CN116559794B - Radar anti-interference intelligent decision method for double-multi-domain complex neural network - Google Patents

Radar anti-interference intelligent decision method for double-multi-domain complex neural network Download PDF

Info

Publication number
CN116559794B
CN116559794B CN202310848136.7A CN202310848136A CN116559794B CN 116559794 B CN116559794 B CN 116559794B CN 202310848136 A CN202310848136 A CN 202310848136A CN 116559794 B CN116559794 B CN 116559794B
Authority
CN
China
Prior art keywords
interference
radar
network
complex
neural network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202310848136.7A
Other languages
Chinese (zh)
Other versions
CN116559794A (en
Inventor
李亚超
岑熙
顾彤
宫竹后
石光明
徐刚锋
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xidian University
Original Assignee
Xidian University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xidian University filed Critical Xidian University
Priority to CN202310848136.7A priority Critical patent/CN116559794B/en
Publication of CN116559794A publication Critical patent/CN116559794A/en
Application granted granted Critical
Publication of CN116559794B publication Critical patent/CN116559794B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Molecular Biology (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention discloses a radar anti-interference intelligent decision method of a double multi-domain complex neural network, which comprises the following steps: designing an algorithm framework for obtaining a radar state matrix based on interference signals, and constructing a memory bank for storing the radar state matrix; constructing a dual multi-domain CD3QNs complex neural network comprising an estimated value network and a target value network; constructing a loss function based on the mean square error output by the estimated value network and the target value network; constructing training data pairs based on the interfered radar echo data, and obtaining a corresponding radar state matrix by utilizing the algorithm frame; training the network by using the training data pair and the corresponding radar state matrix based on the loss function so as to carry out radar anti-interference intelligent decision through the trained network. The method has the advantages of high accuracy, strong real-time performance and stable performance of anti-interference decision aiming at complex active interference signals, and can be used for intelligent decision of the optimal anti-interference method in radar interference countermeasure.

Description

Radar anti-interference intelligent decision method for double-multi-domain complex neural network
Technical Field
The invention belongs to the technical field of radar anti-interference, and particularly relates to a radar anti-interference intelligent decision method for a double-multi-domain complex neural network.
Background
As the battlefield electromagnetic environment becomes increasingly complex, the perception capability of the traditional radar on the environment cannot meet the actual needs. The radar data characteristics are manually extracted, and the anti-interference method cannot rapidly cope with the rapid and changeable interference fight scenes. In addition, when the anti-interference method is selected, the interference condition is judged mainly by the experience of expert personnel, then anti-interference measures are selected from an anti-interference method library, the process is seriously dependent on personal experience, the intelligent degree is low, and the robustness is not achieved.
In recent years, with the development of artificial intelligence technology, machine learning algorithms are gradually applied to radar anti-interference. Among them, reinforcement learning methods are attracting attention because of the process of effectively characterizing anti-interference decisions. For example, criminal strong et al put forward an anti-interference decision method based on Q-learning in the paper published by system engineering and electronic technology (system engineering and electronic technology, 2018,5, 1030-05), which mainly constructs a radar state matrix by manually extracting characteristic parameters such as radar threat level, radar working model and the like, and then decides the best anti-interference measure according to the learned interference pattern-anti-interference return Q voting. In addition, some students have studied an anti-interference intelligent decision method based on deep learning, for example, in the patent literature of the university of western electronic technology (DQN algorithm-based radar anti-interference intelligent decision method) (application publication number CN113341383 a) filed by the university of western electronic technology, a radar anti-interference intelligent decision method based on DQN network is proposed, which can quickly decide an optimal anti-interference method by constructing a deep learning neural network to replace a Q table in Q-learning to calculate an interference pattern-anti-interference return Q table.
However, the anti-interference decision method based on Q-learning mainly adopts manual parameter extraction, the calculated amount is large, the efficiency is low, and the Q value is easy to be over-estimated by updating the Q value by a simple value iteration method, so that the anti-interference effect of the radar is poor, and the decision accuracy is low; in addition, the method stores data in an interference pattern-anti-interference return matrix form, so that the memory requirement under a large amount of interference data is difficult to meet. The two neural networks in the anti-interference intelligent decision-making method of the radar based on the DQN network have the problems of high coupling training difficulty and low convergence speed; and the convolution kernel of the neural network is a real number and is not matched with complex data of radar echo, so that the accuracy of decision making is affected.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a radar anti-interference intelligent decision method of a double-multi-domain complex neural network, so as to avoid the problems of manually separating signal characteristics, large calculated amount, low decision efficiency, low accuracy and network discomfort of radar data in the anti-interference decision process.
The technical idea of the invention is as follows: constructing a double multi-domain CD3QNs (Complex Dueling Double Deep Q Network Softmax) complex neural network, inputting a radar state matrix into the double multi-domain CD3QNs complex neural network, and combiningThe algorithm makes a decision on the optimal anti-interference method in the training process, so that the model has the optimal anti-interference decision capability. The maximum value of the final convergence matrix is obtained through the trained double multi-domain CD3QNs complex neural network, so that the optimal anti-interference method is efficiently and accurately decided.
The technical problems to be solved by the invention are realized by the following technical scheme:
a radar anti-interference intelligent decision method of a double multi-domain complex neural network comprises the following steps:
step 1: designing an algorithm framework for obtaining a radar state matrix based on interference signals, and constructing a memory bank for storing the radar state matrix;
step 2: constructing a dual multi-domain CD3QNs complex neural network comprising an estimated value network and a target value network;
step 3: constructing a loss function of a double multi-domain CD3QNs complex neural network based on the mean square error output by the estimated value network and the target value network;
step 4: constructing training data pairs based on the interfered radar echo data;
step 5: processing the training data pairs by utilizing the algorithm frame in the step 1 to obtain a corresponding radar state matrix; and training the double multi-domain CD3QNs complex neural network by utilizing the training data pairs and the corresponding radar state matrixes based on the loss function so as to carry out radar anti-interference intelligent decision through the trained network.
The invention has the beneficial effects that:
1. according to the intelligent radar anti-interference decision method provided by the invention, by constructing the double multi-domain CD3QNs complex neural network, the problem that the complex radar interference data cannot be adapted by using a real convolution kernel in the prior art is solved, the problems of manually separating signal characteristics, large calculated amount, low decision efficiency, low accuracy, poor stability and network unadapted radar data in the anti-interference decision process are avoided, the effective extraction of radar interference data characteristics is realized, and the accuracy is improved;
2. the invention adopts the memory to store radar interference data, can meet the memory requirement under a large amount of interference data, and realizes the optimal anti-interference method for decision-making while training, thereby ensuring that the invention has the capability of deciding the optimal anti-interference method in real time and improving the decision-making efficiency of the radar anti-interference method;
3. in the invention, the convergence matrix is calculated through the estimated value network of the double multi-domain CD3QNs complex neural network in the process of training the network, and the maximum convergence value is selected from the calculated convergence matrix, so that the direct mapping from the radar state input to the convergence value is realized, the problems of low decision accuracy and high training difficulty and low convergence speed caused by complex calculation steps and easy overestimation in the conventional algorithm are overcome, and the invention has the advantages of simple calculation and high decision accuracy.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
FIG. 1 is a schematic flow chart of a radar anti-interference intelligent decision method of a double multi-domain complex neural network provided by an embodiment of the invention;
FIG. 2 is a schematic diagram of an operation of a complex convolution layer according to an embodiment of the present disclosure;
FIG. 3 is a schematic diagram of an estimated value network according to an embodiment of the present invention;
FIG. 4 is a block diagram of a dual multi-domain CD3QNs complex neural network according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a convergence matrix for anti-interference of radar echo data subjected to linear function shift interference using the method of the present invention and the prior art method.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but embodiments of the present invention are not limited thereto.
Example 1
Referring to fig. 1, fig. 1 is a flow chart of a radar anti-interference intelligent decision method based on a dual multi-domain complex neural network according to an embodiment of the present invention, which includes:
step 1: an algorithm framework for obtaining a radar state matrix based on the interference signals is designed, and a memory bank is constructed for storing the radar state matrix.
In this embodiment, the algorithm framework for obtaining the radar state matrix based on the interference signal includes:
11 Acquiring active interference signals transmitted by an jammer。
Specifically, the jammer transmits an active interference signal, and the radar receives the interference signal transmitted by the jammerThe method comprises the steps of carrying out a first treatment on the surface of the The active interference type can be any one of a linear function frequency-shift signal, a coherent motion false target interference signal or a slice forwarding interference signal.
12 Using)Algorithm, selecting any anti-interference method +.>Calculating anti-interference method->Signal-to-distortion ratio against active interference signal, which is taken as anti-interference benefit +.>。
In general, the anti-interference method library comprises a subspace projection ESP method, an inaccurate Lagrangian multiplier IALM method, a LocNet method, a RecNet method and other anti-interference methods.
Algorithm is denoted by +.>The probability of the magnitude selects the unselected anti-interference method from the anti-interference action library to +.>Selecting an anti-interference method from the selected actions,/probability of->The value range of (2) is +.>。
Specifically, the present embodiment calculates the signal-to-distortion ratio using the following formula:
；
in the formula ,for undisturbed normalized radar echo, +.>In order to perform the radar echo after the anti-jamming method,representing the signal-to-distortion ratio.
13 Construction of a radar state matrixThe expression is:
。
in addition, the embodiment also designs a memory bank with a certain capacity for storing the dataStored as a set of data in the memory bank.
Alternatively, the memory bank may be represented by D, and may have a capacity of 3000 and a size of 3000×3, and when the amount of stored interference data is greater than 3000, the data in the memory bank D may be overwritten.
Step 2: a dual multi-domain CD3QNs complex neural network is constructed that includes an estimated value network and a target value network.
Specifically, in the present embodiment, the estimated value network and the target value network adopt the same network structure. The construction process of the network is described in detail below.
21 A complex convolution kernel is designed to obtain the operation mode of the complex convolution layer.
Specifically, referring to fig. 2, fig. 2 is a schematic diagram of an operation manner of a complex convolution layer according to an embodiment of the present invention, where the operation manner is as follows:
；
in the formula , and />Representing a complex feature map and a complex convolution kernel, respectively, < >> and />Representing the real and imaginary parts of the complex signature, +.>Representing imaginary number listBit (s)/(s)> and />Representing the real and imaginary parts of the complex convolution kernel +.>Representing complex multiplication.
22 A time-frequency domain neural network branch comprising 3 complex convolution layers and 2 complex full connection layers is constructed.
Specifically, referring to fig. 3, fig. 3 is a schematic structural diagram of an estimated value network according to an embodiment of the present invention. The 1 st complex convolution layer, the 2 nd complex convolution layer, the 1 st average pooling layer (namely the 3 rd complex convolution layer), the 1 st full connection layer and the 2 nd full connection layer form a time-frequency domain neural network branch.
23 A time domain neural network branch including 3 complex fully connected layers, such as the 3 rd fully connected layer, the 4 th fully connected layer, and the 5 th fully connected layer in fig. 3 is constructed.
24 A neural network output portion including 1 plurality of fully connected layers and Softmax layers, such as the 6 th fully connected layer and Softmax layer in fig. 3.
The operation mode of the Softmax layer is as follows:
；
in the formula ,indicate->Probability of selecting each feature, corresponding to the probability of selecting the anti-interference method, is within the range of +.>；/>Indicate->Zhang Fushu characteristic map, < >>Representing modulo values, +.>As an exponential function +.>The total number of the feature graphs of the previous layer.
25 Based on the time-frequency domain neural network branch, the time domain neural network branch and the neural network output part, an estimated value network is built, and the whole estimated value network structure is shown in figure 3.
And then the target value network is continuously built.
Optionally, in this embodiment, the structure of the target value network may take the same structure as the estimated value network, and step 2 further includes:
26 The structure of the estimated value network is duplicated to form a target value network, thereby obtaining a dual multi-domain CD3QNs complex neural network including the estimated value network and the target value network, the structure block diagram of which is shown in fig. 4.
After the network structure is built, network parameters also need to be set.
Optionally, in this embodiment, the estimated value network of the dual multi-domain CD3QNs complex neural networkThe layer-specific parameters of (a) are shown in table 1 below:
TABLE 1
Step 3: constructing a loss function of a double multi-domain CD3QNs complex neural network based on the mean square error output by the estimated value network and the target value network;
in particular, the method comprises the steps of,after the dual multi-domain CD3QNs complex neural network structure is obtained, it is trained. The loss function of the double multi-domain CD3QNs complex neural network needs to be constructed before trainingThe expression is:
；
in the formula ,representing a loss function->Representing the next radar status +.>The anti-interference benefit is obtained in the process,representing discount factor, value range->，/>Representing a convergence matrix->Indicating radar status +.>Down through the target value network->Calculating convergence matrix->Action corresponding to maximum value->，/>Representing the current radar status matrix +.>Execution of action->Through the evaluation value network->And calculating the obtained convergence matrix.
Step 4: training data pairs are constructed based on the disturbed radar echo data.
In particular, for radar echo data subject to interferencePerforming short-time Fourier transform to obtain time-frequency diagram corresponding to echo data>And construct training data pair +.>。
Wherein, parameters of the short-time Fourier transform are set as follows: the window function is a hamming window, a window length of 63, a step size of 1, and a fourier transform point number of 256.
Step 5: processing the training data pair by utilizing the algorithm frame in the step 1 to obtain a corresponding radar state matrix; training the double multi-domain CD3QNs complex neural network by using the training data pair and the corresponding radar state matrix based on the loss function so as to carry out intelligent radar anti-interference decision through the trained network.
Specifically, step 5 includes:
51 Setting training parameters including maximum number of iterationsLearning rate->Lot size->。
Optionally, the maximum number of iterationsCan be set to 2000, learning rate +.>Set to 0.001, batch size +.>Set to 16.
52 Selecting)Training data pair->And utilize the algorithm frame pair of step 1 +.>Processing the training data pairs to obtain +.>A radar state matrix; at the same time, this is->The radar state matrix is stored in a constructed memory bank.
The invention adopts the memory to store radar interference data, can meet the memory requirement under a large amount of interference data, realizes the optimal anti-interference method for decision-making while training, ensures that the invention has the capability of deciding the optimal anti-interference method in real time, and improves the decision-making efficiency of the radar anti-interference method.
53 To be taken out)Training data pairs and corresponding->The radar state matrix is used as data of one iteration training, and a momentum method is used for training an estimated value network by combining a loss function.
Specifically, the training of the estimated value network by using the momentum method in the embodiment includes the following steps:
a) Is provided withThe initial value is 1 for the iteration times; />Is->Momentum at iteration, let ∈ ->；
b) Network of estimated valuesThe layer parameters of (a) use the letter +.>Representing, calculate loss function->Layer parameters relative to the estimated value network>Gradient size +.>The calculation formula is as follows:
；
wherein ,representing partial differentiation;
c) According to the gradient magnitudeUse learning rate->Calculate->Momentum ∈ at iteration>The calculation formula is as follows:
；
d) For each layer of parameters of the estimated value networkUpdating, wherein the updating formula is as follows:
；
e) Iteratively performing steps b) -d) untilReach maximum iteration number->And obtaining a trained estimated value network.
54 When estimating the accumulated training times of the value networkWhen the value of (2) is an integer multiple of 100, copying the parameters of the estimated value network to the target value network until the round of training is finished;
55 Repeating the training process of steps 52) -54) until an iteration stop condition is reached, resulting in a trained network.
It can be seen that in the training process of the dual multi-domain CD3QNs complex neural network in this embodiment, the estimated value network is trained first, and then the parameters thereof are copied to the target value network periodically. And the loss function used in the training process calculates a convergence matrix through an estimated value network of a double multi-domain CD3QNs complex neural network, and selects the maximum convergence value from the convergence matrix, so that direct mapping from radar state input to convergence value is realized, the problems of low decision accuracy and high training difficulty and low convergence speed caused by complex calculation steps and easy overestimation in the conventional algorithm are overcome, and the method has the advantages of simplicity in calculation and high decision accuracy.
After the trained network is obtained, the intelligent decision of radar anti-interference can be made through the network.
In particular, first of all radar echo data with interference signalsConstructing data pairs by corresponding time-frequency diagrams. Then, the constructed data pair +.>Inputting into a trained estimated value network, and calculating to obtainIs a q-value convergence matrix of (c). And finally, selecting an anti-interference method corresponding to the row with the maximum convergence value in the convergence matrix as an optimal anti-interference method.
The intelligent radar anti-interference decision method provided by the invention solves the problem that the complex radar interference data cannot be adapted by using a real number convolution kernel in the prior art by constructing the double multi-domain CD3QNs complex neural network, avoids the problems of manually separating signal characteristics, large calculated amount, low decision efficiency, low accuracy, poor stability and network unadapted radar data in the anti-interference decision process, realizes effective extraction of radar interference data characteristics, and improves the accuracy.
Example two
The beneficial effects of the invention are verified by simulation tests.
1. Simulation conditions:
the hardware platform of the simulation experiment of the invention is: intel (R) Core (TM) i7-10700CPU,2.90GHz, 64G memory and NVIDIA GeForce RTX 3080 GPU.
The software platform of the simulation experiment of the invention is: pycharm2021.
2. Simulation content and result analysis:
under the simulation condition, the method provided by the invention is used for carrying out anti-interference intelligent decision simulation on radar echo data subjected to linear function frequency shift interference, and a signal model of the linear function frequency shift interference is as follows:
；
in the formula ,for disturbing pulse width +.>Is the center frequency of the chirp signal, +.>For the frequency modulation rate of the chirp signal, +.>Shifting the initial center frequency of the interference signal for a linear function,/->The initial tone frequency of the interfering signal is shifted as a linear function.
Simulation experiment setting of the inventionIn algorithm +.>Initial value is 1.0, each roundTraining attenuation is 0.05, minimum attenuation is 0.8, and the number of the designed countermeasures is 2000. In the countermeasure process, a q value convergence matrix of the 4 methods is shown in fig. 5, wherein an ordinate q-value represents an anti-interference effective q value of each method under linear frequency shift interference, an abscissa epicode represents the countermeasure times, and the larger the q value is, the better the anti-interference effect of the method is. ESP, IAILM, locNet and RecNET represent different anti-interference treatment methods, respectively.
As can be seen from fig. 5, the q-value of the RecNet method is highest and converges fastest.
In addition, table 2 below also shows the number of times different anti-interference methods are selected under linear frequency shift interference.
TABLE 2
It can be seen that the accuracy of the intelligent decision of the radar is 86.05% when RecNet is selected 1721 times by the method of the invention.
In summary, the method of the invention carries out intelligent decision on the anti-interference method for the interference signal by constructing the double multi-domain CD3QNs complex neural networkThe method has the advantages of rapid and stable values and high decision accuracy, and the method has stable performance and high decision efficiency, and can provide intelligent decisions of the optimal anti-interference method.
The foregoing is a further detailed description of the invention in connection with the preferred embodiments, and it is not intended that the invention be limited to the specific embodiments described. It will be apparent to those skilled in the art that several simple deductions or substitutions may be made without departing from the spirit of the invention, and these should be considered to be within the scope of the invention.

Claims (7)

1. A radar anti-interference intelligent decision method based on a double multi-domain complex neural network is characterized by comprising the following steps:
step 1: designing an algorithm framework for obtaining a radar state matrix based on interference signals, and constructing a memory bank for storing the radar state matrix; comprising the following steps:
11 Acquiring active interference signal s emitted by jammer t ；
12 Using epsilon-greedy algorithm to select any one anti-interference method a) t Calculating the anti-interference method a t Against the signal distortion ratio of the active interference signal, taking the signal distortion ratio as anti-interference benefit R t ；
13 Construction of a radar state matrix S t The expression is: s is S t ＝[s t ,a t ,R t ]；
Step 2: constructing a dual multi-domain CD3QNs complex neural network comprising an estimated value network and a target value network; comprising the following steps:
21 A complex convolution kernel is designed to obtain the operation mode of the complex convolution layer;
22 A time-frequency domain neural network branch comprising 3 complex convolution layers and 2 complex full connection layers is constructed;
23 Constructing a time domain neural network branch comprising 3 complex full-connection layers;
24 Building a neural network output part comprising 1 complex full-connection layer and 1 Softmax layer;
25 Building an estimated value network based on the time-frequency domain neural network branch, the time-domain neural network branch and the neural network output part;
26 Copying the structure of the estimated value network to form a target value network, thereby obtaining a dual multi-domain CD3QNs complex neural network comprising the estimated value network and the target value network;
step 3: constructing a loss function of a double multi-domain CD3QNs complex neural network based on the mean square error output by the estimated value network and the target value network;
step 4: constructing training data pairs based on the interfered radar echo data;
step 5: processing the training data pairs by utilizing the algorithm frame in the step 1 to obtain a corresponding radar state matrix; and training the double multi-domain CD3QNs complex neural network by utilizing the training data pairs and the corresponding radar state matrixes based on the loss function so as to carry out radar anti-interference intelligent decision through the trained network.
2. The intelligent decision-making method for radar anti-interference of double multi-domain complex neural network according to claim 1, wherein the complex convolution layer designed in step 21) operates as follows:
wherein F and M represent complex feature maps and complex convolution kernels, respectively, F R and FI Representing the real and imaginary parts of a complex feature map, i representing the imaginary units, M R and MI Representing the real and imaginary parts of the complex convolution kernel, x represents the complex multiplication.
3. The intelligent decision-making method for radar anti-interference of a dual multi-domain complex neural network according to claim 1, wherein in step 24), the Softmax layer operates as follows:
wherein A (i) represents the probability that the ith feature is selected, and corresponds to the anti-interference method a i The selected probability has a value range of (0, 1); x is x i Representing the i Zhang Fushu feature map, |·| represents the modulo value, exp (·) is an exponential function, and k is the total number of feature maps of the previous layer.
4. The intelligent decision-making method for radar anti-interference of a dual multi-domain complex neural network according to claim 1, wherein in step 3, the expression of the loss function of the dual multi-domain CD3QNs complex neural network is:
where Loss represents a Loss function, R t+1 Representing the next radar state S t+1 The anti-interference benefit obtained during the process is that gamma represents discount factor, and the value range is [0,1]Q () represents the convergence matrix,representing radar status S t+1 Down through a target value network Q net2 Calculating the action a, Q (S) corresponding to the maximum value of the convergence matrix Q t A) represents the current radar state matrix S t Performing action a through an estimate network Q net1 And calculating the obtained convergence matrix.
5. The intelligent decision-making method for radar anti-interference of a dual multi-domain complex neural network according to claim 1, wherein the step 4 comprises:
for disturbed radar echo data E t Performing short-time Fourier transform to obtain a time-frequency diagram M corresponding to echo data t And construct training data pair [ E t ,M t ]。
6. The intelligent decision-making method for radar anti-interference of a dual multi-domain complex neural network according to claim 5, wherein step 5 comprises:
51 Setting training parameters including a maximum iteration number M, a learning rate rho and a batch size N;
52 Selecting N training data pairs [ E ] t ,M t ]Processing the N training data pairs by utilizing the algorithm frame in the step 1, and correspondingly obtaining N radar state matrixes;
53 Taking the N training data pairs and the N corresponding radar state matrixes as data of one iteration training, and training an estimated value network by adopting a momentum method in combination with a loss function;
54 When the value of the accumulated training times n of the estimated value network is an integer multiple of 100, copying the parameters of the estimated value network to the target value network until the round of training is finished;
55 Repeating the training process of steps 52) -54) until an iteration stop condition is reached, resulting in a trained network.
7. The intelligent decision-making method for radar anti-interference of a dual multi-domain complex neural network according to claim 6, wherein in step 53), training the estimated value network using a momentum method comprises:
a) Setting k as iteration times and the initial value as 1; m is m k Let m be the momentum at the kth iteration 0 ＝0；
b) And calculating the gradient size grad of the Loss function Loss relative to each layer of parameter theta of the estimated value network, wherein the calculation formula is as follows:
wherein ,representing partial differentiation;
c) Calculating the momentum m at the kth iteration using the learning rate ρ according to the gradient magnitude grad k The calculation formula is as follows:
m k ＝0.9×m k-1 +ρ·grad；
d) Updating the parameters theta of each layer of the estimated value network, wherein an updating formula is as follows:
θ＝θ-m k ；
e) And (c) iteratively executing the steps b) -d) until K reaches the maximum iteration number K, and obtaining the trained estimated value network.
CN202310848136.7A 2023-07-12 2023-07-12 Radar anti-interference intelligent decision method for double-multi-domain complex neural network Active CN116559794B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310848136.7A CN116559794B (en) 2023-07-12 2023-07-12 Radar anti-interference intelligent decision method for double-multi-domain complex neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310848136.7A CN116559794B (en) 2023-07-12 2023-07-12 Radar anti-interference intelligent decision method for double-multi-domain complex neural network

Publications (2)

Publication Number Publication Date
CN116559794A CN116559794A (en) 2023-08-08
CN116559794B true CN116559794B (en) 2023-09-29

Family

ID=87503972

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310848136.7A Active CN116559794B (en) 2023-07-12 2023-07-12 Radar anti-interference intelligent decision method for double-multi-domain complex neural network

Country Status (1)

Country Link
CN (1) CN116559794B (en)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106228201A (en) * 2016-06-20 2016-12-14 电子科技大学 A kind of anti-Deceiving interference method of synthetic aperture radar based on shade characteristic
CN113109780A (en) * 2021-03-02 2021-07-13 西安电子科技大学 High-resolution range profile target identification method based on complex number dense connection neural network
CN113341383A (en) * 2021-05-31 2021-09-03 西安电子科技大学 Radar anti-interference intelligent decision method based on DQN algorithm
CN115200702A (en) * 2022-06-17 2022-10-18 北京理工大学 Computational imaging method and device based on complex neural network
CN115343680A (en) * 2022-07-28 2022-11-15 南京邮电大学 Radar anti-interference decision method based on deep reinforcement learning and combined frequency hopping and pulse width distribution
WO2023092923A1 (en) * 2021-11-26 2023-06-01 北京理工大学 Composite interference signal recognition method and system
CN116338598A (en) * 2023-05-31 2023-06-27 西安电子科技大学 Anti-interference intelligent decision method for radar based on backtracking DQN

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106228201A (en) * 2016-06-20 2016-12-14 电子科技大学 A kind of anti-Deceiving interference method of synthetic aperture radar based on shade characteristic
CN113109780A (en) * 2021-03-02 2021-07-13 西安电子科技大学 High-resolution range profile target identification method based on complex number dense connection neural network
CN113341383A (en) * 2021-05-31 2021-09-03 西安电子科技大学 Radar anti-interference intelligent decision method based on DQN algorithm
WO2023092923A1 (en) * 2021-11-26 2023-06-01 北京理工大学 Composite interference signal recognition method and system
CN115200702A (en) * 2022-06-17 2022-10-18 北京理工大学 Computational imaging method and device based on complex neural network
CN115343680A (en) * 2022-07-28 2022-11-15 南京邮电大学 Radar anti-interference decision method based on deep reinforcement learning and combined frequency hopping and pulse width distribution
CN116338598A (en) * 2023-05-31 2023-06-27 西安电子科技大学 Anti-interference intelligent decision method for radar based on backtracking DQN

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
余选召 ; 吴迪 ; 王君 ; .BP神经网络在雷达抗干扰效能评估中的应用.信息系统工程.2013,(05),全文. *
对多功能雷达的DQN认知干扰决策方法;张柏开;朱卫纲;;系统工程与电子技术(04);全文 *

Also Published As

Publication number Publication date
CN116559794A (en) 2023-08-08

Similar Documents

Publication Publication Date Title
CN110427846B (en) Face recognition method for small unbalanced samples by using convolutional neural network
GB2547816B (en) Actually-measured marine environment data assimilation method based on sequence recursive filtering three-dimensional variation
CN112949387B (en) Intelligent anti-interference target detection method based on transfer learning
CN109993280A (en) A kind of underwater sound source localization method based on deep learning
CN110456332B (en) Underwater acoustic signal enhancement method based on automatic encoder
CN113361777B (en) Runoff prediction method and system based on VMD decomposition and IHHO optimization LSTM
CN112949820B (en) Cognitive anti-interference target detection method based on generation of countermeasure network
CN109067427B (en) A kind of frequency hop sequences prediction technique based on Optimization-type wavelet neural network
CN114595732B (en) Radar radiation source sorting method based on depth clustering
CN113627066A (en) Displacement prediction method for reservoir bank landslide
CN104331583B (en) A kind of Multifractal Modeling method based on Observed sea clutter
CN113341383A (en) Radar anti-interference intelligent decision method based on DQN algorithm
CN115442191B (en) Communication signal noise reduction method and system based on relative average generation countermeasure network
CN110119540A (en) A kind of multi output gradient promotion tree modeling method for survival risk analysis
CN114895263A (en) Radar active interference signal identification method based on deep migration learning
CN112882115A (en) Magnetotelluric signal denoising method and system based on GWO optimized wavelet threshold
CN116304546A (en) Heat supply system heat station fault diagnosis method and system based on sound signals
CN115051929A (en) Network fault prediction method and device based on self-supervision target perception neural network
CN116559794B (en) Radar anti-interference intelligent decision method for double-multi-domain complex neural network
CN113864665B (en) Fluid pipeline leakage positioning method based on adaptive ICA and improved RLS filter
CN111368653A (en) Low-altitude small target detection method based on R-D (R-D) graph and deep neural network
KR102120150B1 (en) Learning method and learning device for variational interference using neural network and test method and test device for variational interference using the same
CN112487933B (en) Radar waveform identification method and system based on automatic deep learning
CN108960326A (en) A kind of point cloud fast partition method and its system based on deep learning frame
CN116958784A (en) Method for rapidly detecting wheat scab spike and severity in natural environment

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