CN110308432A - A kind of radar self-adaption waveform selection Activity recognition method neural network based - Google Patents
A kind of radar self-adaption waveform selection Activity recognition method neural network based Download PDFInfo
- Publication number
- CN110308432A CN110308432A CN201910628143.XA CN201910628143A CN110308432A CN 110308432 A CN110308432 A CN 110308432A CN 201910628143 A CN201910628143 A CN 201910628143A CN 110308432 A CN110308432 A CN 110308432A
- Authority
- CN
- China
- Prior art keywords
- radar
- neural network
- target
- waveform
- moment
- 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.)
- Pending
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
- G01S7/36—Means for anti-jamming, e.g. ECCM, i.e. electronic counter-counter measures
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention belongs to electronic countermeasure technology fields, are related to a kind of radar self-adaption waveform selection Activity recognition method neural network based.The present invention is fitted the adaptively selected system of the waveform of target radar by building neural network.For the Activity recognition problem of a certain portion's radar, can be equivalent to solve the system response problem of the radar self-adaption system.It is the input and output data by obtaining system to the basic skills that the system response of unknown system solves, thus the transmission characteristic of analysis system.The present invention solve conventional radar Activity recognition can only Discrimination Radar current state, this problem of radar behavior can not be recognized.
Description
Technical field
The invention belongs to electronic countermeasure technology fields, are related to a kind of radar self-adaption waveform selection row neural network based
For recognition methods.
Background technique
Radar system is recognized by perception environment and target, carries out analysis decision according to certain mode and rule, finally
Optimize the working condition of oneself.Wherein, cognition transmitting is the important feature for recognizing radar and being different from conventional radar.In different situations
Under, it can be obviously improved the quality of echo-signal using cognition lift-off technology, system detection performance is substantially improved.And adaptive waveform
Selection technique is exactly the pith recognized in lift-off technology.Radar is for oneself that the adaptive adjustment of transmitted waveform is exactly radar
Adapt to waveform behavior.
It is corresponding, under the continuous development of modern radar, it also proposed bigger challenge to Radar Signal Recognition.It passes
By elint system (ELINT) come the emission parameter information of trailer record difference radar in the radar-reconnaissance of system.It is fighting
In, the radar signal of intercepting and capturing is by carrying out right, identification of the realization to radar target with the record in ELINT.And it is directed to work
Mode and parameter New Type Radar complicated and changeable, if only analyzing and knowing still through the mode of this traditional template matching
The current state of these other radars, it will be difficult to obtain good effect, and recognition result will be lagged seriously, lack timeliness
Property.Therefore, it for flexible and changeable New Type Radar, needs to go to analyze its Behavior law from more high-dimensional, obtains its basic change
Change feature.
Summary of the invention
The present invention is fitted the adaptively selected system of the waveform of target radar by building neural network.To Mr. Yu
The Activity recognition problem of one radar can be equivalent to solve the system response problem of the radar self-adaption system.To unknown system
The basic skills that the system response of system solves is the input and output data by obtaining system, so that the transmitting of analysis system is special
Property.
Angle from radar electronic warfare side, the electromagnetic environments information such as clutter is identical with radar side;Target information is thunder
Up to the radiation source of confrontation side itself, or even if not being one's own side's radiation source, target can also be obtained by reconnaissance equipments such as radars
Information;Interference information is the transmitting of radar electronic warfare side thus can directly acquire;And the transmitted waveform data of radar side can also be with
Receive acquisition by scouting;Therefore, radar electronic warfare side is to have ready conditions to obtain complete input-output data.
Neural network algorithm is the method being most suitable at present for excavating rule from the data of a large amount of structurings, so this
Method excavates the mapping relations in input-output sample data, solves thunder indirectly by building neural network algorithm model
System up to Adaptable System responds, to complete to realize the adaptive waveform of radar to the fitting of radar waveform selection system
Behavior identification.
This method main contents are to construct waveform Activity recognition model neural network based.
For the Activity recognition problem of a certain portion's radar, can be equivalent to solve the system response of the radar self-adaption system
Problem.It is the input and output data by obtaining system to the basic skills that the system response of unknown system solves, thus point
The transmission characteristic of analysis system.
Angle from radar electronic warfare side, the information of acquisition are extremely incomplete.Radar obtain environment, target and
Interference etc. is after information, last decision is that Multiple factors are comprehensive by multiple links generate as a result, such as radar receiver noise
It is that radar electronic warfare side can not know than SNR.It therefore, can not be by outputting and inputting in the case where this information asymmetry
Data solve accurate system response analytical form.
Although the angle from radar electronic warfare side, complete radar self-adaption system information can not be directly acquired, this
There is no complete " loss " for a little information, but are hidden in input-output data.At the same time, it stands at the angle of radar electronic warfare side
Degree, the electromagnetic environments information such as clutter is identical with radar side;Target information is the radiation source of radar electronic warfare side itself, or i.e.
Make not to be one's own side's radiation source, target information can also be obtained by reconnaissance equipments such as radars;Interference information is radar electronic warfare side's hair
That penetrates can thus directly acquire;And the transmitted waveform data of radar side can also receive acquisition by scouting;Therefore, radar pair
Anti- side is to have ready conditions to obtain complete input-output data.So defeated as long as this can be excavated from sample data
Enter-mapping principle is exported, it is equivalent to solve the system response problem of entire Adaptable System.
Neural network algorithm is the method being most suitable at present for excavating rule from the data of a large amount of structurings, so this
Research excavates the mapping relations in input-output sample data, solves thunder indirectly by building neural network algorithm model
System up to Adaptable System responds, to complete to realize the adaptive waveform of radar to the fitting of radar waveform selection system
Behavior identification.
Overall algorithm of the present invention includes two parts, and first part part is radar self-adaption waveform selection system, extraneous
It influences to obtain transmitted waveform parameter by the waveform selection algorithm based on Kalman filtering by the change input system of SNR.
Second part is that system is selected by neural network learning radar waveform, and SNR input is by interference and dbjective state parametrization input
Instead of, it can be understood as, it influences to form one as a subsystem and waveform selection system with the relational model of SNR by extraneous
Total system, what neural network actually learnt fitting is this total system.The data of experimental stage can be by being previously run thunder
It obtains up to analogue system or is obtained by acquisition battlefield data, and in actually fighting, the acquisition of data can also fight
In obtain immediately, with radar waveform selection system input and output, repetitive exercise algorithm, excavate input-output mapping close
System, the neural network algorithm model for completing training and radar system have similar input-output characteristic.
As shown in Figure 1, the technical scheme is that
1, a kind of radar self-adaption waveform selection Activity recognition method neural network based, which is characterized in that including with
Lower step:
S1, the target state data that target radar is obtained by investigation equipment;Including target range and target velocity;
The supplemental characteristic that target radar is obtained by reconnaissance equipment, including the pulsewidth and chirp rate of target radar transmitting signal;
It is known because interference information is one's own side's transmitting.
Assuming that three kinds are interfered the decaying for ultimately causing radar receiver signal-to-noise ratio as shown in table 1 there are three types of interference signal:
Decaying of the 1 three kinds of interference signals of table to the receiver signal-to-noise ratio of radar
S2, according to step S1 obtain target state data and interference information and previous moment waveform parameter, pass through
Neural network algorithm obtains the transmitted waveform parameter of target radar subsequent time.In simulations, ignore targeted attitude and clutter ring
The influence in border, introducing interference as extraneous influences input.Neural network in emulation is inputted including 5 dimensional features: k-1 moment arteries and veins
Width, k-1 moment chirp rate, target range, target velocity and interference type label export as 2 dimensions: k moment pulsewidth, k moment
Chirp rate.
The forward-propagating that radar self-adaption waveform selection Activity recognition is neural network is carried out using neural network algorithm
Process, as shown in figure 3, feature input for 5 dimension: k-1 moment pulsewidth, k-1 moment chirp rate, target range, target range and
Interference type label, export as 2 dimensions: k moment pulsewidth, k moment chirp rate use x1,x2,x3,x4,x5It indicates, then single sample
It indicates are as follows:
In individual node, intermediate variable are as follows:
zj=WjHj-1+b0
Wherein, WjIndicate the weight parameter of jth layer hidden layer, bjIndicate biasing, HjFor hidden layer variable.Intermediate variable z
Input activation primitiveIn after, so that it may obtain jth layer hidden layer variable are as follows:
In first layer input layer,
zl=W1H0+b1
H0=xi
Because in this method, hidden layer shares 2 layers, so in the last layer, output layer output are as follows:
O=W3H2+b3
Wherein,
O=[o1 o2]
o1Indicate pulsewidth o2Indicate chirp rate.
By training neural network, the parameter W=[W of neural network is obtained1 W2 W3] after, so that it may input data xi, complete
The prediction of the adaptively selected behavior of pairs of radar waveform.
Neural network model parameter setting is as shown in table 2:
The setting of 2 neural network model parameter of table
The hidden layer number of plies | 2 |
Input layer number | 5 |
Hidden layer number of nodes | 10 |
Output layer number of nodes | 2 |
Training method | Factor of momentum steepest descent method |
Factor of momentum | 0.3 |
Learning rate | 0.5 |
S3, after constructing neural network model (model parameter of neural network is as shown in table 4-2) acquire data under line
Collection, acquisition method are the method for step S1, use the target state data obtained in data set and interference information as input, mesh
The transmitted waveform parameter of radar is marked as desired output, neural network model is trained, trained neural network is obtained
Model;
Parameter W=[the W of neural network in order to obtain1 W2 W3], enabled neural network can Accurate Prediction radar waveform from
Housing choice behavior is adapted to, the present invention uses error-duration model coaching method:
The mathematical description of neuron propagated forward is as follows:
The neuron has an input x, and weight w, deviation input is b, and target output is t, it is contemplated that is exported as y.In advance
Period error is
To eliminate the caused error compliance problem when error is summed in entire input and output mode, we are used here
Delta rule, the error used in delta rule is square error, is defined as:
According to delta rule, best initial weights (make square error minimum) can in the training process from initial weight,
Decline to obtain along negative gradient direction.Square error is carried out differential, obtained to w, b:
Weight change should be proportional to the negative value of error gradient, introduces learning rate β, and the weight change in each iteration can table
It is shown as:
Learning rate β determines the movement speed along gradient direction, with the new weight of determination.Big β value can accelerate weight
Change, small β value then slows down the change of weight.It is all, theiNew weight after secondary iteration may be expressed as:
In short, being realized as follows using the BP algorithm of delta rule: input pattern (x0,x1,x2,…xn) pass through
Connection is passed, its initial weight is arranged to arbitrary value.To the input of weighting summation, output y is generated, then y and given
Target output t compare determine this mode square error ε.Input and target output are constantly suggested, and are changed each time
Weighed value adjusting is carried out until obtaining possible least squares error using delta rule after generation or training time each time.
S4, using trained neural network model, according to the target state data and interference information obtained in real time, to mesh
The transmitted waveform parameter of mark radar is predicted.
The beneficial effects of the present invention are: this method will use nerve on the basis of radar self-adaption waveform behavior modeling
Network algorithm carries out action learning identification for the radar with adaptive waveform behavior, solves conventional radar Activity recognition
Can only Discrimination Radar current state, this problem of radar behavior can not be recognized.
Detailed description of the invention
Fig. 1 radar waveform Activity recognition neural network based;
Radar tracking waveform Activity recognition model based on machine learning in Fig. 2 emulation;
Fig. 3 is neural network model figure;
Neural network prediction and actual parameter comparative selection under Fig. 4 minimum mean square error criterion: (a) pulsewidth selects;(b) it adjusts
The selection of frequency slope;
Neural network prediction and actual parameter comparative selection under Fig. 5 comentropy criterion: (a) pulsewidth selects;(b) chirp rate
Selection;
Fig. 6 minimum confirms neural network prediction and actual parameter comparative selection under thresholding Volume Criterion: (a) pulsewidth selects;
(b) chirp rate selects.
Specific embodiment
Illustrate the validity of the present invention program below with reference to simulated example.
The waveform selection algorithm for being primarily based on Kalman filtering is emulated, and is then joined using obtained radar tracking waveform
Training sample of the Number Sequence as waveform Activity recognition algorithm neural network based.
(1) the waveform selection algorithm simulating of Kalman filtering
Under certain scene settings, to taking least mean-square error, comentropy and minimum confirmation three kinds of standards of thresholding volume
Then the Kalman filtering of the waveform selection of function is emulated, and specific principle of simulation is introduced in annex 1.And with take fixed wave
The Kalman filtering algorithm of shape compares and analyzes.
Simulating scenes setting: radar tracking object is a uniformly accelrated rectilinear motion object, initial velocity 40m/ in scene one
S, constant acceleration is 2m/s2Prolong radar radial direction away from radar motion, is 20km with radar initial distance.
Gaussian envelope chirp pulse signal is used in emulation, selectable signal parameter is pulsewidth τ and chirp rate
b.Pulsewidth range of choice is τ ∈ [10,300] μ s, 5 μ s of selection interval;Chirp rate range of choice is b ∈ [- 3000, -100] ∪
[100,3000] MHz, selection interval 50MHz.
R0Target when for signal-to-noise ratio being 1 between radar at a distance from, R is taken in emulation0=400km.Put aside that clutter causes
Signal-to-noise ratio rise and fall, and interfere caused by signal-to-noise ratio mutation.
(2) waveform Activity recognition algorithm simulating neural network based
In simulations, the input of the waveform selection system based on Kalman filtering is divided into the wave of extraneous factor and last moment
Two parts of shape parameter.The influence of extraneous factor is finally reflected the decaying in receiver signal-to-noise ratio to act on the waveform of radar
Selection.And waveform parameter part, it is assumed that radar emission signal is chirp pulse signal, at this point, the radar that can be optimized and revised
Parameter is pulsewidth and chirp rate.Since during a secondary tracking, tracking depths also have an impact to radar waveform parameter selection.
Therefore, the specific input in above-mentioned simulation model includes three parts, and first part is the actual parameter at k-1 moment, includes arteries and veins
Wide and chirp rate, second part are dbjective states, include target range and target velocity, Part III is interference type;Tool
Body output is k moment Prediction Parameters, includes pulsewidth and chirp rate.The simulation model figure is Fig. 2.
In emulation, ignore the influence of targeted attitude and clutter environment, dbjective state input is current goal distance and target
Speed, introducing interference as extraneous influences input, and practical threat radar identification focuses more on thunder caused by the interference to radar
It is most valuable to threat radar is successfully managed up to waveform parameter changing rule.Therefore the neural network in emulation includes 5 Wei Te
Sign input: k-1 moment pulsewidth, k-1 moment chirp rate, target range, target range and interference type label export as 2 dimensions:
K moment pulsewidth, k moment chirp rate.Assume that, there are three types of interference signal, three kinds of interference ultimately cause radar receiver letter in emulation
Make an uproar ratio decaying it is as shown in table 1.
Decaying of the 1 three kinds of interference signals of table to the receiver signal-to-noise ratio of radar
Interference type | SNR degradation |
1 | 1 |
2 | 0.02 |
3 | 0.1 |
Under last point of simulating scenes setting, SNR degradation caused by interfering is added and carries out based on Kalman filtering
Radar tracking waveform selection algorithm simulating, obtain simulating tracking waveform parameter sequence conduct of the practical radar under interference effect
The sample data of training neural network.Under minimum mean square error criterion, comentropy criterion and minimum confirmation thresholding Volume Criterion,
The wave that the radar tracking waveform parameter sequence and neural network that waveform selection algorithm based on Kalman filtering obtains are fitted
Shape parameter alignment is as shown in Fig. 3,4,5.
In order to make it easy to understand, being introduced as follows above-mentioned criterion:
1) Kalman filtering
Kalman filter is a kind of algorithm estimated by observing data dbjective state.Pass through state equation and sight
Survey equation description description object space.State equation describes the state transfer of target, with xkIndicate the state at target k moment, xk-1
Indicate that the state at k-1 moment, target state equation form are as follows:
xk=f (xk-1)+vk
F () is state transition function, vkFor the state-noise at k moment.
Observational equation are as follows:
zk=h (xk)+wk
Wherein zkIndicate the observation vector at k moment;H () is state-observation transfer function;wkIt makes an uproar for the measurement survey at k moment
Sound.
Assuming that system is linear discrete time system, the linear forms of state equation and observational equation are as follows:
X (k)=Fx (k-1)+Γ v (k)
Z (k)=Hx (k)+w (k)
Wherein, x (k) is state vector, and F is state-transition matrix, and Γ is state-noise input matrix, and v (k) is system shape
State noise, covariance matrix Q;Z (k) is observation vector, and H is calculation matrix, and w (k) is that measurement noise has with waveform parameter
It closes, covariance matrix is the CRLB for measuring noise covariance, is expressed as R (θ), θ=(τ, b), wherein τ is pulsewidth, and b is to adjust
Frequency slope.
The basic step of Kalman filter algorithm is as follows:
(1) dbjective state one-step prediction:
(2) it is observed by the k moment and calculates innovation process αk:
(3) one-step prediction error covariance is calculated:
Pk|k-1=FPk-1|k-1FT+ΓQΓT
(4) the covariance matrix A of innovation process is calculated:
Ak=HPk|k-1HT+R(θ)
(5) kalman gain K is calculated:
(6) state estimation:
(7) state estimation error co-variance matrix is updated:
In Kalman filter, be added parameter waveform selection course, with certain criterion selection optimum waveform parameter realize with
Track waveform adaptively adjusts.By selecting certain objective function, calculating target function value is selected so that target function value is optimal
Waveform parameter.The selection of objective function determines the standard of parameter of measurement superiority and inferiority.The form of objective function has very much, gives below
Objective function under several different criterion out
2) minimum mean square error criterion
Mean square error is to measure the most common index of accuracy, in Kalman filtering algorithm, state estimation mean square error
The accuracy of tracking is just reflected, it can be to minimize the mean square error of state estimation as parameter optimization criterion.Then objective function
Is defined as:
Wherein, θ=(τ, b), wherein τ is pulsewidth, and b is chirp rate, and x (k) is state vector, and z (k) is observation vector,
| | | | indicate L2 norm.In Kalman filtering algorithm, state estimation error co-variance matrix is defined as:
P (k | k)=E [εk|k(θ)εk|k(θ)T|zk]
WhereinIndicate k moment state estimation error, mark operation is asked in Tr [] expression.State is estimated
Count error co-variance matrix track taking:
Therefore, the mark that state estimation mean square error is equivalent to minimum state evaluated error covariance matrix is minimized, is obtained
Final goal functional form under minimum mean square error criterion:
Wherein, P (θ) indicates that state estimation error co-variance matrix passes through state estimation that is, in Kalman filtering process
Error co-variance matrix carrys out calculating target function value.
3) minimum confirmation thresholding Volume Criterion
Confirmation thresholding volume refers to the observation space cell size during tracking.In clutter and interfere intensive background ring
Under border, smaller thresholding volume just includes fewer clutter and interference, to improve tracking effect.According to derivation, minimize true
Recognize thresholding volume to be equivalent to minimize the determinant of innovation process covariance matrix, objective function indicates are as follows:
Wherein, θ=(τ, b), wherein τ is pulsewidth, and b is chirp rate, and A indicates the covariance matrix of innovation process, det
Matrix determinant operation is asked in [] expression, in Kalman filtering, by innovation process covariance matrix come calculating target function.
4) comentropy criterion
Shannon entropy is introduced to state estimation error:
G=H [εk|k(θ)]
Wherein, H [] is to seek Entropy algorithm,Indicate k moment state estimation error.
Comentropy characterizes the size of information content, and in state estimation, the information content in evaluated error is smaller, then estimated value
Accuracy it is higher.Therefore, minimizing evaluated error comentropy can be used as the objective function of parameter optimization.According to derivation, state
Evaluated error comentropy is proportional to the determinant of state estimation error co-variance matrix, therefore objective function can be expressed as follows:
Wherein, P (k | k) indicates state estimation error co-variance matrix, and matrix determinant operation is asked in det [] expression.
As can be seen that waveform Activity recognition neural network based can be very good to intend from the simulation result of Fig. 3,4,5
It closes actual waveform and selects system, by current time waveform parameter, target status information, interference type as input, export thunder
Up to the predicted value of subsequent time waveform parameter, the characteristic of guinea pig waveform selection system.Interference and dbjective state are as radar
Identification side is to know information, and radar waveform Activity recognition method as described herein will be interfered and dbjective state is as inputting, directly
The basic reason of radar self-adaption waveform behavior will be caused to associate with the selection of waveform parameter, the input and output of system are
Known to Radar Signal Recognition side.The result of neural network forecast is almost coincide with actual parameter selection trend, is only drawn with noise
The small size error jitter risen.
Claims (1)
1. a kind of radar self-adaption waveform selection Activity recognition method neural network based, which is characterized in that including following step
It is rapid:
S1, the target state data that target radar is obtained by investigation equipment, including target range and target velocity;Pass through scouting
Equipment obtains the supplemental characteristic of target radar, pulsewidth and chirp rate including target radar transmitting signal;Interference information simultaneously
It is one's own side's transmitting, therefore is known;
S2, building neural network model, neural network model is according to the step S1 target state data obtained and interference information, knot
Waveform parameter, that is, target radar the supplemental characteristic for closing previous moment obtains under target radar for the moment by neural network algorithm
The transmitted waveform parameter at quarter;The neural network is inputted including 5 dimensional features: k-1 moment pulsewidth, k-1 moment chirp rate, target
Distance, target velocity and interference type label export as 2 dimensions: k moment pulsewidth, k moment chirp rate;
S3, after constructing neural network model, acquire data set under line, acquisition method is to pass through the method for step S1 to obtain data
Collection uses training data as input, the transmitting of target radar using the data set of acquisition as training data according to step S2
Waveform parameter is trained neural network model as desired output, obtains trained neural network model;
S4, using trained neural network model, according to the target state data and interference information obtained in real time, to target thunder
The transmitted waveform parameter reached is predicted.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910628143.XA CN110308432A (en) | 2019-07-12 | 2019-07-12 | A kind of radar self-adaption waveform selection Activity recognition method neural network based |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910628143.XA CN110308432A (en) | 2019-07-12 | 2019-07-12 | A kind of radar self-adaption waveform selection Activity recognition method neural network based |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110308432A true CN110308432A (en) | 2019-10-08 |
Family
ID=68080070
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910628143.XA Pending CN110308432A (en) | 2019-07-12 | 2019-07-12 | A kind of radar self-adaption waveform selection Activity recognition method neural network based |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110308432A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111337918A (en) * | 2020-02-17 | 2020-06-26 | 南京航空航天大学 | Airborne radar radio frequency stealth waveform selection method based on neural network |
CN111366899A (en) * | 2020-03-27 | 2020-07-03 | 电子科技大学 | Cognitive radar anti-reconnaissance waveform selection method based on criterion switching |
CN112198484A (en) * | 2020-09-24 | 2021-01-08 | 清华大学 | Anti-interference-oriented radar frequency agility method |
CN112596031A (en) * | 2020-12-22 | 2021-04-02 | 电子科技大学 | Target radar threat degree assessment method based on knowledge graph |
CN113050045A (en) * | 2021-02-09 | 2021-06-29 | 中国人民解放军空军研究院战略预警研究所 | Intelligent comprehensive main and side lobe interference resisting system and method |
CN114417247A (en) * | 2022-01-19 | 2022-04-29 | 中国电子科技集团公司第五十四研究所 | Hyperspectral image waveband selection method based on subspace |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103336276A (en) * | 2013-07-11 | 2013-10-02 | 太原理工大学 | Cognitive radar detection device based on neural network |
CN105388461A (en) * | 2015-10-31 | 2016-03-09 | 电子科技大学 | Radar adaptive behavior Q learning method |
CN106680775A (en) * | 2016-12-12 | 2017-05-17 | 清华大学 | Method and system for automatically identifying radar signal modulation modes |
WO2018198453A1 (en) * | 2017-04-27 | 2018-11-01 | 日立オートモティブシステムズ株式会社 | Radar circuit, radar system, and radar program |
-
2019
- 2019-07-12 CN CN201910628143.XA patent/CN110308432A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103336276A (en) * | 2013-07-11 | 2013-10-02 | 太原理工大学 | Cognitive radar detection device based on neural network |
CN105388461A (en) * | 2015-10-31 | 2016-03-09 | 电子科技大学 | Radar adaptive behavior Q learning method |
CN106680775A (en) * | 2016-12-12 | 2017-05-17 | 清华大学 | Method and system for automatically identifying radar signal modulation modes |
WO2018198453A1 (en) * | 2017-04-27 | 2018-11-01 | 日立オートモティブシステムズ株式会社 | Radar circuit, radar system, and radar program |
Non-Patent Citations (3)
Title |
---|
王肖: "基于机器学习的雷达信号分选和目标识别", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
许智: "基于机器学习的认知跟踪雷达波形优化研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
贾鑫: "认知电子战概念及关键技术", 《装备学院学报》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111337918A (en) * | 2020-02-17 | 2020-06-26 | 南京航空航天大学 | Airborne radar radio frequency stealth waveform selection method based on neural network |
CN111366899A (en) * | 2020-03-27 | 2020-07-03 | 电子科技大学 | Cognitive radar anti-reconnaissance waveform selection method based on criterion switching |
CN111366899B (en) * | 2020-03-27 | 2022-08-05 | 电子科技大学 | Cognitive radar anti-reconnaissance waveform selection method based on criterion switching |
CN112198484A (en) * | 2020-09-24 | 2021-01-08 | 清华大学 | Anti-interference-oriented radar frequency agility method |
CN112198484B (en) * | 2020-09-24 | 2024-01-02 | 清华大学 | Anti-interference-oriented radar frequency agility method |
CN112596031A (en) * | 2020-12-22 | 2021-04-02 | 电子科技大学 | Target radar threat degree assessment method based on knowledge graph |
CN113050045A (en) * | 2021-02-09 | 2021-06-29 | 中国人民解放军空军研究院战略预警研究所 | Intelligent comprehensive main and side lobe interference resisting system and method |
CN114417247A (en) * | 2022-01-19 | 2022-04-29 | 中国电子科技集团公司第五十四研究所 | Hyperspectral image waveband selection method based on subspace |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110308432A (en) | A kind of radar self-adaption waveform selection Activity recognition method neural network based | |
CN110363151A (en) | Based on the controllable radar target detection method of binary channels convolutional neural networks false-alarm | |
CN109993280A (en) | A kind of underwater sound source localization method based on deep learning | |
CN106407677A (en) | Multi-target tracking method in case of loss of measurement data | |
CN112904290B (en) | Generation method of intelligent cognitive anti-interference strategy of radar | |
CN107193009A (en) | A kind of many UUV cooperative systems underwater target tracking algorithms of many interaction models of fuzzy self-adaption | |
CN104820993B (en) | It is a kind of to combine particle filter and track the underwater weak signal target tracking for putting preceding detection | |
CN108682023A (en) | Close coupling Unscented kalman tracking filter algorithm based on Elman neural networks | |
CN111965632B (en) | Radar target detection method based on Riemann manifold dimensionality reduction | |
CN109116311A (en) | Knowledge based assists the clutter suppression method of sparse iteration covariance estimation | |
Matuszewski et al. | Neural network application for emitter identification | |
CN111366899B (en) | Cognitive radar anti-reconnaissance waveform selection method based on criterion switching | |
CN107015214B (en) | Space-time adaptive processing method based on sparse Bayesian learning | |
Chang et al. | Applied state estimation and association | |
CN110033043B (en) | Radar one-dimensional range profile rejection method based on condition generation type countermeasure network | |
CN114690790A (en) | GRU-based rapid identification method for guidance law of intercepted aircraft | |
CN117744505A (en) | Deep learning-based inversion method for electromagnetic wave resistivity of azimuth while drilling | |
Li et al. | Sea/land clutter recognition for over-the-horizon radar via deep CNN | |
CN114565020B (en) | Aircraft sensor signal fusion method based on deep belief network and extended Kalman filtering | |
RU2724990C1 (en) | Method of detecting and classifying sea targets using a mathematical model of determining the type of target | |
RU2726992C1 (en) | System for detection and classification of sea targets using mathematical model of target type determination | |
CN113126086B (en) | Life detection radar weak target detection method based on state prediction accumulation | |
CN110208770B (en) | Deep neural network-based radar waveform design method | |
Yang et al. | Radar waveform design based on multi-agent reinforcement learning | |
RU2697719C1 (en) | Marine monitoring system with programmable neuron network control system |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20191008 |