the three dimensions of the data points. Then for the calculated result DiSorting in ascending order, as used herein
Figure BDA0001851198250000091
Representing the sorted data set;
step 2.2): ascending data set obtained according to step 2.1)
Figure BDA0001851198250000092
Based on the k-distance principle, each one is selectedThe 4 th value is ordered in the data set and constitutes the data set. Further, sorting the data set in an ascending order to obtain a k-distance value sequence;
step 2.3): and drawing a k-distance curve graph according to the k-distance value sequence obtained in the step 2.2). Further calculating the place with the maximum change of the slope of the curve according to the graph, wherein the y value is the Eps value; secondly, the user can also select the Eps value according to the graph;
step 2.4): the min _ samples parameter setting is typically consistent with the value of k, and is therefore set to 4 here as well.
The step 3) comprises the following steps:
clustering the data by using a DBScan algorithm according to the dimension selected in the step 1.3) and the clustering parameters obtained in the step 2);
assume that after all the above steps, the obtained clustering result is { c }1,c2,…,ct,…,ckT is more than or equal to 1 and less than or equal to k.
The step 4) comprises the following steps:
step 4.1): based on the clustering result { c) of step 3)1,c2,…,ct,…,ckInquiring direction angle record information of all signals in each cluster in a direction angle table; herein make
Figure BDA0001851198250000093
Representing a cluster ctWherein the direction angles of all signals are recorded and arranged in ascending order according to the time stamp attributes of the corresponding records, wherein fiRepresenting the direction angle of a signal, nctRepresenting a cluster ctThe number of all signal direction angle entries in.
Further, for the cluster ctThe direction angle information of each signal in the system is respectively counted. F is to bectThe direction angle value in (1) is classified according to the signal to which the direction angle value belongs, and the classification result is
Figure BDA0001851198250000101
Figure BDA0001851198250000102
Wherein m isctRepresents the number of signals, i.e. the number of elements, in the cluster;
Figure BDA0001851198250000103
representing a sequence of directional angles of a signal. Let CFRepresenting the direction angle record number of the signal for the step 5);
step 4.2): and carrying out anomaly detection on the direction angle time sequence data of each cluster according to a wavelet transform method.
The direction angle time series is first simulated using the ARIMA method. For the method, three parameters (p, d, q) need to be determined, wherein d represents the difference times when the time sequence is changed into a stable sequence, whether the time sequence is stable can be determined by ADF inspection, and thus a parameter d is obtained, and p and q can be determined by combining an ACF graph, a PACF graph, an AIC criterion and a BIC criterion respectively, namely, the possible value ranges of p and q are determined by the ACF graph and the PACF graph, and then p and q which enable the value of AIC and BIC to be minimum are selected as the parameters. The formula for AIC is: AIC ═ 2ln (l) +2k, the formula of BIC is: BIC ═ 2ln (L) + ln (n) × k, where L is the maximum likelihood of the model when (p, d, q) takes a certain set of values, n is the number of data, and k is the number of variables of the model. Suppose that the direction angle timing data of a certain cluster is { f1,f2,…,fn} analog sequence by ARIMA (p, d, q) of { f'1,f′2,…,f′nIs the residual sequence is { e }1,e2,…,enIn which en＝fn-f′n. Finally calculating the mean value of the residual error sequence
Figure BDA0001851198250000104
Sum variance
Figure BDA0001851198250000105
If | eiMu | is greater than 3 sigma, then the corresponding f is considerediIs an anomaly. In the mentioned ARIMA method, p, d and q are parameters of ARIMA, and after the three parameters are determined, the ARIMA model can be usedThe fitted time series, say the original time series, is (t1, f1), (t2, f2) …, where t denotes time and f denotes the azimuth. Then, after determining the p, d, and q parameters of the ARIMA model, t may be transmitted to the ARIMA model, and f' is calculated as ARIMA (t).
Step 4.3): assume that after this step, cluster ctSet of outliers of
Figure BDA0001851198250000106
Figure BDA0001851198250000111
Wherein n istIndicating the number of candidate outliers in the cluster.
The step 5) comprises the following steps:
step 5.1): based on the result obtained in the step 4.3), clustering c according to the signal to which the abnormal point belongstCandidate outlier set N oftAnd (6) classifying. The classification result can be expressed as
Figure BDA0001851198250000112
Wherein m istRepresenting the number of signals to which all candidate outliers belong, a subset X of candidate outliers1＝{f1,f2,…,fi,…,fx1-1,fx1An anomaly point f iniBelonging to a single signal, x1 being represented at NtThe number of abnormal points belonging to the same signal. Not for a particular signal, in general we use CnfRepresenting the number of outliers of the signal;
step 5.2): comparing the clusters c according to the statistical results of 4.1) and 5.1)tDirection angle recording number C of the same signalFAnd number of outliers Cnf. If it is not
Figure BDA0001851198250000113
The signal is not considered to belong to the cluster, and the signal is deleted from the cluster; if it is not
Figure BDA0001851198250000114
Considering part of the direction angle of the signal as noise, and removing the noise from the signal;
step 5.3): let t be 1, loop all clusters until t k ends.

Claims (9)

1. A method for detecting a moving object facing radar signal data is characterized by comprising the following steps:
step 1): selecting clustering characteristics from radar signal data: firstly, calculating the occurrence frequency of each dimension in radar signal data points by using a histogram method, then calculating the information entropy of each dimension according to the information entropy principle, and selecting the dimension with larger information entropy as a clustering feature to perform subsequent clustering;
step 2): calculating clustering parameters of DBScan: on the basis of the clustering characteristics selected in the step 1), calculating the distance between each signal data point and other signal data points and performing ascending sorting on the calculation results to obtain ascending data sets with the same number as the signal data points, then selecting values in fixed sorting positions in each ascending data set based on a k-distance principle and performing ascending sorting to obtain a k-distance value sequence, and simultaneously taking the serial numbers of the fixed sorting positions as min _ samples parameters of the DBSCAn clustering parameters; then drawing a k-distance curve graph according to the k-distance value sequence, and taking the y value at the position with the maximum curve slope change in the graph as the Eps value of the DBSCAn clustering parameter;
step 3): clustering: clustering the signal data points based on the selected clustering characteristics and the calculated clustering parameters;
step 4): and (3) detecting candidate abnormal points: searching direction angle information recorded by all signal data points in each cluster in radar signal data according to the clustering result obtained in the step 3), performing ascending arrangement according to the attribute of the timestamp, simulating a direction angle time sequence based on an ARIMA method, and finally performing anomaly point detection through residual errors to establish a candidate anomaly point set for each cluster;
step 5): evaluating candidate outliers: adjusting elements in the cluster or adjusting direction angle data of the signal data points according to the candidate abnormal point set detected in the step 4).
2. The method of claim 1, wherein the step 1) comprises the following steps:
step 1.1): calculating the occurrence frequency of each dimension in radar signal data points by using a histogram method;
step 1.2): and calculating the entropy value of each dimensionality according to an information entropy calculation formula, and selecting the dimensionality with the largest entropy value as a subsequent clustering dimensionality.
3. The method for detecting moving objects facing radar signal data according to claim 2, wherein in step 1.1), the frequency of occurrence of each dimension is calculated according to dimension sample x of discrete attribute1，x2，…，xi，…，xn-1，xnWherein n represents the number of data of the attribute, and is consistent with the number of data in the data set, and the set T ═ x of the sample elements is counted1，x2，…，xi，…，xn-1，xnAnd for each element in the set T, the number of times c (x) it appears in the sample is calculatedi) Finally, the approximate probability of an element in the set, i.e. the frequency with which it appears in the sample, is calculated by the formula
Figure FDA0003285461000000021
For a dimension sample of continuous attributes, firstly discretizing the value range of the attributes into m disjoint intervals Rm＝(xi，xi+1](i is 0, …, m-1), and placing the signal data point in the corresponding interval by comparing the size of the attribute value, and further, counting the number of data c (R) in each intervalm) And calculating the probability of each interval
Figure FDA0003285461000000022
4. The method of claim 3, wherein the dimensions of the continuous attributes comprise: frequency, pulse period, pulse width, pulse period maximum, pulse period minimum, maximum intra-pulse width, and minimum intra-pulse width, the dimensions of the discrete attributes including: bandwidth type, frequency type, pulse period type, pulse width type.
5. The method according to claim 3, wherein in step 1.2), in calculating the entropy h (x) of each dimension, the dimension with discrete attributes is adopted
Figure FDA0003285461000000023
Performing calculations for dimensions of continuous attributes
Figure FDA0003285461000000024
And (6) performing calculation.
6. The method for detecting moving objects facing radar signal data according to claim 1, wherein the step 2) includes the following specific steps:
step 2.1): calculating the distance between each data point and the rest of the data points based on the number of the dimensions selected as the clustering characteristics in the step 1)
Figure FDA0003285461000000031
Figure FDA0003285461000000032
Wherein d isi，jRepresents the data point xiAnd xjThe Euclidean distance between the data points, l represents the number of the dimensionalities as the clustering feature, n represents the number of the data points, and then the calculation result D is comparediSequencing in ascending order to obtain a sequenced data set
Figure FDA0003285461000000033
Represents;
step 2.2): data set obtained according to step 2.1)
Figure FDA0003285461000000034
Selecting a 4 th value of each data set based on a k-distance principle, namely selecting 4 k values of the k-distance principle, and performing ascending sorting to obtain a k-distance value sequence;
step 2.3): according to the k-distance value sequence obtained in the step 2.2), taking the value of each element in the k-distance value sequence as a vertical coordinate, taking the position of each element in the sequence as a horizontal coordinate, drawing a k-distance curve graph, and calculating a y value at the maximum change of the slope of the curve in the graph as a required Eps value; secondly, the user can also select the Eps value according to the graph;
step 2.4): the min _ samples parameter setting is consistent with the value of k, i.e., set to 4.
7. The method of claim 1, wherein the step 3) comprises the following steps:
clustering the data by using a DBScan algorithm according to the dimension selected in the step 1) and the clustering parameters obtained in the step 2); the obtained clustering result is { c1，c2，…，ct，…，ckT is more than or equal to 1 and less than or equal to k.
8. The method of claim 1, wherein the step 4) comprises the following steps:
step 4.1): taking the clustering result of the step 3) as { c1，c2，…，ct，…，ckT is more than or equal to 1 and less than or equal to k, and direction angle record information of all signals in each cluster is inquired in a direction table; to be provided with
Figure FDA0003285461000000035
Representing a cluster ctDirection of all signals inThe corners record information and are arranged in ascending order according to the timestamp attribute of the corresponding record, wherein fiRepresenting the direction angle of a signal, nctRepresenting a cluster ctThe number of all signal direction angle records in;
further, for the cluster ctRespectively counting the direction angle information of each signal, and converting the direction angle information into FctThe direction angle value in (1) is classified according to the signal to which the direction angle value belongs, and the classification result is
Figure FDA0003285461000000041
Figure FDA0003285461000000042
Wherein m isctRepresents the number of signals, i.e. the number of elements, in the cluster;
Figure FDA0003285461000000043
a sequence of direction angle data representing a signal; and with CFRepresenting the number of direction angle records of a signal;
step 4.2): and (3) carrying out anomaly detection on the direction angle time sequence data of each cluster based on an ARIMA method:
the method comprises the steps of simulating a direction angle time sequence by using an ARIMA method, firstly determining three parameters of p, d and q, wherein d represents the difference times when the time sequence is changed into a stable sequence, determining whether the time sequence is stable by ADF (automatic frequency analysis) test to obtain a parameter d, firstly determining a possible value range by an ACF (anisotropic conductive film) diagram and a PACF (picture archiving and communication) diagram for the parameter p and the parameter q, then selecting the p and the q which enable the value of AIC and BIC to be minimum as parameters, and adopting the formula of AIC as follows: AIC ═ 2ln (l) +2k, the formula of BIC is: BIC-2 ln (L) + ln (n) × k, where L is the maximum likelihood of the model when (p, d, q) takes a certain set of values, n is the number of data, and k is the number of variables of the model; the time sequence data of the direction angle of a certain cluster is f1，f2，…，fn} analog sequence by ARIMA (p, d, q) of { f'1，f′2，…，f′nIs the residual sequence is { e }1，e2，…，enIn which en＝fn-f′nFinally, the mean value of the residual sequence is calculated
Figure FDA0003285461000000044
Sum variance
Figure FDA0003285461000000045
If | eiMu | is greater than 3 sigma, then the corresponding f is considerediIs an anomaly point;
step 4.3): via step 4.2) a cluster c is obtainedtSet of outliers of
Figure FDA0003285461000000046
Wherein n istIndicating the number of candidate outliers in the cluster.
9. The method of claim 1, wherein the step 5) comprises the following steps:
step 5.1): based on the result obtained in the step 4.3), clustering c according to the signal to which the abnormal point belongstCandidate outlier set N oftClassifying; the classification result is expressed as
Figure FDA0003285461000000047
Wherein m istRepresenting the number of signals to which all candidate outliers belong, a subset X of candidate outliers1＝{f1，f2，…，fi，…，fx1-1，fx1An anomaly point f iniBelonging to a single signal, x1 being represented at NtThe number of abnormal points belonging to the same signal; with combined use of CnfNumber of outliers representing a signal;
step 5.2): comparing the clusters c according to the statistical results of 4.1) and 5.1)tDirection angle recording number C of the same signalFAnd number of outliers CnfIf, if
Figure FDA0003285461000000051
The signal is not considered to belong to the cluster, and the signal is deleted from the cluster; if it is not
Figure FDA0003285461000000052
Considering part of the direction angle of the signal as noise, and removing the noise from the signal;
step 5.3): step 5 is executed in a loop until all clusters are traversed.
CN201811296013.2A 2018-11-01 2018-11-01 Radar signal data-oriented moving object detection method Active CN109447163B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811296013.2A CN109447163B (en) 2018-11-01 2018-11-01 Radar signal data-oriented moving object detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811296013.2A CN109447163B (en) 2018-11-01 2018-11-01 Radar signal data-oriented moving object detection method

Publications (2)

Publication Number Publication Date
CN109447163A CN109447163A (en) 2019-03-08
CN109447163B true CN109447163B (en) 2022-03-22

Family

ID=65549921

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811296013.2A Active CN109447163B (en) 2018-11-01 2018-11-01 Radar signal data-oriented moving object detection method

Country Status (1)

Country Link
CN (1) CN109447163B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112288561A (en) * 2020-05-25 2021-01-29 百维金科(上海)信息科技有限公司 Internet financial fraud behavior detection method based on DBSCAN algorithm

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US3261016A (en) * 1962-03-08 1966-07-12 Burr Donald William Non-rigid servo-controlled aerial structures
CN105427301A (en) * 2015-11-17 2016-03-23 西安电子科技大学 Sea and land clutter scene segmentation method based on direct current component ratio measure
CN105897488A (en) * 2016-06-13 2016-08-24 中南大学 Visualization method of radio signal data
CN106022359A (en) * 2016-05-12 2016-10-12 武汉理工大学 Fuzzy entropy space clustering analysis method based on orderly information entropy
CN107301409A (en) * 2017-07-18 2017-10-27 云南大学 Learn the system and method for processing electrocardiogram based on Wrapper feature selectings Bagging
CN108073553A (en) * 2017-12-27 2018-05-25 西北师范大学 The unsupervised discretization method of connection attribute data based on comentropy
CN108197647A (en) * 2017-12-28 2018-06-22 中南大学 A kind of Fast Speed Clustering of automobile starter durable test data
CN108449306A (en) * 2017-02-16 2018-08-24 上海行邑信息科技有限公司 One kind degree of peeling off detection method

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US3261016A (en) * 1962-03-08 1966-07-12 Burr Donald William Non-rigid servo-controlled aerial structures
CN105427301A (en) * 2015-11-17 2016-03-23 西安电子科技大学 Sea and land clutter scene segmentation method based on direct current component ratio measure
CN106022359A (en) * 2016-05-12 2016-10-12 武汉理工大学 Fuzzy entropy space clustering analysis method based on orderly information entropy
CN105897488A (en) * 2016-06-13 2016-08-24 中南大学 Visualization method of radio signal data
CN108449306A (en) * 2017-02-16 2018-08-24 上海行邑信息科技有限公司 One kind degree of peeling off detection method
CN107301409A (en) * 2017-07-18 2017-10-27 云南大学 Learn the system and method for processing electrocardiogram based on Wrapper feature selectings Bagging
CN108073553A (en) * 2017-12-27 2018-05-25 西北师范大学 The unsupervised discretization method of connection attribute data based on comentropy
CN108197647A (en) * 2017-12-28 2018-06-22 中南大学 A kind of Fast Speed Clustering of automobile starter durable test data

Non-Patent Citations (7)

* Cited by examiner, † Cited by third party
Title
A study on prediction of rainfall using datamining technique;R. S. Kumar等;《2016 International Conference on Inventive Computation Technologies (ICICT)》;20161231;第1-9页 *
DBSCAN Revisited, Revisited: Why and How You Should (Still) Use DBSCAN;Erich Schubert等;《ACM Transactions on Database Systems》;20170731;第42卷(第3期);第1-19页 *
Velocity/Shape Estimation Algorithm Using Tracking Filter and Data Fusion of Dual Doppler Radar Interferometers;Kenshi Saho等;《International Journal of Computer and Electrical Engineering》;20151231;第283-295页 *
一种基于近类点和模糊点的未知雷达信号分选算法;张荣等;《航船电子对抗》;20111031;第34卷(第5期);第12-14页 *
信息熵时序和树图用于NetFlow可视化的研究;张胜等;《高技术通讯》;20141231;第24卷(第9期);第903-909页 *
现代雷达信号分选技术综述;王杰贵等;《雷达科学与技术》;20060430(第2期);第104-120页 *
舰载雷达探测精度标定方法及关键技术研究;刘冬利;《中国博士学位论文全文数据库·工程科技Ⅱ辑》;20161015(第10期);第C036-2页 *

Also Published As

Publication number Publication date
CN109447163A (en) 2019-03-08

Similar Documents

Publication Publication Date Title
CN107728140B (en) Multi-target multi-channel parallel tracking processing method for warning radar
US9906963B2 (en) Big-data-mining-based wireless channel modeling method
Hansen et al. Measurement of galaxy cluster sizes, radial profiles, and luminosity functions from SDSS photometric data
US10908261B2 (en) Target identification and clutter mitigation in high resolution radar systems
CN111767930A (en) Method for detecting abnormal time series data of Internet of things and related equipment thereof
CN104392250A (en) Image classification method based on MapReduce
CN112986925B (en) Radar pulse sequence sorting method based on image features
CN107576948B (en) Radar target identification method based on high-resolution range profile IMF (inertial measurement framework) features
CN109348416B (en) Fingerprint indoor positioning method based on binary k-means
CN107832778A (en) A kind of same target recognition methods based on spatial synthesis similarity
Janßen et al. K-means clustering of extremes
CN109447163B (en) Radar signal data-oriented moving object detection method
US11754704B2 (en) Synthetic-aperture-radar image processing device and image processing method
CN111738319A (en) Clustering result evaluation method and device based on large-scale samples
CN116362425B (en) Method for analyzing playable area based on weather modification airborne detection data
CN107729903A (en) SAR image object detection method based on area probability statistics and significance analysis
CN115374880B (en) Offshore target identification-oriented multistage incremental data fusion system
CN111666439B (en) Working method for rapidly extracting and dividing medical image big data aiming at cloud environment
Coppa et al. The bimodality of the 10k zCOSMOS-bright galaxies up to z~ 1: a new statistical and portable classification based on optical galaxy properties
CN110781963B (en) K-means clustering-based aerial target clustering method
JP2014206382A (en) Target type identification device
CN113313138A (en) Intrusion behavior feature conversion method and detection method based on probability generation model
Ngan et al. Modeling of traffic data characteristics by dirichlet process mixtures
CN111666262B (en) Working method for extracting feature points of massive medical images in network attached storage NAS state
CN110659700B (en) KNN-based image sample generation 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