CN109444831A - A kind of radar chaff decision-making technique based on transfer learning - Google Patents
A kind of radar chaff decision-making technique based on transfer learning Download PDFInfo
- Publication number
- CN109444831A CN109444831A CN201811034422.5A CN201811034422A CN109444831A CN 109444831 A CN109444831 A CN 109444831A CN 201811034422 A CN201811034422 A CN 201811034422A CN 109444831 A CN109444831 A CN 109444831A
- Authority
- CN
- China
- Prior art keywords
- radar
- priori knowledge
- unknown threat
- parameter
- knowledge library
- 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
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/38—Jamming means, e.g. producing false echoes
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
A kind of radar chaff decision-making technique based on transfer learning provided by the invention, by the low-dimensional concealed space for constructing radar parameter priori knowledge library and unknown threat data collection, training sample is extracted from concealed space, it is trained by support vector machines, it realizes to the interfering well cluster of unknown threat data collection, effectively improves intelligent countermeasure system interfering well cluster accuracy.
Description
Technical field
The invention belongs to electronic interferences technical field more particularly to a kind of radar chaff decision-making parties based on transfer learning
Method.
Background technique
Electronic interferences refer to make enemy's electronic equipment and system lose or reduce the electric wave that efficiency is taken and upsetting measure, are
The important link of Modern Information war has run through the overall process of entire war, and the purpose is to weaken or destroy enemy to use respectively
Kind of electronic equipment and system carry out the ability of battle reconnaissance, operational commanding, liaison and weapons control and guidance, for it is hidden oneself
The survival ability of Fang Qitu and raising one's own side's aircraft, naval vessels creates favorable conditions.Core ring of the interfering well cluster as electronic interferences
Section, refers on the basis of radar priori knowledge library, is suitable for radar different operating by comparison matching or respective algorithms selection
The process of the jamming signal type of mode is basis and the key factor grasped electromagnetism processed and weigh, win battlefield active.
With Radar emitter number sharply increase and the use of Different Modulations, Battle Field Electromagnetic become day
Beneficial complicated, so that detecing the target emanation source signal that receives, there are following outstanding features: (1) due to multifunction radar signal
Polyphyly flexible and changeable, that information obtains, and signal is being propagated, is being detectd in receipts and treatment process by factors such as multipath effect, noises
Influence, cause radar priori knowledge library parameter attribute to there is incomplete phenomenon;(2) wartime radar running parameter from it is usually different,
Radar priori knowledge library sample is incomplete;(3) new system radar is constantly equipped, and is constructed without corresponding radar priori knowledge library.
In conventional radar antagonistic process, the radar parameter that countermeasure system obtains is matched with priori knowledge library, if it does, then system
Select corresponding jamming signal type;If it does not match, system interference decision accuracy is low.The more function flexible and changeable to running parameter
Energy radar, conditional electronic countercheck face the low problem of above-mentioned interfering well cluster accuracy.
In the age in last century 70-80, artificial intelligence technology, which is suggested, to be applied in electronic warfare, thereafter few open source literatures
Report.Until 2010, the U.S. issued the projects such as ARC, BLADE, cognitive interference machine, CommEx bulletin in succession, and machine learning exists
Application in electronic warfare rapidly develops.For the select permeability of jamming signal type, Tang Wenlong etc. has studied with different criterion and is based on
The jamming signal type selection method of game theory, this method are based on sufficient priori knowledge, need to establish the anti-interference countermeasure square of radar chaff-
Gust, the game matrix of both sides is mutually unknown in real process;The introducings such as Li Yunjie recognize principle, propose to calculate based on Q- study
The cognition radar electronic warfare Process Design of method, according to radar state, dynamic adjusts jamming exposure area so that interference more targetedly,
Flexibility and intelligence, but to establish under known radar status condition;Chen Kai proposes that Jamming Effect Evaluation corrects knowledge of interference
The interfering well cluster system in library, the system construct parameter library and knowledge base two parts, and structure is more complex.The studies above equips thunder to new
It reaches, priori knowledge library unknown situation can not be applied.
Summary of the invention
To solve the above problems, the present invention provides a kind of radar chaff decision-making technique based on transfer learning, can be improved
Intelligent countermeasure system interfering well cluster accuracy.
A kind of radar chaff decision-making technique based on transfer learning, comprising the following steps:
S1: using migration componential analysis, training sample is extracted from concealed space, wherein the concealed space is thunder
The same data space being mapped to up to parameter priori knowledge library and unknown threat data collection by converting φ;
S2: using the method for support vector machines, the training sample is trained, obtains interfering well cluster model;
S3: new unknown threat data is inputted into the interfering well cluster model, obtains the interference of new unknown threat data
The result of decision.
Further, described using migration componential analysis, extraction training sample specifically includes following from concealed space
Step:
Assuming that radar parameter priori knowledge library XSEdge distribution be P (XS), XSIt is after φ is convertedUnknown threat
Data set XTEdge distribution be Q (XT), XTIt is after φ is converted
Obtain edge distribution P (XS) and edge distribution Q (XT) between maximum mean difference MMD:
Wherein, tr is to seek mark,For from radar parameter priori knowledge library and unknown prestige
The data space map of data set composition is coerced to the matrix of m dimension concealed space, T is transposition, and K is transformation φ radar parameter priori
Knowledge base and unknown threat data collection are mapped into the nuclear matrix of concealed space, L is defined as:(other), xpWith xqRespectively radar parameter
Two samples arbitrarily chosen in the set that priori knowledge library and unknown threat data collection are constituted, n1For radar parameter priori knowledge
Library XSSample number, n2For unknown threat data collection XTSample number;
Function Ω to be optimized is obtained according to maximum mean difference MMD1:
Wherein, max is to be maximized, and μ > 0 is preset tradeoff parameter, Im∈Rm×mFor unit matrix, H=In1+n2-(1/
(n1+n2))11TCentered on change matrix,For complete 1 column vector,For unit matrix, R is
Real number matrix;
Obtain function Ω to be optimized1In (KLK+ μ Im)-1The maximum m characteristic value of KHK, the then corresponding feature of m characteristic value
Matrix is W1, to obtain the training sample of concealed space
Further, described that function Ω to be optimized is obtained according to maximum mean difference MMD1Specifically:
Neighborhood matching objective function Γ is constructed according to maximum mean difference MMD1:
s.t.WTKHKW=Im
Using method of Lagrange multipliers by constraint condition WTKHKW=ImNeighborhood matching objective function Γ is added1In, by neighborhood
Match objective function Γ1It is transformed to function Ω to be optimized1。
Further, described using migration componential analysis, extraction training sample specifically includes following from concealed space
Step:
Assuming that radar parameter priori knowledge library XSEdge distribution be P (XS), XSIt is after φ is convertedUnknown threat
Data set XTEdge distribution be Q (XT), XTIt is after φ is converted
Obtain edge distribution P (XS) and edge distribution Q (XT) between maximum mean difference MMD:
Wherein, tr is to seek mark,For from radar parameter priori knowledge library and unknown prestige
Side of body data set is mapped to the matrix of m dimension concealed space, and T is transposition, and K is transformation φ radar parameter priori knowledge library and unknown prestige
Side of body data set is mapped into the nuclear matrix of concealed space, L is defined as:(other), xpWith xqRespectively radar parameter
Two samples arbitrarily chosen in the sample set that priori knowledge library and unknown threat data collection are constituted, n1For radar parameter priori
Knowledge base XSSample number, n2For unknown threat data collection XTSample number;
Function Ω to be optimized is obtained according to maximum mean difference MMD2:
Wherein, max is to be maximized, and μ > 0 is default tradeoff parameter, Im∈Rm×mFor unit matrix, H=In1+n2-(1/(n1
+n2))11TIt is centralization matrix,It is complete 1 column vector,It is unit matrix, R is real
Matrix number, λ are predetermined coefficient, S=D-M, M=[mab],For table
It levies in the sample set that radar parameter priori knowledge library is constituted with unknown threat data collection, the two sample x arbitrarily chosenaAnd xb
The affinity parameter of affinity, dabFor sample xaAnd xbDistance, σ is auto-selecting parameter,For radar parameter priori knowledge library with
Unknown threat data collection, which is mapped to after concealed space, utilizes principal component analytical method treated nuclear matrix;
Obtain function Ω to be optimized2InMaximum m characteristic value,
Then the corresponding eigenmatrix of m characteristic value is W2, to obtain the training sample of concealed space
Further, described that function Ω to be optimized is obtained according to maximum mean difference MMD2Specifically:
It enablesWherein, γ ∈ [0,1] is preset tradeoff coefficient, Kl(u, v)=Kyy
(yu,yv)(u,v≤n1), KyyFor the nuclear matrix of unknown threat data collection, yu,yvIt is in radar parameter priori knowledge library any two
The jamming signal type label of a sample;
It willHilbert-Schmidt discrimination standard HSIC is substituted into, objective function Γ is obtained2:
By objective function Γ2Be converted to objective function Γ3:
According to objective function Γ3Construct function Ω to be optimized2。
The utility model has the advantages that
1, a kind of radar chaff decision-making technique based on transfer learning provided by the invention, by constructing radar parameter priori
The low-dimensional concealed space of knowledge base and unknown threat data collection extracts training sample from concealed space, is trained by support vector machines,
It realizes to the interfering well cluster of unknown threat data collection, effectively improves intelligent countermeasure system interfering well cluster accuracy.
2, a kind of radar chaff decision-making technique based on transfer learning provided by the invention, using unsupervised migration ingredient point
Analysis and semi-supervised migration constituent analysis, using Largest Mean difference MMD as weighing criteria, solve radar parameter priori knowledge library with not
The edge distribution distance for knowing threat data collection, the mark for solving nuclear matrix is converted by core embedded mode, and construction radar parameter is first
The low-dimensional concealed space for testing knowledge base Yu unknown threat data collection extracts training sample from concealed space, is instructed by support vector machines
Practice, realizes to the interfering well cluster of unknown threat data collection, be not only able to improve intelligent countermeasure system interfering well cluster accuracy, to thunder
Up to the research also important in inhibiting of interfering well cluster forward position algorithm.
Detailed description of the invention
Fig. 1 is a kind of flow chart of the radar chaff decision-making technique based on transfer learning provided by the invention;
Fig. 2 is provided by the invention interfering well cluster that tetra- kinds of methods of U-TCA, SSTCA, SVM and TM obtain to be respectively adopted just
True rate is with biased error variation tendency schematic diagram;
Fig. 3 is provided by the invention interfering well cluster that tetra- kinds of methods of U-TCA, SSTCA, SVM and TM obtain to be respectively adopted just
True rate box diagram;
Fig. 4 is provided by the invention interfering well cluster that tetra- kinds of methods of U-TCA, SSTCA, SVM and TM obtain to be respectively adopted just
True rate violin figure.
Specific embodiment
In order to make those skilled in the art more fully understand application scheme, below in conjunction in the embodiment of the present application
Attached drawing, the technical scheme in the embodiment of the application is clearly and completely described.
Referring to Fig. 1, which is a kind of radar chaff decision-making technique based on transfer learning provided by the embodiments of the present application
Flow chart.A kind of radar chaff decision-making technique based on transfer learning, comprising the following steps:
S1: using migration componential analysis, training sample is extracted from concealed space, wherein the concealed space is thunder
The same data space being mapped to up to parameter priori knowledge library and unknown threat data collection by converting φ.
It should be noted that migration constituent analysis TCA (Transfer Component Analysis-Support) is divided into
It unsupervised migration constituent analysis (unsupervised transfer component analysis, U-TCA) and semi-supervised moves
It moves constituent analysis (semi-supervised transfer component analysis, SSTCA).
It is described below using unsupervised migration constituent analysis U-TCA, the specific side of training sample is extracted from concealed space
Method.
Using migration componential analysis described in step S1, training sample is extracted from concealed space and specifically includes following step
It is rapid:
Assuming that radar parameter priori knowledge library XSEdge distribution be P (XS), XSIt is after φ is convertedUnknown threat
Data set XTEdge distribution be Q (XT), XTIt is after φ is converted
Obtain edge distribution P (XS) and edge distribution Q (XT) between maximum mean difference MMD:
Wherein, tr is to seek mark,For from radar parameter priori knowledge library and unknown prestige
The data space map of data set composition is coerced to the matrix of m dimension concealed space, T is transposition, and K is transformation φ radar parameter priori
Knowledge base and unknown threat data collection are mapped into the nuclear matrix of concealed space, L is defined as:(other), xpWith xqRespectively radar parameter
Two samples arbitrarily chosen in the set that priori knowledge library and unknown threat data collection are constituted, n1For radar parameter priori knowledge
Library XSSample number, n2For unknown threat data collection XTSample number;
Function Ω to be optimized is obtained according to maximum mean difference MMD1:
Wherein, max is to be maximized, and μ > 0 is default tradeoff parameter, Im∈Rm×mFor unit matrix, H=In1+n2-(1/(n1
+n2))11TCentered on change matrix,For complete 1 column vector,For unit matrix, R is real
Matrix number;
Obtain function Ω to be optimized1In (KLK+ μ Im)-1The maximum m characteristic value of KHK, the then corresponding feature of m characteristic value
Matrix is W1, to obtain the training sample of concealed space
It should be noted that K=[KS,SKS,T;KT,SKT,T] ∈ R is mapping phi radar parameter priori knowledge library and unknown
Threat data collection is embedded in the nuclear matrix of concealed space, KS,S、KT,T、KS,T、KT,SRespectively radar parameter priori knowledge library, unknown prestige
Coerce the Gram matrix of the crossing domain of data set and radar parameter priori knowledge library and unknown threat data collection.
It should be noted that radar parameter priori knowledge library XSEdge distribution be P (XS) and unknown threat data collection XT's
Edge distribution is Q (XT) and unequal, i.e. P (XS)≠Q(XT), but there is transformation φ, make P (φ (XS))≈Q(φ(XT)), P (YS
|φ(XS))≈P(YT|φ(XT)), wherein φ (XS)、φ(XT) respectively refer to radar parameter priori knowledge library and unknown threat number
φ transformation is carried out according to collection.In order to enable P (φ (XS))≈Q(φ(XT)), P (φ (XS)) and P (φ (XT)) distance should be minimum
Change, then have:
Wherein,Respectively refer to every concentrated to radar parameter priori knowledge library and unknown threat data
Data or each sample carry out φ transformation.
Then by core embedding grammar, formula (3) is transformed to ask mark tr (KL), just obtains obtaining edge distribution P (XS)
With edge distribution Q (XT) between maximum mean difference MMD, i.e. formula (1).
Further, described that function Ω to be optimized is obtained according to maximum mean difference MMD1Specifically:
Neighborhood matching objective function Γ is constructed according to maximum mean difference MMD1:
Using method of Lagrange multipliers by constraint condition WTKHKW=ImOptimization method is addedBy neighborhood matching objective function Γ1It is transformed to function Ω to be optimized1。
It is described below using semi-supervised migration constituent analysis SSTCA, the specific side of training sample is extracted from concealed space
Method.Described in step S1 using migration componential analysis, from concealed space extract training sample specifically includes the following steps:
Assuming that radar parameter priori knowledge library XSEdge distribution be P (XS), XSIt is after φ is convertedUnknown threat
Data set XTEdge distribution be Q (XT), XTIt is after φ is converted
Obtain edge distribution P (XS) and edge distribution Q (XT) between maximum mean difference MMD:
Wherein, tr is to seek mark,For from radar parameter priori knowledge library and unknown prestige
Side of body data set is mapped to the matrix of m dimension concealed space, and T is transposition, and K is transformation φ radar parameter priori knowledge library and unknown prestige
Side of body data set is mapped into the nuclear matrix of concealed space, L is defined as:(other), xpWith xqRespectively radar parameter
Two samples arbitrarily chosen in the set that priori knowledge library and unknown threat data collection are constituted, n1For radar parameter priori knowledge
Library XSSample number, n2For unknown threat data collection XTSample number;
Function Ω to be optimized is obtained according to maximum mean difference MMD2:
Wherein, max is to be maximized, and μ > 0 is default tradeoff parameter, Im∈Rm×mFor unit matrix, H=In1+n2-(1/(n1
+n2))11TIt is centralization matrix,It is complete 1 column vector,It is unit matrix, R is real
Matrix number, λ are predetermined coefficient, S=D-M, M=[mab],For table
It levies in the sample set that radar parameter priori knowledge library is constituted with unknown threat data collection, the two sample x arbitrarily chosenaAnd xb
The affinity parameter of affinity, dabFor sample xaAnd xbDistance, σ is auto-selecting parameter,For radar parameter priori knowledge library with
Unknown threat data collection, which is mapped to after concealed space, utilizes principal component analytical method treated nuclear matrix;
Obtain function Ω to be optimized2InMaximum m characteristic value,
Then the corresponding eigenmatrix of m characteristic value is W2, to obtain the training sample of concealed space
Further, described that function Ω to be optimized is obtained according to maximum mean difference MMD2Specifically:
It enablesWherein, γ ∈ [0,1] is preset tradeoff coefficient, Kl(u, v)=Kyy(yu,
yv)(u,v≤n1), KyyFor the nuclear matrix of unknown threat data collection, yu,yvFor any two sample in radar parameter priori knowledge library
This jamming signal type label;
It willHilbert-Schmidt discrimination standard HSIC is substituted into, objective function Γ is obtained2:
By objective function Γ2Be converted to objective function Γ3:
According to objective function Γ3Construct function Ω to be optimized2。
It should be noted that if xaAnd xbIt is adjacent in the input space, then distance after transforming to concealed space also very little,
Data transform to m dimension data space RmAfter be transformed to WTK, [WTK]fIt is any one sample xfEmbedded coordinate, then by target letter
Number Γ2Be converted to objective function Γ3Before, formula (6) can be converted into formula (8) first, obtain formula further according to formula (8)
(7)。
S2: using the method for support vector machines, the training sample is trained, obtains interfering well cluster model.
(1) it is trained by taking the training sample that U-TCA is obtained as an example below.
To concealed space sample after migration linearly can decision data, training sampleThere is n1+n2A data then take trained sample
ThisPreceding n1A data (x1,y1),(x2,y2),...,(xh,yh),...(xn1,yn1), wherein xhIndicate h-th of sample, yhIt is
Corresponding jamming signal type output.It is for the hyperplane equation of decision then: ωTX-t=0, wherein ω is adjustable weight vector,
X is input vector, and t is biasing.The problem of for giving ω and t, solving optimal hyperlane, translates into following optimization problem:
Interval maximum is set to be equivalent to minimize | | ω | |, optimization problem is converted to constrained double optimization problem:
To linearly can not decision data, for each sample introduce slack variable ξr, form soft margin optimization problem:
Wherein c > 0 is known as punishment parameter: c is larger to mean that the punishment dynamics to interval error will be increased, smaller, indicates
There is biggish tolerance to interval error.
For non-linear decision problem, SVM is mapped to feature space for concealed space sample is inputted by nonlinear transformation,
The linear problem for converting feature space by defining kernel function for input space nonlinear problem.Common kernel function is linear
Kernel function (linear), Radial basis kernel function (RBF), Polynomial kernel function (polynomial), perceptron kernel function
(sigmoid) etc..
(2) parameter optimization
SVM training process is mainly influenced by two parameters, i.e. penalty parameter c and kernel functional parameter g.For SVM parameter
Optimization, does not generally acknowledge unified best method in the world, and common method is K-CV cross validation at present.Enable c, g [2-10,
210] value in range, for given parameter c, g, training set is obtained as initial data progress cross validation in this group of c, g
Under accuracy is selected to the jamming signal type of training set, highest that group of c, g will be obtained as optimized parameter.
Parameter optimization result horizontal axis and the longitudinal axis indicate to take with 2 that as the value after the logarithm at bottom, contour expression takes corresponding c
With the interfering well cluster accuracy of cross validation corresponding after g.C value is excessive to will lead to overfitting, makes the extensive energy of interfering well cluster
Power reduces, and the feelings very high and very low to unknown threat data interfering well cluster accuracy to training sample interfering well cluster accuracy occurs
Condition.When different parameter combinations simultaneously reach optimum jamming decision accuracy when, take the smallest one group of parameter c, g as optimal ginseng
Number.
S3: new unknown threat data is inputted into the interfering well cluster model, obtains the interference of new unknown threat data
The result of decision.
The present embodiment proposition method is denoted as U-TCA and SSTCA, for the validity for verifying the proposed method of the present embodiment, is divided
Interfering well cluster accuracy under different biased errors horizontal (Error Deviation Level, EDL) is analysed, TM, SVM, UTCA are utilized
And 4 kinds of methods such as SSTCA compare experiment.
TM method distance threshold is set as 0.03.The parameters such as repetition, carrier frequency are in lesser range under certain fixation operating mode
Interior variation carries out a test experiments at interval of 2%, surveys every time to the biased error between test sample addition 0%-20%
Examination carries out 100 Monte Carlo experiments, calculates under different biased error level conditions, is carrying out interfering well cluster just using distinct methods
The mean value of true rate, interfering well cluster accuracy are defined as selecting the ratio between sample number and test sample number of corresponding jamming signal type label.
U-TCA, SSTCA, SVM and TM interfering well cluster accuracy variation tendency are as shown in Figure 2.Air-air scene fire control radar works in X more
Wave band, bandwidth are generally less than 10e3MHz, therefore to certain operating mode using agile parameter, the Parameters variations such as PRI, CF are less than
10%, when biased error level is 10%, interfering well cluster accuracy box diagram and violin figure are as shown in Figure 3, Figure 4.
It can be seen that traditional TM method interfering well cluster accuracy is lower than 14.81% when biased error level is less than 20%,
The SVM interfering well cluster accuracy and U-TCA, SSTCA interfering well cluster accuracy not migrated are 87% or more, wherein being based on
The interfering well cluster accuracy average value of TCA is greater than 93%, and higher than the SVM method not migrated, SVM, SSTCA and U-TCA are with deviation
It is gentle that error increases the variation of interfering well cluster accuracy, demonstrate this 3 kinds of methods all have stronger generalization ability and robustness this
One conclusion.U-TCA and SSTCA algorithm itself has no point of superiority and inferiority.When cross-cutting differentiation direction be it is similar, believed using label
Breath, can make different classes of sample that can more divide in concealed space, and semi-supervised TCA algorithm is better than unsupervised TCA algorithm at this time.
Certainly, the invention may also have other embodiments, without deviating from the spirit and substance of the present invention, ripe
Various corresponding changes and modifications can be made according to the present invention certainly by knowing those skilled in the art, but these it is corresponding change and
Deformation all should fall within the scope of protection of the appended claims of the present invention.
Claims (5)
1. a kind of radar chaff decision-making technique based on transfer learning, which comprises the following steps:
S1: using migration componential analysis, training sample is extracted from concealed space, wherein the concealed space is radar ginseng
The same data space that number priori knowledge library and unknown threat data collection are mapped to by converting φ;
S2: using the method for support vector machines, the training sample is trained, obtains interfering well cluster model;
S3: new unknown threat data is inputted into the interfering well cluster model, obtains the interfering well cluster of new unknown threat data
As a result.
2. a kind of radar chaff decision-making technique based on transfer learning as described in claim 1, which is characterized in that the use
Migrate componential analysis, from concealed space extract training sample specifically includes the following steps:
Assuming that radar parameter priori knowledge library XSEdge distribution be P (XS), XSIt is after φ is convertedUnknown threat data
Collect XTEdge distribution be Q (XT), XTIt is after φ is converted
Obtain edge distribution P (XS) and edge distribution Q (XT) between maximum mean difference MMD:
Wherein, tr is to seek mark,For from radar parameter priori knowledge library and unknown threat number
According to the data space map of collection composition to the matrix of m dimension concealed space, T is transposition, and K is transformation φ radar parameter priori knowledge
Library and unknown threat data collection are mapped into the nuclear matrix of concealed space, L is defined as:xpWith xqRespectively radar parameter
Two samples arbitrarily chosen in the set that priori knowledge library and unknown threat data collection are constituted, n1For radar parameter priori knowledge
Library XSSample number, n2For unknown threat data collection XTSample number;
Function Ω to be optimized is obtained according to maximum mean difference MMD1:
Wherein, max is to be maximized, and μ > 0 is preset tradeoff parameter, Im∈Rm×mFor unit matrix, H=In1+n2-(1/(n1+
n2))11TCentered on change matrix,For complete 1 column vector,For unit matrix, R is real number
Matrix;
Obtain function Ω to be optimized1In (KLK+ μ Im)-1The maximum m characteristic value of KHK, the then corresponding eigenmatrix of m characteristic value
For W1, to obtain the training sample of concealed space
3. a kind of radar chaff decision-making technique based on transfer learning as claimed in claim 2, which is characterized in that the basis
Maximum mean difference MMD obtains function Ω to be optimized1Specifically:
Neighborhood matching objective function Γ is constructed according to maximum mean difference MMD1:
s.t.WTKHKW=Im
Using method of Lagrange multipliers by constraint condition WTKHKW=ImNeighborhood matching objective function Γ is added1In, by neighborhood matching
Objective function Γ1It is transformed to function Ω to be optimized1。
4. a kind of radar chaff decision-making technique based on transfer learning as described in claim 1, which is characterized in that the use
Migrate componential analysis, from concealed space extract training sample specifically includes the following steps:
Assuming that radar parameter priori knowledge library XSEdge distribution be P (XS), XSIt is after φ is convertedUnknown threat data
Collect XTEdge distribution be Q (XT), XTIt is after φ is converted
Obtain edge distribution P (XS) and edge distribution Q (XT) between maximum mean difference MMD:
Wherein, tr is to seek mark,For from radar parameter priori knowledge library and unknown threat number
The matrix of m dimension concealed space is mapped to according to collection, T is transposition, and K is transformation φ radar parameter priori knowledge library and unknown threat number
The nuclear matrix of concealed space, L are mapped into according to collection is defined as:xpWith xqRespectively radar is joined
Two samples arbitrarily chosen in the sample set that number priori knowledge library and unknown threat data collection are constituted, n1For radar parameter elder generation
Test knowledge base XSSample number, n2For unknown threat data collection XTSample number;
Function Ω to be optimized is obtained according to maximum mean difference MMD2:
Wherein, max is to be maximized, and μ > 0 is default tradeoff parameter, Im∈Rm×mFor unit matrix, H=In1+n2-(1/(n1+n2))
11TIt is centralization matrix,It is complete 1 column vector,It is unit matrix, R is real number square
Battle array, λ is predetermined coefficient, S=D-M, M=[mab],To characterize thunder
Up to the two sample x in the sample set of parameter priori knowledge library and unknown threat data collection composition, arbitrarily chosenaAnd xbIt is affine
The affinity parameter of degree, dabFor sample xaAnd xbDistance, σ is auto-selecting parameter,For radar parameter priori knowledge library with it is unknown
Threat data collection, which is mapped to after concealed space, utilizes principal component analytical method treated nuclear matrix;
Obtain function Ω to be optimized2InMaximum m characteristic value, then m
The corresponding eigenmatrix of characteristic value is W2, to obtain the training sample of concealed space
5. a kind of radar chaff decision-making technique based on transfer learning as claimed in claim 4, which is characterized in that the basis
Maximum mean difference MMD obtains function Ω to be optimized2Specifically:
It enablesWherein, γ ∈ [0,1] is preset tradeoff coefficient, Kl(u, v)=Kyy(yu,
yv)(u,v≤n1), KyyFor the nuclear matrix of unknown threat data collection, yu,yvFor any two sample in radar parameter priori knowledge library
This jamming signal type label;
It willHilbert-Schmidt discrimination standard HSIC is substituted into, objective function Γ is obtained2:
By objective function Γ2Be converted to objective function Γ3:
According to objective function Γ3Construct function Ω to be optimized2。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811034422.5A CN109444831B (en) | 2018-09-06 | 2018-09-06 | Radar interference decision method based on transfer learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811034422.5A CN109444831B (en) | 2018-09-06 | 2018-09-06 | Radar interference decision method based on transfer learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109444831A true CN109444831A (en) | 2019-03-08 |
CN109444831B CN109444831B (en) | 2023-03-10 |
Family
ID=65530142
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811034422.5A Active CN109444831B (en) | 2018-09-06 | 2018-09-06 | Radar interference decision method based on transfer learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109444831B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111199127A (en) * | 2020-01-13 | 2020-05-26 | 西安电子科技大学 | Radar interference decision method based on deep reinforcement learning |
CN111597873A (en) * | 2020-03-27 | 2020-08-28 | 中国人民解放军海军工程大学 | Support vector data description-based ultrashort wave threat signal sensing method |
CN112949387A (en) * | 2021-01-27 | 2021-06-11 | 西安电子科技大学 | Intelligent anti-interference target detection method based on transfer learning |
CN113625740A (en) * | 2021-08-27 | 2021-11-09 | 北京航空航天大学 | Unmanned aerial vehicle air combat game method based on transfer learning pigeon swarm optimization |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104635215A (en) * | 2013-11-07 | 2015-05-20 | 北京航天长征飞行器研究所 | Jamming decision device based on SOPC (System-on-a-Programmable-Chip) |
CN106406162A (en) * | 2016-08-12 | 2017-02-15 | 广东技术师范学院 | Alternating current servo control system based on transfer neural network |
CN107301381A (en) * | 2017-06-01 | 2017-10-27 | 西安电子科技大学昆山创新研究院 | Recognition Method of Radar Emitters based on deep learning and multi-task learning strategy |
CN107390180A (en) * | 2017-07-18 | 2017-11-24 | 中国人民解放军国防科学技术大学 | The multifunction radar signal estimation method of model is represented based on predicted state |
CN107390189A (en) * | 2017-07-18 | 2017-11-24 | 中国人民解放军国防科学技术大学 | Multifunction radar behavior identification and method for quick predicting under the conditions of low prior information |
US9993166B1 (en) * | 2013-06-21 | 2018-06-12 | Fitbit, Inc. | Monitoring device using radar and measuring motion with a non-contact device |
-
2018
- 2018-09-06 CN CN201811034422.5A patent/CN109444831B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9993166B1 (en) * | 2013-06-21 | 2018-06-12 | Fitbit, Inc. | Monitoring device using radar and measuring motion with a non-contact device |
CN104635215A (en) * | 2013-11-07 | 2015-05-20 | 北京航天长征飞行器研究所 | Jamming decision device based on SOPC (System-on-a-Programmable-Chip) |
CN106406162A (en) * | 2016-08-12 | 2017-02-15 | 广东技术师范学院 | Alternating current servo control system based on transfer neural network |
CN107301381A (en) * | 2017-06-01 | 2017-10-27 | 西安电子科技大学昆山创新研究院 | Recognition Method of Radar Emitters based on deep learning and multi-task learning strategy |
CN107390180A (en) * | 2017-07-18 | 2017-11-24 | 中国人民解放军国防科学技术大学 | The multifunction radar signal estimation method of model is represented based on predicted state |
CN107390189A (en) * | 2017-07-18 | 2017-11-24 | 中国人民解放军国防科学技术大学 | Multifunction radar behavior identification and method for quick predicting under the conditions of low prior information |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111199127A (en) * | 2020-01-13 | 2020-05-26 | 西安电子科技大学 | Radar interference decision method based on deep reinforcement learning |
CN111199127B (en) * | 2020-01-13 | 2022-09-06 | 西安电子科技大学 | Radar interference decision method based on deep reinforcement learning |
CN111597873A (en) * | 2020-03-27 | 2020-08-28 | 中国人民解放军海军工程大学 | Support vector data description-based ultrashort wave threat signal sensing method |
CN112949387A (en) * | 2021-01-27 | 2021-06-11 | 西安电子科技大学 | Intelligent anti-interference target detection method based on transfer learning |
CN112949387B (en) * | 2021-01-27 | 2024-02-09 | 西安电子科技大学 | Intelligent anti-interference target detection method based on transfer learning |
CN113625740A (en) * | 2021-08-27 | 2021-11-09 | 北京航空航天大学 | Unmanned aerial vehicle air combat game method based on transfer learning pigeon swarm optimization |
CN113625740B (en) * | 2021-08-27 | 2024-02-27 | 北京航空航天大学 | Unmanned aerial vehicle air combat game method based on transfer learning pigeon swarm optimization |
Also Published As
Publication number | Publication date |
---|---|
CN109444831B (en) | 2023-03-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109444831A (en) | A kind of radar chaff decision-making technique based on transfer learning | |
Lv et al. | Radar deception jamming recognition based on weighted ensemble CNN with transfer learning | |
Zhang et al. | Cross-scene hyperspectral image classification with discriminative cooperative alignment | |
Pei et al. | SAR automatic target recognition based on multiview deep learning framework | |
CN109376556B (en) | Attack method for EEG brain-computer interface based on convolutional neural network | |
CN111163472B (en) | Signal identification attack defense method based on generative countermeasure network | |
CN107992891A (en) | Based on spectrum vector analysis multi-spectral remote sensing image change detecting method | |
CN101833667A (en) | Pattern recognition classification method expressed based on grouping sparsity | |
CN107274360A (en) | A kind of high spectrum image denoising method based on Fisher dictionary learnings, low-rank representation | |
CN116106878A (en) | Big data analysis system and method | |
Yao et al. | Radar-based human activity recognition using hyperdimensional computing | |
CN116451145A (en) | Cooling device and method for stem cell transportation | |
Sun et al. | Radar emitter individual identification based on convolutional neural network learning | |
Zha et al. | LT-SEI: Long-tailed specific emitter identification based on decoupled representation learning in low-resource scenarios | |
CN115951315A (en) | Radar deception jamming identification method and system based on improved wavelet packet energy spectrum | |
CN109784318A (en) | The recognition methods of Link16 data-link signal neural network based | |
CN104809444A (en) | Half-bridge directional filter method for infrared small target detection based on tensor extension | |
Wang et al. | Clustering based on kernel density estimation: nearest local maximum searching algorithm | |
Feng et al. | FCGCN: Feature Correlation Graph Convolution Network for Few-Shot Individual Identification | |
Li et al. | Battlefield target identification based on improved grid-search svm classifier | |
CN113177520A (en) | Intelligent radiation source identification method based on ensemble learning | |
Bing et al. | Modulation recognition of communication signal based on wavelet RBF neural network | |
CN110189361A (en) | A kind of multi-channel feature and the target following preferentially updated parallel | |
Wang et al. | Image retrieval based on fireworks algorithm optimizing convolutional neural network | |
Li et al. | Radiation source individual identification using machine learning 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 |