CN106597425A - Radar object positioning method based on machine learning - Google Patents
Radar object positioning method based on machine learning Download PDFInfo
- Publication number
- CN106597425A CN106597425A CN201611032540.3A CN201611032540A CN106597425A CN 106597425 A CN106597425 A CN 106597425A CN 201611032540 A CN201611032540 A CN 201611032540A CN 106597425 A CN106597425 A CN 106597425A
- Authority
- CN
- China
- Prior art keywords
- target
- radar
- orientation
- neutral net
- dual pathways
- 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
- 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
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/02—Systems using reflection of radio waves, e.g. primary radar systems; Analogous systems
- G01S13/06—Systems determining position data of a target
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention proposes a radar object positioning method based on machine learning. According to one embodiment of the invention, the method comprises the steps: carrying out the range-direction pulse compression of a target echo signal received by an along-track two-channel SAR (synthetic aperture radar), obtaining an interference phase matrix, sequentially inputting the interference phase vectors corresponding to all distance units in the interference phase matrix into a neural network, and obtaining a result of judgment whether each distance unit has a moving object or not and the orientation position x0 of the moving object. According to the embodiment of the invention, the method avoids problems of complex orientation pulse compression, radial speed estimation and faced phase fuzziness of a conventional ATI (along-track interference) method, thereby solving a problem that the positioning precision of the moving object is decreased because of error accumulation in the above process. The method is higher in positioning precision of the moving object, is lower in time consumption, and can meet the demands of high instantaneity and high positioning precision.
Description
Technical field
The application is related to signal processing technology field, and in particular to Radar Targets'Detection and identification field, more particularly to one
Plant based on the radar target localization method of machine learning, the method can be used for radar moving targets detection, moving targets location and tracking
Deng.
Background technology
Existing radar system is based on more strict theoretical model and traditional signal processing means, is primarily present following asking
Topic:(1) complex environment modeling is difficult, and systematic error accumulation, high robustness is difficult to;(2) linear period processing mode is (such as Fu
In leaf transformation etc.), ambiguity solution (range ambiguity, doppler ambiguity, phase ambiguity, velocity ambiguity etc.) needs complicated computing and high
Expensive hardware cost;(3) radar system is complicated, and poor universality, portability are weaker, and the upgrading cycle is long.
In recent years, with the lifting of big data and computing capability, started again with machine learning, deep learning as representative
Artificial intelligence's tide.Using machine learning techniques, radar processing procedure is equivalent to into black box, by training neutral net, directly
Connect and build raw radar data or apart from pulse pressure numeric field data and the non-linear relation of output information, be capable of achieving following functions:(1) profit
It is trained with the data comprising environment and systematic error, improves system robustness;(2) Nonlinear Processing ambiguity solution;(3) system
Framework is presented modularity, portable strong, achievable quick upgrading.Therefore, radar target of the research based on machine learning is positioned
Method is significant.
The content of the invention
The purpose of the application is the deficiency for above-mentioned prior art, proposes a kind of radar target based on machine learning
Localization method, to improve the robustness and Rapid transplant ability of system realization.
This application provides a kind of radar target localization method based on machine learning, methods described includes:To utilizing edge
The target echo signal that flight path dual pathways synthetic aperture radar SAR is receivedWithEnter row distance to pulse pressure process,
The dual pathways is obtained apart from pulse pressure domain signalWithWherein,For fast time, tmFor the slow time,
R0For the minimum distance of target to radar platform running track, x0It is the target in tm=0 moment is relative to the radar platform
Position of orientation;By the dual pathways apart from pulse pressure domain signalWithCarry out conjugate multiplication,
Obtain interference matrixWherein,It isConjugate matrices;To the interference matrixCarry out taking phase operation, be calculated interference
Phasing matrixWherein, angle [] is represented and is taken phase operation;Will be described
Interferometric phase matrixIn the corresponding interferometric phase vector of each range cell sequentially input neutral net, obtain
Position of orientation x of target movement properties value M and the target of respective distances unit relative to the radar platform0, that is, obtain
Each range cell is with the presence or absence of moving target and the position of orientation of moving target.
In certain embodiments, the radar target localization method based on machine learning, also includes:Build in advance and instruct
Practice based on the neutral net of error back-propagating BP.
In certain embodiments, the advance structure and the neutral net based on error back-propagating BP is trained, bag
Include following steps:
(1) target component (x of target setting0,R0,vr,va) and along the system of flight path dual pathways synthetic aperture radar SAR
Parameter, wherein, the target refers to the target for training the neutral net, comprising moving target and static target, R0For
The target to radar platform running track minimum distance, x0It is the target in tm=0 moment is relative to the radar platform
Position of orientation, tmFor slow time, vrAnd vaThe radial velocity and orientation speed of respectively described target;
(2) according to the target component and the systematic parameter, the dual pathways apart from pulse pressure domain for generating the target is imitated
True signalWithAnd target movement properties value is calculated, wherein,For the fast time, if described
Radial velocity v of targetrWith orientation speed vaIt is simultaneously zero, then sets target movement properties value M=0, otherwise, M=is set
1;
(3) signal is emulated according to the dual pathways of the targetWithIt is calculated described
The data vector s of target place range cell1′(tm,x0) and s2′(tm,x0);
(4) by s1′(tm,x0) and s '2(tm,x0) conjugate multiplication, calculate the interferometric phase vector of the target;
(5) neutral net is built, calculates length L of the interferometric phase vector of the target, set the nerve net
The number of the input node of network is equal to L, the number of output node is equal to the 2, number of hidden layer node and is equal toIts
In,Extraction of square root computing is represented, floor () represents downward rounding operation, initializes frequency of training T=0;
(6) using the interferometric phase vector as the neutral net input data, by target movement properties value M
With the target relative to the radar platform position of orientation x0As the label data of the neutral net, will M and x0Make
For the output of the neutral net, a neural metwork training is completed using steepest descent method, update frequency of training T=T+1;
(7) if T<L2, then target component (the x is updated0,R0,vr,va) value, systematic parameter is constant, according to step
(2) the movement properties value and interferometric phase vector of the target are regenerated to step (4), is performed once according still further to step (6)
Neural metwork training;Otherwise, the training of the neutral net is completed.
In certain embodiments, it is described to be according to the target component and along flight path dual pathways synthetic aperture radar SAR
System parameter, the dual pathways apart from pulse pressure domain for generating the target emulates signalWithBag
Include:
(1) do not consider that range migration affects caused by the speed due to the target, and according to equation below the mesh is calculated
Mark to instantaneous oblique distance R of the radar platform1And R2:
Wherein, R0For the minimum distance of the target to the radar platform running track, vrFor the radial direction speed of the target
Degree, tmFor slow time, x0It is the target in tm=0 moment is relative to radar platform position of orientation, vsFor the radar platform
The speed of service, vaFor the orientation speed of the target, d is spacing twin-channel along flight path.
(2) dual pathways apart from pulse pressure domain for calculating the target according to equation below emulates signalWith
Wherein, A is the signal amplitude in the target range pulse pressure domain, BrTo launch linear FM signal bandwidth,For it is fast when
Between, c is the light velocity, and λ is the corresponding wavelength of transmitting centre carrier frequency, R1And R2For the instantaneous oblique distance of the target to radar platform,
J is the imaginary part of symbol, and π is pi, and sinc is SIN function, and exp is exponential function.
In certain embodiments, it is described that signal is emulated according to the dual pathways of the targetWithIt is calculated the data vector s of target place range cell1′(tm,x0) and s '2(tm,x0), including:
The data vector s of target place range cell is calculated according to equation below1′(tm,x0) and s2′(tm,
x0):
WhenWhen minimum,
WhenWhen minimum,
Wherein, | | represent the computing that takes absolute value.
A kind of radar target localization method based on machine learning that the application is provided has compared with prior art following
Advantage:
1) neutral net of the application constructs moving target apart from pulse pressure numeric field data and the nonlinear dependence of its position of orientation
System, it is to avoid the phase ambiguity that traditional ATI (Along-track interferometry, Along Track Interferometry) method is faced
Problem, with higher positioning precision.
2) the application in pulse pressure numeric field data directly from extracting moving target position of orientation information, it is to avoid traditional ATI
The processes such as the complicated orientation pulse pressure of method, radial velocity estimation, improve the real-time of moving target positioning.
3) the application in pulse pressure numeric field data directly from extracting moving target position of orientation information, it is to avoid traditional ATI
The complicated orientation pulse pressure of method, radial velocity such as estimate at the process, so as to solve above procedure in move caused by error accumulation
Target location accuracy declines problem.
Description of the drawings
By reading the detailed description made to non-limiting example made with reference to the following drawings, the application other
Feature, objects and advantages will become more apparent upon:
Fig. 1 is the flow chart of one embodiment of the radar target localization method based on machine learning of the application;
The deviations that Fig. 2 is the application to be realized to moving target positioning from tradition ATI methods under different radial velocities
Contrast simulation figure;
The required used time that Fig. 3 is the application to be realized to moving target positioning from tradition ATI methods under different radial velocities
Contrast simulation figure;
Fig. 4 is the flow chart of another embodiment of the radar target localization method based on machine learning of the application.
Specific embodiment
The application is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining related invention, rather than the restriction to the invention.It also should be noted that, in order to
Be easy to description, illustrate only in accompanying drawing to about the related part of invention.
It should be noted that in the case where not conflicting, the feature in embodiment and embodiment in the application can phase
Mutually combination.Below with reference to the accompanying drawings and in conjunction with the embodiments describing the application in detail.
Fig. 1 shows the flow chart of one embodiment of the radar target localization method based on machine learning of the application
100.The radar target localization method based on machine learning, comprises the following steps:
Step 101, to using along flight path dual pathways synthetic aperture radar SAR receive target echo signal enter row distance to
Pulse pressure process, obtains the dual pathways apart from pulse pressure domain signal matrix.
In the present embodiment, to using the target echo signal received along flight path dual pathways synthetic aperture radar SARWithEnter row distance to pulse pressure process, obtain the dual pathways apart from pulse pressure domain signal matrixWithMethod can resolve into following steps:
1) target echo signal is received using along flight path dual pathways synthetic aperture radar SARWith
Wherein,For fast time, tmFor the slow time, σ for target scattering coefficient,It is apart from window function, W (tm) be
Orientation window function, exp () is exponential function, and j is the imaginary part of symbol, and π is pi, and γ is the frequency modulation for launching linear FM signal
Rate, c is the light velocity, fcFor centre carrier frequency, R1And R2Represent that target is twin-channel instantaneous oblique along flight path to radar platform respectively
Away from:
Wherein, R0For the minimum distance of above-mentioned target to the running track of above-mentioned radar platform, vrFor the radial direction of above-mentioned target
Speed, x0It is above-mentioned target in tm=0 moment relative to above-mentioned radar platform position of orientation, tmFor slow time, vsFor above-mentioned thunder
Up to the speed of service of platform, vaFor the orientation speed of above-mentioned target, d is spacing twin-channel along flight path.
2) target echo signal to receivingWithEnter row distance to Fourier transform, obtain target away from
Descriscent frequency domain signal X1(fr,tm) and X2(fr,tm) be:
Wherein, A1(fr,tm) and A2(fr,tm) target range is respectively to frequency domain signal X1(fr,tm) and X2(fr,tm)
Amplitude, frFor frequency of distance.
3) by target range to frequency domain signal X1(fr,tm) and X2(fr,tm) distance is multiplied by respectively to adaptation function Sr
(fr), and by distance to inverse Fourier transform, obtain target range pulse pressure domain signal matrixWithFor:
Wherein, A for target range pulse pressure domain signal amplitude, BrTo launch linear FM signal bandwidth, λ is transmitting carrier wave
The corresponding wavelength of mid frequency, π is pi, and sinc () is SIN function, and exp () is exponential function, and distance is to matching
Function Sr(fr) representation formula be:
Step 102, conjugate multiplication is carried out by the above-mentioned dual pathways apart from pulse pressure domain signal matrix, obtains interference matrix.
In the present embodiment, by the above-mentioned dual pathways apart from pulse pressure domain signal matrixWith
Conjugate multiplication is carried out, interference matrix is obtainedFormula is expressed as:
Wherein,It isConjugate matrices.
Step 103, carries out taking phase operation to above-mentioned interference matrix, is calculated interferometric phase matrix.
In the present embodiment, to above-mentioned interference matrixCarry out taking phase operation, be calculated interference phase
Bit matrixFormula is expressed as:
Wherein, angle [] is represented and is extracted phase operation, and the extraction phase operation is prior art, be will not be described here.
Step 104, by the corresponding interferometric phase vector of each range cell in above-mentioned interference phasing matrix god is sequentially input
Jing networks, obtain the orientation position of target movement properties value M and above-mentioned target relative to above-mentioned radar platform of respective distances unit
Put x0, that is, each range cell is obtained with the presence or absence of moving target and the position of moving target.
In this embodiment, above-mentioned interference phasing matrixIn represent that each distance is single per a line or each row
The corresponding interferometric phase vector of unit, if the corresponding interferometric phase vector input of a certain range cell of interferometric phase matrix is refreshing
Jing networks, then the neutral net will export the range cell target movement properties value M and above-mentioned target relative to above-mentioned radar
Position of orientation x of platform0, wherein, the value of above-mentioned target movement properties value M is that 0 or 1, M=0 represent that the range cell is present
Static target, M=1 represents that the range cell has moving target.
In the present embodiment, it is the corresponding interferometric phase vector of each range cell in above-mentioned interference phasing matrix is defeated successively
Enter neutral net, obtain the side of target movement properties value M and above-mentioned target relative to above-mentioned radar platform of respective distances unit
Position position x0, that is, each range cell is obtained with the presence or absence of moving target and the position of moving target.
The advantage of the application can be further illustrated by following emulation data processing.
1. systematic parameter and target component are set
It is as shown in table 1 that systematic parameter is set:
The systematic parameter of table 1
The theoretical value of the target component of moving target is set:Radial velocity vrRespectively [- 20:0.5:20] m/s, i.e., with
0.5m/s is step-length, and interval range is [- 20,20] m/s, orientation speed vaFor 0, position of orientation x0For 100.3m, target is to thunder
Up to the minimum distance R of platform running track0For 9000m.
2. data processing
The echo data of target is generated according to the systematic parameter and target component of above-mentioned setting, according in flow process Figure 100
Step 101 generates the interferometric phase matrix of moving target to step 103, and the interferometric phase Input matrix of above-mentioned generation has been instructed
The neutral net perfected, the target movement properties value of Output simulation result, i.e. moving target and position of orientation, and record the process
Required used time tnet;Estimate moving target position of orientation using tradition ATI methods simultaneously, and record used time t needed for the processATI。
The basic step of existing traditional ATI methods positioning target is:
1) using target range pulse pressure domain signal matrixWithThrough following orientation arteries and veins
Pressure process:
Wherein,ifft
[] to represent and carry out Inverse Fast Fourier Transforms along orientation, and fft [] is represented carries out fast Fourier transform along orientation,Represent vectorLength value.
2) by y1′(i,tm;x0) and y2′(i,tm;x0) interference treatment is done, detect that moving target is located according to CFAR Methods
Position of orientation and parasang, be designated as respectivelyAnd I, and estimate moving target interferometric phase
WhenAnd during i=I
Wherein,It is y '2(i,tm;x0) conjugate function.
3) moving target radial velocity is estimated according to following formula:
4) moving target is calculated due to azimuth deviation amount △ x caused by radial velocity:
5) byMoving target is repositioned with △ x, obtaining its position of orientation is:
Moving target is positioned deviations comparing result using the application method and tradition ATI methods as shown in Fig. 2
Calculate used time comparing result as shown in Figure 3.
From the simulation result of Fig. 2, traditional ATI methods are affected by phase fuzzy problem, cause its positioning to occur larger
Deviation;And the application adopts Neural Network Based Nonlinear processing mode, there is no fuzzy problem, positioning precision is higher.
From the simulation result of Fig. 3, traditional ATI methods are needed through complicated orientation pulse pressure, radial velocity estimation, side
The steps such as position side-play amount estimation realize the positioning of moving target, and required time is longer;And the application is using the good god of off-line training
Jing networks, using moving target signal as neutral net input, direct output campaign target bearing position, operation efficiency compared with
It is high.
To sum up, less using the application processing mode operand, estimated accuracy is higher, there is no fuzzy problem.Therefore, originally
Application can meet high real-time and high position precision requirement.
With continued reference to Fig. 4, the flow process of another embodiment based on the radar target localization method of machine learning is shown
Figure 40 0.Mainly description builds in advance and trains the flow process of neutral net to the flow chart 400, comprises the following steps:
Step 401, the target component of target setting and the systematic parameter along flight path dual pathways synthetic aperture radar SAR.
In the present embodiment, the target component of target setting and the system along flight path dual pathways synthetic aperture radar SAR are joined
Number.Wherein, above-mentioned target refers to the target for training above-mentioned neutral net, the target can be moving target, or
Static target.
In the present embodiment, above-mentioned target component at least includes:x0、R0、vr、va.Wherein, R0It is flat to radar for above-mentioned target
The minimum distance of platform running track, x0It is above-mentioned target in tm=0 moment relative to above-mentioned radar platform position of orientation, tmFor
Slow time, vrAnd vaThe radial velocity and orientation speed of respectively above-mentioned target.As an example, radial velocity vrFor 4m/s, side
Position is to speed vaFor 0, position of orientation x0For [0:1:300] m, i.e., with step-length as 1m, interval range is [0,300] m.It is double along flight path
The systematic parameter of passage synthetic aperture radar SAR at least includes:Centre carrier frequency, signal bandwidth, synthetic aperture time, sampling
Frequency, the radar platform speed of service, along flight path dual pathways spacing, pulse recurrence frequency.Wherein, the value of systematic parameter can join
According to table 1.
Step 402, according to target component and systematic parameter, generates the dual pathways emulation letter apart from pulse pressure domain of above-mentioned target
Number, and calculate target movement properties value.
In the present embodiment, in above-mentioned target component, if radial velocity v of above-mentioned targetrWith orientation speed vaTogether
When be zero, then target movement properties value M=0 is set, to represent that above-mentioned target is static target;Otherwise, M=1 is set, to table
Show that above-mentioned target is moving target.
In the present embodiment, first, do not consider that range migration affects caused by the speed due to the target, according to as follows
Formula calculates instantaneous oblique distance R of above-mentioned target to above-mentioned radar platform1And R2:
Wherein, R0For the minimum distance of above-mentioned target to above-mentioned radar platform running track, vrFor the radial direction speed of above-mentioned target
Degree, tmFor slow time, x0It is above-mentioned target in tm=0 moment is relative to above-mentioned radar platform position of orientation, vsIt is flat for above-mentioned radar
The speed of service of platform, vaFor the orientation speed of above-mentioned target, d is spacing twin-channel along flight path.
Then, the dual pathways apart from pulse pressure domain for calculating above-mentioned target according to equation below emulates signal
With
Wherein, A is the signal amplitude in above-mentioned target range pulse pressure domain, BrTo launch linear FM signal bandwidth,For it is fast when
Between, c is the light velocity, and λ is the corresponding wavelength of transmitting centre carrier frequency, R1And R2For the instantaneous oblique distance of above-mentioned target to radar platform,
J is the imaginary part of symbol,π is pi, and sinc is SIN function, and exp is exponential function.
Step 403, according to the dual pathways of above-mentioned target signal is emulated, and is calculated the number of above-mentioned target place range cell
According to vector.
In this embodiment, the data vector s of above-mentioned target place range cell is calculated according to equation below1′(tm,
x0) and s '2(tm,x0):
WhenWhen minimum,
WhenWhen minimum,
Wherein, | | represent the computing that takes absolute value, data vector s1′(tm,x0) beWhen minimum,Value;Data vector s2′(tm,x0) beWhen minimum,Value.
Step 404, by the data vector conjugate multiplication of above-mentioned target place range cell, calculates the interference phase of above-mentioned target
Bit vector.
In the present embodiment, first, by the data vector s of above-mentioned target place range cell1′(tm,x0) and s '2(tm,
x0) conjugate multiplication, calculate the interference vector △ s ' (t of above-mentioned targetm,x0), formula is expressed as:
Wherein,It is s '2(tm,x0) conjugate matrices.
Then, to above-mentioned interference vector △ s ' (tm,x0) carry out taking phase operation, it is calculated interferometric phase vector
Step 405, builds neutral net, and initializes frequency of training.
In the present embodiment, the interferometric phase vector of above-mentioned target is calculatedLength L, set above-mentioned nerve net
The number of the input node of network is equal to L, the number of output node is equal to the 2, number of hidden layer node and is equal toIts
In,Extraction of square root computing is represented, floor () represents downward rounding operation, initializes frequency of training T=0.
Step 406, using above-mentioned interference phase vectors as neutral net input data, by above-mentioned target movement properties value
With target relative to radar platform position of orientation as the label data of above-mentioned neutral net, complete one using steepest descent method
Secondary neural metwork training, updates frequency of training.
In the present embodiment, by above-mentioned interference phase vectorsAs the input data of above-mentioned neutral net, will be upper
State position of orientation x of target movement properties value M and above-mentioned target relative to above-mentioned radar platform0As the mark of above-mentioned neutral net
Data are signed, will M and x0As the output of above-mentioned neutral net, a neural metwork training is completed using steepest descent method, more
New frequency of training T=T+1.
Whether step 407, training of judgement terminates.
In the present embodiment, if T<L2, then it represents that training is not over, and need to continue executing with step 408, otherwise represents right
Neutral net has completed training.
Step 408, updates target component.
In the present embodiment, in response to judging to be not over the training of neutral net in step 407, then update above-mentioned
Target component (the x of target0,R0,vr,va), then, according to step 402 to step 404 the target fortune of above-mentioned target is regenerated
Dynamic property value and interferometric phase vector, afterwards, perform a neural metwork training, finally, according to step 407 according to step 406
Judge whether the training of above-mentioned neutral net is terminated.
In the present embodiment, by building and training neutral net, moving target is obtained apart from pulse pressure numeric field data and its side
The non-linear relation of position position, it is to avoid the complicated orientation pulse pressure of traditional Along-track interferometry ATI methods, radial velocity are estimated to wait
Journey and its phase fuzzy problem for being faced, so as to reduce above procedure in moving target positioning precision caused by error accumulation
Decline problem, with higher moving target positioning precision.
Above description is only the preferred embodiment of the application and the explanation to institute's application technology principle.People in the art
Member should be appreciated that invention scope involved in the application, however it is not limited to the technology of the particular combination of above-mentioned technical characteristic
Scheme, while also should cover in the case of without departing from the inventive concept, is carried out by above-mentioned technical characteristic or its equivalent feature
Combination in any and other technical schemes for being formed.Such as features described above has similar work(with (but not limited to) disclosed herein
The technical scheme that the technical characteristic of energy is replaced mutually and formed.
Claims (5)
1. a kind of radar target localization method based on machine learning, it is characterised in that methods described includes:
To using the target echo signal received along flight path dual pathways synthetic aperture radar SARWithEnter row distance
To pulse pressure process, the dual pathways is obtained apart from pulse pressure domain signal matrixWithWherein,For fast
Time, tmFor slow time, R0For the minimum distance of target to radar platform running track, x0It is the target in tm=0 moment phase
For the position of orientation of the radar platform;
By the dual pathways apart from pulse pressure domain signal matrixWithConjugate multiplication is carried out, is obtained
Interference matrixWherein,It isConjugate matrices;
To the interference matrixCarry out taking phase operation, be calculated interferometric phase matrixWherein, angle [] is represented and is taken phase operation;
By the interferometric phase matrixIn the corresponding interferometric phase vector of each range cell sequentially input nerve
Network, obtains the position of orientation of target movement properties value M and the target relative to the radar platform of respective distances unit
x0, that is, each range cell is obtained with the presence or absence of moving target and the position of orientation of moving target.
2. the radar target localization method based on machine learning according to claim 1, it is characterised in that methods described is also
Including:
The neutral net based on error back-propagating BP is built and trained in advance.
3. the radar target localization method based on machine learning according to claim 2, it is characterised in that the advance structure
The neutral net based on error back-propagating BP is built and trained, is comprised the steps:
(1) target component (x of target setting0,R0,vr,va) and along the systematic parameter of flight path dual pathways synthetic aperture radar SAR,
Wherein, the target refers to the target for training the neutral net, comprising moving target and static target, R0For the mesh
Mark the minimum distance of radar platform running track, x0It is the target in tm=0 moment the orientation relative to the radar platform
Position, tmFor slow time, vrAnd vaThe radial velocity and orientation speed of respectively described target;
(2) according to the target component and the systematic parameter, the dual pathways emulation letter apart from pulse pressure domain of the target is generated
NumberWithAnd target movement properties value is calculated, wherein,For the fast time, if the target
Radial velocity vrWith orientation speed vaIt is simultaneously zero, then sets target movement properties value M=0, otherwise, M=1 is set;
(3) signal is emulated according to the dual pathways of the targetWithIt is calculated the target
The data vector s ' of place range cell1(tm,x0) and s '2(tm,x0);
(4) by s '1(tm,x0) and s '2(tm,x0) conjugate multiplication, calculate the interferometric phase vector of the target;
(5) neutral net is built, calculates length L of the interferometric phase vector of the target, set the neutral net
The number of input node is equal to L, the number of output node is equal to the 2, number of hidden layer node and is equal toWherein,Extraction of square root computing is represented, floor () represents downward rounding operation, initializes frequency of training T=0;
(6) using the interferometric phase vector as the neutral net input data, by target movement properties value M and institute
State position of orientation x of the target relative to the radar platform0As the label data of the neutral net, will M and x0As institute
The output of neutral net is stated, using steepest descent method a neural metwork training is completed, update frequency of training T=T+1;
(7) if T<L2, then target component (the x is updated0,R0,vr,va) value, systematic parameter is constant, according to step (2) extremely
Step (4) regenerates the movement properties value and interferometric phase vector of the target, and according still further to step (6) nerve net is performed
Network training;Otherwise, the training of the neutral net is completed.
4. the radar target localization method based on machine learning according to claim 3, it is characterised in that described according to institute
State target component and the systematic parameter along flight path dual pathways synthetic aperture radar SAR, generate the target apart from pulse pressure domain
The dual pathways emulates signalWithIncluding:
(1) do not consider that range migration affects caused by the speed due to the target, and the target is calculated extremely according to equation below
Instantaneous oblique distance R of the radar platform1And R2:
Wherein, R0For the minimum distance of the target to the radar platform running track, vrFor the radial velocity of the target,
tmFor slow time, x0It is the target in tm=0 moment is relative to the radar platform position of orientation, vsFor the radar platform
The speed of service, vaFor the orientation speed of the target, d is spacing twin-channel along flight path;
(2) dual pathways apart from pulse pressure domain for calculating the target according to equation below emulates signalWith
Wherein, A is the signal amplitude in the target range pulse pressure domain, BrTo launch linear FM signal bandwidth,For fast time, c
For the light velocity, λ is to launch the corresponding wavelength of centre carrier frequency, R1And R2For the instantaneous oblique distance of the target to radar platform, j is
The imaginary part of symbol, π is pi, and sinc is SIN function, and exp is exponential function.
5. the radar target localization method based on machine learning according to claim 4, it is characterised in that described according to institute
State the dual pathways emulation signal of targetWithIt is calculated target place range cell
Data vector s '1(tm,x0) and s '2(tm,x0), including:
The data vector s ' of target place range cell is calculated according to equation below1(tm,x0) and s '2(tm,x0):
WhenWhen minimum,
WhenWhen minimum,
Wherein, | | represent the computing that takes absolute value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611032540.3A CN106597425B (en) | 2016-11-18 | 2016-11-18 | A kind of radar target localization method based on machine learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611032540.3A CN106597425B (en) | 2016-11-18 | 2016-11-18 | A kind of radar target localization method based on machine learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106597425A true CN106597425A (en) | 2017-04-26 |
CN106597425B CN106597425B (en) | 2019-02-12 |
Family
ID=58591610
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611032540.3A Active CN106597425B (en) | 2016-11-18 | 2016-11-18 | A kind of radar target localization method based on machine learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106597425B (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107942659A (en) * | 2017-11-10 | 2018-04-20 | 宁夏杰唯智能科技有限公司 | Transmitting device control method and equipment |
CN108535730A (en) * | 2018-04-16 | 2018-09-14 | 青海大学 | A kind of Doppler weather radar solution velocity ambiguity method and system |
CN108931771A (en) * | 2018-06-06 | 2018-12-04 | 电子科技大学 | A kind of method for tracking target based on synthetic aperture radar image-forming technology |
CN109709536A (en) * | 2019-01-24 | 2019-05-03 | 电子科技大学 | A kind of SAR moving target detection method based on convolutional neural networks |
CN111161227A (en) * | 2019-12-20 | 2020-05-15 | 成都数之联科技有限公司 | Target positioning method and system based on deep neural network |
CN111199162A (en) * | 2020-01-11 | 2020-05-26 | 华南理工大学 | RFID reader fault self-adaptive positioning method |
CN112180338A (en) * | 2020-06-10 | 2021-01-05 | 四川九洲电器集团有限责任公司 | Holographic digital array radar target quantity estimation method and system |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103135100A (en) * | 2013-01-31 | 2013-06-05 | 西安电子科技大学 | Moving-target parameter estimation method of common-rail bistatic synthetic aperture radar (SAR) |
CN103364783A (en) * | 2013-07-04 | 2013-10-23 | 西安电子科技大学 | Moving target radial velocity non-fuzzy estimation method based on single-channel SAR (synthetic aperture radar) |
CN103630899A (en) * | 2013-03-29 | 2014-03-12 | 中国科学院电子学研究所 | Method for high-resolution radar compressed sensing imaging of moving object on ground |
CN105974414A (en) * | 2016-06-24 | 2016-09-28 | 西安电子科技大学 | High resolution spotlight SAR self-focusing imaging method based on two-dimensional self-focusing |
US9459344B1 (en) * | 2011-01-14 | 2016-10-04 | Lockheed Martin Corporation | Ship position and velocity using satellite ephemerides and radar range measurement of satellite |
-
2016
- 2016-11-18 CN CN201611032540.3A patent/CN106597425B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9459344B1 (en) * | 2011-01-14 | 2016-10-04 | Lockheed Martin Corporation | Ship position and velocity using satellite ephemerides and radar range measurement of satellite |
CN103135100A (en) * | 2013-01-31 | 2013-06-05 | 西安电子科技大学 | Moving-target parameter estimation method of common-rail bistatic synthetic aperture radar (SAR) |
CN103630899A (en) * | 2013-03-29 | 2014-03-12 | 中国科学院电子学研究所 | Method for high-resolution radar compressed sensing imaging of moving object on ground |
CN103364783A (en) * | 2013-07-04 | 2013-10-23 | 西安电子科技大学 | Moving target radial velocity non-fuzzy estimation method based on single-channel SAR (synthetic aperture radar) |
CN105974414A (en) * | 2016-06-24 | 2016-09-28 | 西安电子科技大学 | High resolution spotlight SAR self-focusing imaging method based on two-dimensional self-focusing |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107942659A (en) * | 2017-11-10 | 2018-04-20 | 宁夏杰唯智能科技有限公司 | Transmitting device control method and equipment |
CN107942659B (en) * | 2017-11-10 | 2024-05-14 | 宁夏杰唯智能科技有限公司 | Transmission device control method and equipment |
CN108535730A (en) * | 2018-04-16 | 2018-09-14 | 青海大学 | A kind of Doppler weather radar solution velocity ambiguity method and system |
CN108931771A (en) * | 2018-06-06 | 2018-12-04 | 电子科技大学 | A kind of method for tracking target based on synthetic aperture radar image-forming technology |
CN109709536A (en) * | 2019-01-24 | 2019-05-03 | 电子科技大学 | A kind of SAR moving target detection method based on convolutional neural networks |
CN111161227A (en) * | 2019-12-20 | 2020-05-15 | 成都数之联科技有限公司 | Target positioning method and system based on deep neural network |
CN111199162A (en) * | 2020-01-11 | 2020-05-26 | 华南理工大学 | RFID reader fault self-adaptive positioning method |
CN111199162B (en) * | 2020-01-11 | 2021-10-26 | 华南理工大学 | RFID reader fault self-adaptive positioning method |
CN112180338A (en) * | 2020-06-10 | 2021-01-05 | 四川九洲电器集团有限责任公司 | Holographic digital array radar target quantity estimation method and system |
CN112180338B (en) * | 2020-06-10 | 2022-03-01 | 四川九洲电器集团有限责任公司 | Holographic digital array radar target quantity estimation method and system |
Also Published As
Publication number | Publication date |
---|---|
CN106597425B (en) | 2019-02-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106597425A (en) | Radar object positioning method based on machine learning | |
CN104076351B (en) | Phase-coherent accumulation detection method for high-speed high maneuvering target | |
CN103744068B (en) | The moving-target detection formation method of dual pathways Continuous Wave with frequency modulation SAR system | |
CN103869311B (en) | Real beam scanning radar super-resolution imaging method | |
CN104237871B (en) | Delay inequality estimation method based on phase compensation | |
CN107167783A (en) | A kind of sparse reconstructing method of conformal array clutter covariance matrix | |
US11581967B2 (en) | Wireless channel scenario identification method and system | |
CN107132534A (en) | A kind of optimization method of High-Speed RADAR target frequency domain detection | |
CN104597434B (en) | Improve the multiframe coherent TBD methods of envelope shift compensation and Fourier Transform of Fractional Order | |
CN111551922B (en) | Three-dimensional space double/multi-base radar high-speed target detection method | |
CN102749621B (en) | Bistatic synthetic aperture radar (BSAR) frequency domain imaging method | |
CN104849708B (en) | High speed machine moving target parameter estimation method based on the conversion of frequency domain polynomial-phase | |
CN107843892A (en) | A kind of high-speed target Doppler velocity measurement method based on least square method | |
CN106842166A (en) | A kind of solution velocity ambiguity method suitable for LFMCW radar system | |
CN108459308A (en) | A kind of analogue echoes method and device based on time-varying RCS data | |
CN109521410A (en) | High-speed maneuver target phase-coherent accumulation detection method based on time reversal transformation | |
CN101526614A (en) | SAR echo rapid simulation method based on sub-aperture and equivalent scatterer | |
CN105738879A (en) | Radar clutter time space adaptive pre-filtering method based on sparse recovery | |
CN103135100B (en) | Moving-target parameter estimation method of common-rail bistatic synthetic aperture radar (SAR) | |
CN106383340A (en) | Speed false target identifying method of random pulse initial phase radar | |
CN106249212A (en) | The polarization discrimination method of active decoy under main lobe compacting jamming pattern | |
CN103529441A (en) | Method and system for detecting and distinguishing passive synthetic aperture target signal | |
CN104730517A (en) | Bistatic MIMO radar multi-target tracking method | |
CN106646395A (en) | Radar echo deduction method for flight target | |
CN102608587B (en) | Air mobile target detection method based on nonlinear least square |
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 |