CN109447163A - A kind of mobile object detection method towards radar signal data - Google Patents
A kind of mobile object detection method towards radar signal data Download PDFInfo
- Publication number
- CN109447163A CN109447163A CN201811296013.2A CN201811296013A CN109447163A CN 109447163 A CN109447163 A CN 109447163A CN 201811296013 A CN201811296013 A CN 201811296013A CN 109447163 A CN109447163 A CN 109447163A
- Authority
- CN
- China
- Prior art keywords
- cluster
- signal
- data
- value
- dimension
- 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.)
- Granted
Links
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
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a kind of mobile object detection method towards radar signal data, steps are as follows: 1) calculates the comentropy of each dimension according to information entropy principle, the biggish dimension of comentropy is selected to carry out subsequent cluster；2) clustering parameter of DBScan is calculated by k- distance method；3) cluster feature based on selection and calculated clustering parameter cluster signal data；4) it is carried out abnormality detection based on deflection timing information of the ARIMA method to signal, establishes candidate abnormal point set for each cluster；5) the abnormal point set of the candidate detected according to step 4), is adjusted the element in cluster or is adjusted to the direction angular data of signal.The present invention is detected and corrected the mobile object (aircraft, ship) hidden in radar signal in terms of signal data and bearing data two according to cluster and abnormality detection theory, takes a firm foundation for the analysis of subsequent movement object behavior mode.
Description
Technical field
The invention belongs to technical field of computer information processing, are related to a kind of mobile object towards radar signal data
Detection method.
Background technique
Under normal conditions, signal analysis personnel need to sort a large amount of signal data, those are belonged to same
The signal of signal source sort out and analyze on this basis the behavioural characteristic of signal source.For the behavior to signal source
Feature has correct analysis, and signal sorting result needs higher accuracy, and not so subsequent analysis task can all be established
On the basis of a mistake.Had benefited from technology development in recent years, many signal acquiring systems being capable of automatic identification and classification one
A little signals, but the signal that these systems can only be very close to signal attribute sort out and in signal table with unique one
Row record identification signal source, if the attribute value of signal biggish fluctuation occurs and will be identified as another signal source
The signal of sending.Thus a record in signal table cannot represent a signal source completely, it would be desirable to by poly-
" signal source " that these are scattered by class method is sorted out.
It is just identified as the signal that multiple signal sources issue since signal attribute value differs more acquisition system, so single
Purely by signal attribute it is clustered can only rough trap signal, reduce quantity, be not effective to ensure that cluster result
Accuracy.In order to improve the accuracy of cluster result, we by means of signal bearing data.By believing in one cluster of detection
Number deflection variation whether link up to correct cluster result, to improve the accuracy of signal source detection.A signal herein
Source is likely to be an airplane or a steamer, therefore the present invention is directed to be carried out by radar signal data to these mobile objects
Accurate detection, the analysis for subsequent analysis personnel provide solid foundation.
Summary of the invention
Technical problem solved by the invention is in view of the deficiencies of the prior art, to provide one kind towards radar signal number
According to mobile object detection method, according to cluster and abnormality detection theory in terms of signal data and bearing data two to radar
The mobile object (aircraft, ship) hidden in signal is detected and corrected, and is the analysis of subsequent movement object behavior mode
It takes a firm foundation.
A kind of mobile object detection method towards radar signal data, comprising the following steps:
Step 1): cluster feature is chosen from radar signal data: calculating radar signal number first with histogram method
Then the frequency that each dimension occurs in strong point calculates the comentropy of each dimension according to information entropy principle, select comentropy compared with
Big dimension carries out subsequent cluster as cluster feature；
Step 2): it calculates the clustering parameter of DBScan: on the basis of the cluster feature selected by step 1), calculating each signal
The distance between data point and other signal data points simultaneously carry out ascending sort to calculated result, obtain identical as signal data point
The ascending order data set of quantity is then based on k- distance principle and selects value in fixation sorting position in each ascending order data set
And ascending sort is carried out, k- is obtained apart from value sequence, while using the serial number of fixed sorting position as DBScan clustering parameter
Min_samples parameter；Then k- distance Curve figure is drawn apart from value sequence according to k-, and maximum with slope of curve variation in scheming
Eps value of the y value at place as DBScan clustering parameter；
Step 3): cluster: cluster feature and calculated clustering parameter based on selection cluster signal data point；
Step 4): candidate outlier detection: the cluster result obtained according to step 3) searches all signals in each cluster
The direction angle information that data point records in radar signal data, and ascending order arrangement is carried out according to timestamp attribute, it is then based on
ARIMA method simulates deflection time series, carries out outlier detection finally by residual error, establishes for each cluster candidate abnormal
Point set；
Step 5): candidate abnormal point is assessed: the abnormal point set of candidate detected according to step 4), to the element in cluster
It is adjusted or the direction angular data of signal data point is adjusted.
A kind of mobile object detection method towards radar signal data, the step 1) the following steps are included:
Step 1.1): the frequency that each dimension occurs in radar signal data point is calculated using histogram method；
Step 1.2): calculating the entropy of each dimension according to comentropy calculation formula, and chooses the maximum l dimension of entropy
As subsequent cluster dimension.
A kind of mobile object detection method towards radar signal data, it is every calculating in the step 1.1)
When the frequency that a dimension occurs, for the dimension sample x of Category Attributes1,x2,…,xi,…,xn-1,xn, wherein n indicates the attribute
Data amount check, it is consistent with the data amount check in data set, count set T={ x of the sample elements1,x2,…,xi,…,xm-1,
xm, and its number c (x occurred in the sample is calculated to each of set T elementi), element in last set of computations
Approximation probability, i.e. its frequency occurred in the sample, calculation formula areFor the dimension sample of connection attribute
This is divided into m disjoint section R first by the value range discretization of attributem=(xi,xi+1] (i=0 ..., m-1), and
Signal data point is placed into corresponding section by comparing the size of the attribute value, further, counts the number in each section
According to number c (Rm), and calculate the probability in each section
A kind of mobile object detection method towards radar signal data, the dimension of connection attribute include: frequency,
Pulse period, pulse width, pulse period maximum value, pulse period minimum value, maximum arteries and veins insied width and most scun insied width, from
The dimension for dissipating attribute includes: bandwidth types, frequency type, pulse period type, pulse width type.
A kind of mobile object detection method towards radar signal data, it is every calculating in the step 1.2)
When entropy H (X) of a dimension, the dimension of Category Attributes is usedIt is counted
It calculates, the dimension of connection attribute is usedIt is calculated.
A kind of mobile object detection method towards radar signal data, specific step is as follows for the step 2):
Step 2.1): the number based on the selected dimension as cluster feature of step 1) calculates each data point and residue
The distance between data point Wherein di,jTable
Registration strong point xiAnd xjBetween Euclidean distance, l indicates dimension number as cluster feature, the number of n expression data point, so
Afterwards to calculated result DiCarry out ascending sort, the data set after being sortedIt indicates；
Step 2.2): the data set obtained according to step 2.1)Each data set is selected based on k- distance principle
The value of middle sequence the 4th, i.e., take 4 for the k value of k- distance principle, and carry out ascending sort and obtain k- apart from value sequence；
Step 2.3): according to k- obtained by step 2.2) apart from value sequence, with k-, the value of each element is in value sequence
Ordinate draws k- distance Curve figure, and calculate the slope of curve in figure using the position of each element in the sequence as abscissa
Change maximum y value as required Eps value；Secondly, user can also voluntarily choose Eps value according to the figure；
Step 2.4): min_samples parameter setting is consistent with k value, that is, is set as 4.
A kind of mobile object detection method towards radar signal data, the step 3) the following steps are included:
According to the clustering parameter that the dimension of step 1) selection and step 2) obtain, data are gathered using DBScan algorithm
Class；Gained cluster result is { c1,c2,…,ct,…,ck, wherein 1≤t≤k.
A kind of mobile object detection method towards radar signal data, the step 4) the following steps are included:
Step 4.1): with the cluster result of step 3) for { c1,c2,…,ct,…,ck, wherein 1≤t≤k, in direction table
The deflection of all signals records information in the middle each cluster of inquiry；WithTable
Show cluster ctIn all signals deflection record information and arranged according to the timestamp attribute ascending order of respective record, wherein fiTable
Show the deflection value of some signal, nctIndicate cluster ctIn all sense angles record quantity；
Further, to cluster ctIn the direction angle information of each signal counted respectively, by FctIn direction angle value root
Classify according to its affiliated signal, classification results are Wherein mctIt indicates
The number of signal in the cluster, the i.e. number of element；Indicate the deflection data sequence an of signal.And with CFIndicate one
The deflection of a signal records number, for step 5) use；
Step 4.2): it is carried out abnormality detection based on deflection time series data of the ARIMA method to each cluster:
Deflection time series is simulated using ARIMA method, it is first determined tri- parameters of p, d, q, wherein d is indicated
The difference number done when time series is become stationary sequence is examined by ADF and determines whether time series is steady to obtain
Parameter d, parameter p and parameter q determine possible value range by ACF figure, PACF figure first, and then choosing takes AIC and BIC
It is worth the smallest p and q as parameter, the formula of AIC are as follows: the formula of AIC=-2ln (L)+2k, BIC are as follows: BIC=-2ln (L)+ln
(n) * k, wherein L is the maximum likelihood of model when (p, d, q) takes certain class value, and nn is data bulk, and kk is the variable of the model
Number；With the deflection time series data of some cluster for { f1,f2,…,fn, the simulated series obtained by ARIMA (p, d, q)
For { f '1,f′2,…,f′n, then its residual sequence is { e1,e2,…,en, wherein en=fn-f′n, finally calculate residual sequence
Mean valueAnd varianceIf | ei- μ | 3 σ of >, then it is assumed that corresponding fiIt is different
Chang Dian；
Step 4.3): through step 4.2), cluster c is obtainedtAbnormal point set beWherein ntIndicate the number of candidate abnormal point in the cluster.
A kind of mobile object detection method towards radar signal data, the step 5) the following steps are included:
Step 5.1): step 4.3) acquired results are based on, c will be clustered according to the affiliated signal of abnormal pointtCandidate abnormal point
Set NtClassify；Classification results are expressed asWherein mtIndicate all times
Select the quantity of the belonged to signal of abnormal point, a candidate abnormal point subset X1={ f1,f2,…,fi,…,fx1-1,fx1In it is different
Normal point fiA signal is belonged to, x1 is indicated in NtThe abnormal point number of the middle same signal of ownership；And use CnfIndicate a letter
Number abnormal point number；
4.1) and 5.1) step 5.2): according to statistical result, comparison cluster ctIn the same signal deflection record
Number CFWith abnormal point quantity Cnf.IfThen think that the signal is not belonging to the cluster, signal is deleted from cluster
It removes；IfThen think that the part deflection of the signal is recorded as noise, it is removed from signal；
Step 5.3): circulation executes step 5 until traversing all clusters.
The technical effects of the invention are that according to cluster and abnormality detection theory from two sides of signal data and bearing data
The mobile object (aircraft, ship) hidden in radar signal is detected and corrected, and is subsequent movement object behavior mould
The analysis of formula is taken a firm foundation.
Detailed description of the invention
Fig. 1 is the method for the invention flow chart；
Fig. 2 is radar signal clustering views；
Fig. 3 is EPS curve view.
Specific embodiment
To keep the purpose of the present invention, mentality of designing and advantage clearer, below in conjunction with specific example, and referring to attached drawing,
Invention is further described in detail.
The present invention comprises the steps of:
Step 1): cluster feature is chosen.Since signal data feature is more, and some of them feature may be noise number
According to not contributing cluster result not only it is also possible to reduce Clustering Effect, therefore calculate each dimension according to information entropy principle
Comentropy, select the biggish dimension of comentropy to carry out subsequent cluster；
Step 2): the clustering parameter of DBScan is calculated.The clustering parameter of DBScan is calculated by k- distance method；
Step 3): cluster.Cluster feature and calculated clustering parameter based on selection cluster signal data；
Step 4): candidate outlier detection.Abnormal inspection is carried out based on deflection timing information of the ARIMA method to signal
It surveys, establishes candidate abnormal point set for each cluster；
Step 5): candidate abnormal point is assessed.The abnormal point set of candidate detected according to step 4), to the element in cluster
It is adjusted or the direction angular data of signal is adjusted.
The step 1) the following steps are included:
Step 1.1): the present invention institute towards radar signal data be made of two tables, wherein one be signal table, the table
Have recorded the primary attribute value an of logical signal, i.e., by the same signal source (such as: aircraft or ship) issue signal frequency,
The attributes such as pulse period；Other one is direction table, which has recorded a logical signal (such as: aircraft or ship) in signal table
Deflection change information, signal table and the table are one-to-many relationship.
Since there are errors for receiving system, it is possible to by the same signal source S1The signal identification of sending is multiple letters
Number source S1,S1′,S″1... the signal of sending, and the different record of a plurality of signal id is stored as (with signal source S in signal table1,
S′1,S″1... number it is consistent).In order to help signal analysis personnel to have correct understanding, Wo Menxu to the signal of the same signal source
Identify that these substantially belong to the signal of the same signal source.Herein it should be noted that, although system is not known correctly
Level signal source (S1,S′1,S″1...) it is classified as one kind, but directional information of each signal source in the table of direction is by the letter
Generation is moved in number source itself, there is no the directional information of other signal sources is identified mistake, and is stored under the signal source
Situation.
In order to help signal analysis personnel to have correct understanding to " behavioural characteristic " of the same signal source, it would be desirable to letter
" signal " in number table carries out clustering --- " signal " that those itself belong to the same signal source is classified as one kind.Due to
The feature recorded in signal table is more, and some of them feature may be noise data, does not contribute cluster result not only
It is also possible to reducing Clustering Effect, it would be desirable to which selecting those includes that the more feature of information is clustered.In this, it is assumed that should
Data set has n data.
Step 1.2): the probability distribution of each dimension in radar direction data is calculated using histogram method.Herein I
Need to distinguish Category Attributes and connection attribute.For Category Attributes, it is assumed that its sample is x1,x2,…,xi,…,xn-1,xn,
Middle n indicates that the data amount check of the attribute is consistent with the data amount check in data set.Count the set T={ x of the sample elements1,
x2,…,xi,…,xm-1,xm, and it is calculated to each of set T element and reppears existing number c (x in samplei), finally
The approximation probability of element in set of computations, i.e. its frequency occurred in the sample, calculation formula areFor even
Continuous attribute, the value range discretization of the attribute need to be divided into m disjoint section R by usm=(xi,xi+1] (i=0 ...,
M-1), and by data point by comparing the size of the attribute value it is placed into corresponding section.Further, it counts in each section
Data amount check c (Rm), and calculate the probability in each section
Step 1.3): according to comentropy calculation formulaOrThe entropy of each dimension is calculated, and chooses the maximum l dimension of entropy and makees
For subsequent cluster dimension.Meanwhile Visual Interactive system also supports the customized cluster dimension of user.
The step 2) the following steps are included:
Step 2.1): it based on subsequent cluster dimension selected by step 1), calculates between each data point and remainder strong point
DistanceWherein di,jIndicate data point xi
And xjBetween Euclidean distance, n indicates the number of data point, and l indicates dimension number as cluster feature, for example in step
1.3) maximum 3 dimensions of entropy are taken in, then l is 3 herein, that is, calculates corresponding Euclidean to these three dimensions of data point
Distance.Then to calculated result DiAscending sort is carried out, we use hereinIndicate the data set after sequence；
Step 2.2): the ascending order data set obtained according to step 2.1)Every number is selected based on k- distance principle
According to the value and composition data collection for concentrating sequence the 4th.Further, ascending sort is carried out to the data set and obtains k- apart from value sequence；
Step 2.3): according to k- obtained by step 2.2) apart from value sequence, k- distance Curve figure is drawn.Further according to the figure
The maximum place of calculated curve slope variation, y value is required Eps value herein；Secondly, user can also voluntarily choose according to the figure
Eps value；
Step 2.4): min_samples parameter setting is consistent with k value under normal conditions, therefore is equally set herein
It is set to 4.
The step 3) the following steps are included:
According to the dimension that step 1.3) selects, the clustering parameter that step 2) obtains carries out data using DBScan algorithm
Cluster；
Assuming that gained cluster result is { c after above-mentioned all steps1,c2,…,ct,…,ck, wherein 1≤t≤
k。
The step 4) the following steps are included:
Step 4.1): it is based on the cluster result { c of step 3)1,c2,…,ct,…,ck, it is inquired in deflection table each
The deflection of all signals records information in cluster；It enables hereinIndicate cluster
ctIn all signals deflection record information and arranged according to the timestamp attribute ascending order of respective record, wherein fiIndicate some
The deflection value of signal, nctIndicate cluster ctIn all sense angles record quantity.
Further, to cluster ctIn the direction angle information of each signal counted respectively.By FctIn direction angle value root
Classify according to its affiliated signal, classification results are Wherein mctIt indicates
The number of signal in the cluster, the i.e. number of element；Indicate the deflection sequence an of signal.Enable CFIndicate the side of signal
Number is recorded to angle, for step 5) use；
Step 4.2): it is carried out abnormality detection according to deflection time series data of the small wave converting method to each cluster.
Deflection time series is simulated first with ARIMA method.There are (p, d, q) three parameters for this method
It needs to be determined that wherein d indicates the difference number done when time series is become stationary sequence, whether time series can steadily lead to
It crosses ADF and examines determination, so that parameter d is obtained, and p and q can then be schemed by ACF figure, PACF respectively, AIC criterion and BIC criterion phase
In conjunction with mode determine, i.e., the possible value range of p and q is determined by ACF figure, PACF figure first, then choosing makes AIC
With BIC value the smallest p and q as parameter.The formula of AIC are as follows: the formula of AIC=-2ln (L)+2k, BIC are as follows: BIC=-2ln
(L)+ln (n) * k, wherein L is the maximum likelihood of model when (p, d, q) takes certain class value, and n is data bulk, and k is the model
Variable number.Assuming that the deflection time series data of some cluster is { f1,f2,…,fn, the mould obtained by ARIMA (p, d, q)
Quasi-ordering is classified as { f '1,f′2,…,f′n, then its residual sequence is { e1,e2,…,en, wherein en=fn-f′n.Finally calculate residual error
The mean value of sequenceAnd varianceIf | ei- μ | 3 σ of >, then it is assumed that corresponding
fiFor abnormal point.In the ARIMA method referred to, p, d, q are the parameters of ARIMA, when these three parameters determine after, ARIMA this
A model can be used to fit time sequence, such as original time series is (t1, f1), (t2, f2) ..., wherein t table
Show the time, f indicates deflection.It is determined that t can be transmitted to this model after p, d of ARIMA model, q parameter, count
Calculate f '=arima (t).
Step 4.3): assuming that after this step, c is clusteredtAbnormal point set be Wherein ntIndicate the number of candidate abnormal point in the cluster.
The step 5) the following steps are included:
Step 5.1): step 4.3) acquired results are based on, c will be clustered according to the affiliated signal of abnormal pointtCandidate abnormal point
Set NtClassify.Classification results are represented byWherein mtIndicate all
The quantity of candidate the belonged to signal of abnormal point, a candidate abnormal point subset X1={ f1,f2,…,fi,…,fx1-1,fx1In
Abnormal point fiA signal is belonged to, x1 is indicated in NtThe abnormal point number of the middle same signal of ownership.It is not specific for some
Signal, we use C under common situationnfIndicate the abnormal point number of signal；
4.1) and 5.1) step 5.2): according to statistical result, comparison cluster ctIn the same signal deflection record
Number CFWith abnormal point quantity Cnf.IfThen think that the signal is not belonging to the cluster, signal is deleted from cluster
It removes；IfThen think that the part deflection of the signal is recorded as noise, it is removed from signal；
Step 5.3): enabling t=1, recycles all clusters until t=k terminates.
Claims (9)
1. a kind of mobile object detection method towards radar signal data, which comprises the following steps:
Step 1): cluster feature is chosen from radar signal data: calculating radar signal data point first with histogram method
In the frequency that occurs of each dimension, the comentropy of each dimension is then calculated according to information entropy principle, selects comentropy biggish
Dimension carries out subsequent cluster as cluster feature；
Step 2): it calculates the clustering parameter of DBScan: on the basis of the cluster feature selected by step 1), calculating each signal data
The distance between point and other signal data points simultaneously carry out ascending sort to calculated result, obtain quantity identical as signal data point
Ascending order data set, be then based on the value that k- distance principle selects in fixation sorting position in each ascending order data set and go forward side by side
Row ascending sort obtains k- apart from value sequence, while using the serial number of fixed sorting position as the min_ of DBScan clustering parameter
Samples parameter；Then k- distance Curve figure is drawn apart from value sequence according to k-, and with slope of curve variation maximum in scheming
Eps value of the y value as DBScan clustering parameter；
Step 3): cluster: cluster feature and calculated clustering parameter based on selection cluster signal data point；
Step 4): candidate outlier detection: the cluster result obtained according to step 3) searches all signal datas in each cluster
The direction angle information that point records in radar signal data, and ascending order arrangement is carried out according to timestamp attribute, it is then based on ARIMA
Method simulates deflection time series, carries out outlier detection finally by residual error, establishes candidate abnormal point set for each cluster
It closes；
Step 5): assess candidate abnormal point: the abnormal point set of candidate detected according to step 4) carries out the element in cluster
Adjustment is adjusted the direction angular data of signal data point.
2. a kind of mobile object detection method towards radar signal data according to claim 1, which is characterized in that institute
State step 1) the following steps are included:
Step 1.1): the frequency that each dimension occurs in radar signal data point is calculated using histogram method；
Step 1.2): calculating the entropy of each dimension according to comentropy calculation formula, and chooses the maximum l dimension conduct of entropy
Subsequent cluster dimension.
3. a kind of mobile object detection method towards radar signal data according to claim 2, which is characterized in that institute
It states in step 1.1), when calculating the frequency that each dimension occurs, for the dimension sample x of Category Attributes1, x2...,
xi..., xn-1, xn, wherein n indicates the data amount check of the attribute, it is consistent with the data amount check in data set, count the sample
Set T={ the x of element1, x2..., xi..., xm-1, xm, and it is calculated to each of set T element and is occurred in the sample
Number c (xi), the approximation probability of element, i.e. its frequency occurred in the sample, calculation formula are in last set of computationsFor the dimension sample of connection attribute, first by the value range discretization of attribute, it is a non-intersecting to be divided into m
Section Rm=(xi, xi+1] (i=0 ..., m-1), and signal data point is placed into phase by comparing the size of the attribute value
The section answered further counts the data amount check c (R in each sectionm), and calculate the probability in each section
4. a kind of mobile object detection method towards radar signal data according to claim 3, which is characterized in that even
The dimension of continuous attribute includes: frequency, pulse period, pulse width, pulse period maximum value, pulse period minimum value, maximum arteries and veins
Insied width and most scun insied width, the dimension of Category Attributes include: that bandwidth types, frequency type, pulse period type, pulse are wide
Spend type.
5. a kind of mobile object detection method towards radar signal data according to claim 3, which is characterized in that institute
It states in step 1.2), when calculating entropy H (X) of each dimension, the dimension of Category Attributes is usedIt is calculated, the dimension of connection attribute is usedIt is calculated.
6. a kind of mobile object detection method towards radar signal data according to claim 1, which is characterized in that institute
Stating step 2), specific step is as follows:
Step 2.1): the number based on the selected dimension as cluster feature of step 1) calculates each data point and remaining data
The distance between point Wherein dI, jIndicate number
Strong point xiAnd xjBetween Euclidean distance, l indicates that dimension number as cluster feature, the number of n expression data point are then right
Calculated result DiCarry out ascending sort, the data set after being sortedIt indicates；
Step 2.2): the data set obtained according to step 2.1)It is selected in each data set and is sorted based on k- distance principle
The k value of k- distance principle is taken 4 by the 4th value, and carry out ascending sort and obtain k- apart from value sequence；
Step 2.3): being vertical sit with the value of k- each element in value sequence according to k- obtained by step 2.2) apart from value sequence
Mark draws k- distance Curve figure using the position of each element in the sequence as abscissa, and calculates the slope of curve in figure and change
Maximum y value is as required Eps value；Secondly, user can also voluntarily choose Eps value according to the figure；
Step 2.4): min_samples parameter setting is consistent with k value, that is, is set as 4.
7. a kind of mobile object detection method towards radar signal data according to claim 1, which is characterized in that institute
State step 3) the following steps are included:
According to the clustering parameter that the dimension of step 1) selection and step 2) obtain, data are clustered using DBScan algorithm；
Gained cluster result is { c1, c2..., ct..., ck, wherein 1≤t≤k.
8. a kind of mobile object detection method towards radar signal data according to claim 1, which is characterized in that institute
State step 4) the following steps are included:
Step 4.1): with the cluster result of step 3) for { c1, c2..., ct..., ck, wherein 1≤t≤k, is looked into the table of direction
Ask the deflection record information of all signals in each cluster；WithIndicate poly-
Class ctIn all signals deflection record information and arranged according to the timestamp attribute ascending order of respective record, wherein fiIndicate certain
The deflection value of a signal, nctIndicate cluster ctIn all sense angles record quantity；
Further, to cluster ctIn the direction angle information of each signal counted respectively, by FctIn direction angle value according to it
Affiliated signal is classified, and classification results are Wherein mctIndicate that this is poly-
The number of signal in class, the i.e. number of element；Indicate the deflection data sequence an of signal；And with CFIndicate a letter
Number deflection record number；
Step 4.2): it is carried out abnormality detection based on deflection time series data of the ARIMA method to each cluster:
Deflection time series is simulated using ARIMA method, it is first determined tri- parameters of p, d, q, wherein d indicate by when
Between sequence become the difference number done when stationary sequence, examined by ADF and determine whether time series steady to obtain parameter
D, parameter p and parameter q determine possible value range by ACF figure, PACF figure first, and then choosing makes AIC and BIC value most
Small p and q is as parameter, the formula of AIC are as follows: the formula of AIC=-2ln (L)+2k, BIC are as follows: BIC=-2ln (L)+ln (n) *
K, wherein L is the maximum likelihood of model when (p, d, q) takes certain class value, and n is data bulk, and k is the variable number of the model；With
The deflection time series data of some cluster is { f1, f2..., fn, it is { f ' by the simulated series that ARIMA (p, d, q) is obtained1,
f′2..., f 'n, then its residual sequence is { e1, e2..., en, wherein en=fn-f′n, finally calculate the mean value of residual sequenceAnd varianceIf | ei- μ | 3 σ of >, then it is assumed that corresponding fiFor abnormal point；
Step 4.3): through step 4.2), cluster c is obtainedtAbnormal point set beWherein ntIndicate the number of candidate abnormal point in the cluster.
9. a kind of mobile object detection method towards radar signal data according to claim 1, which is characterized in that institute
State step 5) the following steps are included:
Step 5.1): step 4.3) acquired results are based on, c will be clustered according to the affiliated signal of abnormal pointtThe abnormal point set N of candidatet
Classify；Classification results are expressed asWherein mtIndicate that all candidates are different
The often quantity of the belonged to signal of point, a candidate abnormal point subset X1={ f1, f2..., fi..., fx1-1, fx1In exception
Point fiA signal is belonged to, x1 is indicated in NtThe abnormal point number of the middle same signal of ownership；And use CnfIndicate a signal
Abnormal point number；
4.1) and 5.1) step 5.2): according to statistical result, comparison cluster ctIn the same signal deflection record number CFWith
Abnormal point quantity Cnf.IfThen think that the signal is not belonging to the cluster, signal is deleted from cluster；Such as
FruitThen think that the part deflection of the signal is recorded as noise, it is removed from signal；
Step 5.3): circulation executes step 5 until traversing all clusters.
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 true CN109447163A (en) | 2019-03-08 |
| CN109447163B 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) |
Cited By (1)
| 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)
| 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 |
-
2018
- 2018-11-01 CN CN201811296013.2A patent/CN109447163B/en active Active
Patent Citations (8)
| 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)
| Title |
|---|
| ERICH SCHUBERT等: "DBSCAN Revisited, Revisited: Why and How You Should (Still) Use DBSCAN", 《ACM TRANSACTIONS ON DATABASE SYSTEMS》 * |
| KENSHI SAHO等: "Velocity/Shape Estimation Algorithm Using Tracking Filter and Data Fusion of Dual Doppler Radar Interferometers", 《INTERNATIONAL JOURNAL OF COMPUTER AND ELECTRICAL ENGINEERING》 * |
| R. S. KUMAR等: "A study on prediction of rainfall using datamining technique", 《2016 INTERNATIONAL CONFERENCE ON INVENTIVE COMPUTATION TECHNOLOGIES (ICICT)》 * |
| 刘冬利: "舰载雷达探测精度标定方法及关键技术研究", 《中国博士学位论文全文数据库·工程科技Ⅱ辑》 * |
| 张胜等: "信息熵时序和树图用于NetFlow可视化的研究", 《高技术通讯》 * |
| 张荣等: "一种基于近类点和模糊点的未知雷达信号分选算法", 《航船电子对抗》 * |
| 王杰贵等: "现代雷达信号分选技术综述", 《雷达科学与技术》 * |
Cited By (1)
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109447163B (en) | 2022-03-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10936906B2 (en) | Training data acquisition method and device, server and storage medium | |
| CN105027162B (en) | Image analysis apparatus, image analysis system, method for analyzing image | |
| CN102105901B (en) | Annotating images | |
| CN105243388B (en) | Waveform classification based on dynamic time warping and partitioning algorithm | |
| CN109409400A (en) | Merge density peaks clustering method, image segmentation system based on k nearest neighbor and multiclass | |
| CN104392250A (en) | Image classification method based on MapReduce | |
| CN112986925B (en) | Radar pulse sequence sorting method based on image features | |
| US20180210897A1 (en) | Model generation method, word weighting method, device, apparatus, and computer storage medium | |
| CN105989001B (en) | Image search method and device, image search system | |
| CN114116829A (en) | Abnormal data analysis method, abnormal data analysis system, and storage medium | |
| CN112529945A (en) | Registration method for multi-view three-dimensional ISAR scattering point set | |
| JP6879749B2 (en) | Support device and support method | |
| Pakgohar et al. | A comparative study of hard clustering algorithms for vegetation data | |
| CN109447163A (en) | A kind of mobile object detection method towards radar signal data | |
| Paturkar et al. | Plant trait segmentation for plant growth monitoring | |
| CN113987243A (en) | Image file gathering method, image file gathering device and computer readable storage medium | |
| Pinto et al. | Are vision transformers always more robust than convolutional neural networks? | |
| CN105354845A (en) | Method for semi-supervised detection on changes in remote sensing images | |
| CN107423319B (en) | Junk web page detection method | |
| Minija et al. | Image processing based Classification and Segmentation using LVS based Multi-Kernel SVM | |
| CN110502660B (en) | Multi-distance measurement image retrieval method under weak supervision | |
| CN110781963B (en) | K-means clustering-based aerial target clustering method | |
| Zubair et al. | Content-based image retrieval system using second-order statistics | |
| Wang et al. | FCM algorithm and index CS for the signal sorting of radiant points | |
| CN113313138A (en) | Intrusion behavior feature conversion method and detection method based on probability generation model |
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 |