CN109726653A - Radar Signal Recognition method based on RNN-DenseNet network - Google Patents
Radar Signal Recognition method based on RNN-DenseNet network Download PDFInfo
- Publication number
- CN109726653A CN109726653A CN201811554579.0A CN201811554579A CN109726653A CN 109726653 A CN109726653 A CN 109726653A CN 201811554579 A CN201811554579 A CN 201811554579A CN 109726653 A CN109726653 A CN 109726653A
- Authority
- CN
- China
- Prior art keywords
- signal
- network
- rnn
- densenet
- indicate
- 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
Landscapes
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a kind of Radar Signal Recognition method based on RNN-DenseNet network, mainly solve the problems, such as that the prior art is insufficient to radar signal feature information extraction low with accuracy of identification.Its scheme are as follows: the data set for generating radar signal is emulated with business software；The signal of radar data collection is first exported in the form of sequence, then data set signal is subjected to time-frequency conversion, and export in the form of signal time-frequency figure；Classification belonging to signal is marked in the data set of two kinds of forms output, and training set and test set is made；The network parameter of RNN and DenseNet are set, by the last layer Fusion Features of the two networks；Use training set signal training RNN-DenseNet network；Test set signal is sent into trained network, network output is that radar signal predicts classification.The present invention can sufficiently extract radar signal feature, improve the discrimination of signal, the Radar Signal Recognition that can be used under complex electromagnetic environment.
Description
Technical field
The invention belongs to signal processing technology field, in particular to a kind of Radar Signal Recognition method can be used for electronics feelings
Report is scouted, in electronic support and threat warning system.
Background technique
With the development of electronic information field, electronic countermeasure is in electronic intelligence reconnaissance, electronic support and threat warning system
In play an important role, radar emitter signal identification is important link in electronic countermeasure.
As electromagnetic environment becomes extremely complex in electronic countermeasure, it is in particular in various, empty in Radar emitter number
Between distribution it is wide and signal is serious in time-domain and frequency-domain aliasing.The radar signal occurred in a relatively short period of time is up to tens of thousands of
Or even it is hundreds of thousands of, can at a time occur a large amount of signals simultaneously.At the same time, the modulation system of signal becomes complicated, signal
The Parameters variations such as frequency it is rapid.With the raising of radar hardware device performance, radar can generate not according to the demand of people
Modulation with the signal of modulation system, especially in frequency and phase；Nowadays more and more researchers put into this
Field, everybody wishes to introduce into the method for update to overcome difficulties.
Patent document " Radar emitter letter based on one-dimensional convolutional neural networks of the Xian Electronics Science and Technology University in its application
Number recognition methods " disclose in (201710361523.2 application publication number CN of application number, 107220606 A) it is a kind of based on one
Tie up the recognition methods of the radar emitter signal of convolutional neural networks.Radar signal is extracted by convolutional Neural, is avoided
Complicated artificial design features process in conventional method.But the shortcoming that this method still has is, network is shallower,
Structure is single, is not enough to characteristic processing, the space that discrimination is still improved.
Summary of the invention
The present invention be directed to radar emitter signals to identify the shortcomings of the prior art, propose a kind of based on RNN-
The method of DenseNet network, with from two different sides of time-frequency figure of radar emitter signal sequence and radar emitter signal
Feature extraction is carried out in face of radar emitter signal, improves Radar Signal Recognition rate.
To achieve the above object, implementation of the invention includes the following:
1) with MATLAB software emulation generate radar signal data set, the data set signal include general pulse signal,
Linear FM signal, NLFM signal, Coded Signals, four phase encoded signals, two frequency encoded signals and four frequencies encode
This seven kinds of signals of signal, wherein every kind of signal generates 3000 samples from -10dB to 2dB every 2dB signal-to-noise ratio；
2) data set signal is pre-processed:
2a) 1) signal for generating data set is first exported in the form of sequence, then data set signal is subjected to time-frequency conversion,
And it is exported in the form of signal time-frequency figure；
Classification belonging to signal 2b) is marked in the data set of two kinds of forms output, is extracted out at random from every class signal
2400 samples are as training sample, and 600 samples are as test sample；
3) RNN-DenseNet network is constructed:
RNN network parameter 3a) is set:
Memory network LSTM constructs RNN-DenseNet network to the length selected in Recognition with Recurrent Neural Network RNN in short-term, i.e., will
Network cellular number in LSTM network is set as 2, will forget parameter and is set as 1, sets the loss function of LSTM network to
Logarithm loss function carrys out Schistosomiasis control rate using AdamOptimizer algorithm, uses index linear unit function as activation letter
Number；
DenseNet network parameter 3b) is set:
If DenseNet network is made of 3 intensive blocks, the convolution operation and 2* of 1*1 are successively carried out between adjacent intensive block
2 average pondization operation, and 32 are set by the growth rate of the DenseNet network；
3c) the last layer feature of RNN network and DenseNet network is merged, and softmax classifier is made
For the output layer of network；
4) training RNN-DenseNet network:
The number of iterations that RNN-DenseNet network 4a) is arranged is 3000, and learning rate is set as 0.01；
The training sample data of signal sequence 4b) are input to 3a) in the LSTM network that sets, by signal time-frequency figure
Training sample data input 3b) in the DenseNet network that sets, when the number of iterations reaches 3000, terminates training, obtain
Trained network model；
5) data of test set are input in trained RNN-DenseNet network, export the pre- of each test signal
Survey classification.
The present invention has the advantage that
First, the DenseNet network application in CNN network in Radar Signal Recognition, is alleviated traditional network by the present invention
In performance degradation is led to by gradient disappearance problem, strengthen the transmitting between characteristic pattern.
Second, the present invention in RNN-DenseNet network in terms of the sequential extraction procedures of radar signal and characteristic pattern two to radar
Signal carries out feature extraction, merges to obtained characteristic information, so that the information of radar signal is fully used, shows
Work improves discrimination.
Detailed description of the invention
Fig. 1 is implementation flow chart of the invention.
Specific embodiment
Referring to Fig.1, radar emitter signal recognition methods of the invention, implementation step are as follows:
Step 1: generating radar signal data set.
The data set of radar signal is generated with MATLAB software emulation, which includes general pulse signal, line
Property FM signal, NLFM signal, Coded Signals, four phase encoded signals, two frequency encoded signals and four frequency coding letter
Number this seven kinds of signals, wherein every kind of signal generates 3000 samples from -10dB to 2dB every 2dB signal-to-noise ratio；
The sample frequency of this 7 kinds of radar signals is disposed as 2GHz, and sampling number is disposed as 512；
General pulse signal, linear FM signal, NLFM signal, Coded Signals, four phase encoded signals this 5
The Carrier frequency configuration of kind radar signal is 150MHz.
Coded Signals use 13 Barker codes.
Four phase encoded signal signals use 16 Frank codes.
Two carrier frequency of two frequency encoded signals are respectively set to 200MHz and 400MHz, and use 13 Barker codes.
Four carrier frequency of four frequency encoded signals are respectively set to 100MHz, 300MHz, 500MHz and 700MHz.
Step 2: radar signal being pre-processed, training sample and test sample are obtained.
The signal for generating data set to step 1 is first exported in the form of sequence, then data set signal is carried out time-frequency conversion,
And it is exported in the form of signal time-frequency figure；
Classification belonging to signal is marked in the data set of two kinds of forms output, extracts 2400 out at random from every class signal
Sample is as training sample, and 600 samples are as test sample；
Step 3: building RNN-DenseNet network.
RNN network parameter 3a) is set:
LSTM network cellular 3a1) is set:
Memory network LSTM constructs RNN-DenseNet network to the length selected in Recognition with Recurrent Neural Network RNN in short-term, by LSTM
Network cellular number in network is set as 2, will forget parameter and is set as 1, wherein LSTM network cellular is expressed as follows:
ht=σ (Wo·[ht-1,xt]+bo)*tanh(ft*Ct-1+Rt),
In formula, t represents time, htFor the output of t moment LSTM network cellular, σ is Sigmoid activation primitive, WOIndicate defeated
The weight gone out, ht-1Indicate the output of a cellular, xtIndicate the input of t moment LSTM network cellular, boIndicate out gate
Biasing, ftIndicate the output that door is forgotten in t moment LSTM network cellular, Ct-1Indicate the cellular state at t-1 moment, RtIndicate t
Tanh operation is done in the output of Memory-Gate in moment LSTM network cellular, tanh expression；
LSTM network losses function 3a2) is set:
Common loss function have L1 loss function, L2 loss function, Huber loss function, Log-cosh loss function and
Logarithm loss function, this example are selected but are not limited to logarithm loss function；
The LSTM network optimization algorithm 3a3) is set:
Common network optimization algorithm have stochastic gradient descent algorithm, small lot gradient descent algorithm, Adagrad algorithm and
AdamOptimizer algorithm, this example are selected but are not limited to AdamOptimizer algorithm；
LSTM network activation function 3a4) is set:
Common activation primitive have sigmoid activation primitive, tanh activation primitive, non-linear unit activation primitive and
Index linear unit activating function, this example are selected but are not limited to index linear unit activating function, linear unit activation primitive
It is expressed as follows:
In formula, x indicates input value, and α is a constant coefficient, and f (x) indicates the output of index linear unit function；
DenseNet network parameter 3b) is set:
If DenseNet network is made of 3 intensive blocks, the convolution operation and 2* of 1*1 are successively carried out between adjacent intensive block
2 average pondization operation, and 32 are set by the growth rate of the DenseNet network,
Wherein intensive block is expressed as follows:
xl=Hl([x0,x1,...xl-1]),
In formula, l indicates the number of plies of intensive block, and H indicates a nonlinear transformation, HlExpression does one to l layers of intensive block
Nonlinear transformation, [x0,x1,...,xl-1] indicate x0To xl-1Output characteristic pattern do and link, xlIt is the output of intensive block；
3c) Fusion Features:
The feature of LSTM network and DenseNet network the last layer is merged, is input in softmax classifier,
Using softmax classifier as the final output layer of RNN-DenseNet network, wherein softmax classifier is expressed as follows:
In formula, yiIndicate the scoring vector of i-th of element,Expression seeks index to the scoring vector of i-th of element,Indicate that the index to all elements scoring vector is summed, LiIndicate the score value of i-th of element output.
Step 4: training RNN-DenseNet network.
The number of iterations that RNN-DenseNet network 4a) is arranged is 3000, and learning rate is set as 0.01；
The training sample data of signal sequence 4b) are input to 3a) in the LSTM network that sets, by signal time-frequency figure
Training sample data input 3b) in the DenseNet network that sets, training is iterated to RNN-DenseNet network, when repeatedly
When generation number reaches 3000, terminates training, obtain trained RNN-DenseNet network model.
Step 5: the data of test set being input in trained RNN-DenseNet network, each test signal is exported
Prediction classification.
Above description is only example of the present invention, does not constitute any limitation of the invention, it is clear that for
It, all may be without departing substantially from inventive principle, structure after understand the contents of the present invention and principle for one of skill in the art
In the case of, modifications and changes in form and details are carried out, but these modifications and variations based on inventive concept are still at this
Within the claims of invention.
Claims (6)
1. the Radar Signal Recognition method based on RNN-DenseNet network, which is characterized in that include the following:
1) data set of radar signal is generated with MATLAB software emulation, which includes general pulse signal, linear
FM signal, NLFM signal, Coded Signals, four phase encoded signals, two frequency encoded signals and four frequency encoded signals
This seven kinds of signals, wherein every kind of signal generates 3000 samples from -10dB to 2dB every 2dB signal-to-noise ratio；
2) data set signal is pre-processed:
2a) 1) signal for generating data set is first exported in the form of sequence, then by data set signal progress time-frequency conversion, and with
The form of signal time-frequency figure exports；
Classification belonging to signal 2b) is marked in the data set of two kinds of forms output, extracts 2400 out at random from every class signal
Sample is as training sample, and 600 samples are as test sample；
3) RNN-DenseNet network is constructed:
RNN network parameter 3a) is set:
Memory network LSTM constructs RNN-DenseNet network to the length selected in Recognition with Recurrent Neural Network RNN in short-term, i.e., by LSTM net
Network cellular number in network is set as 2, will forget parameter and is set as 1, sets logarithm damage for the loss function of LSTM network
Function is lost, carrys out Schistosomiasis control rate using AdamOptimizer algorithm, uses index linear unit function as activation primitive；
DenseNet network parameter 3b) is set:
If DenseNet network is made of 3 intensive blocks, the convolution operation and 2*2 of 1*1 are successively carried out between adjacent intensive block
Average pondization operation, and 32 are set by the growth rate of the DenseNet network；
3c) the last layer feature of RNN network and DenseNet network is merged, and using softmax classifier as net
The output layer of network；
4) training RNN-DenseNet network:
The number of iterations that RNN-DenseNet network 4a) is arranged is 3000, and learning rate is set as 0.01；
The training sample data of signal sequence 4b) are input to 3a) in the LSTM network that sets, by the training of signal time-frequency figure
Sample data inputs 3b) in the DenseNet network that sets, when the number of iterations reaches 3000, terminates training, trained
Good network model；
5) data of test set are input in trained RNN-DenseNet network, export the prediction class of each test signal
Not.
2. parameter setting is such as the method according to claim 1, wherein 7 kinds of different radar signals in 1)
Under:
The sample frequency of 7 kinds of radar signals is disposed as 2GHz, and sampling number is disposed as 512；
This 5 kinds of thunders of general pulse signal, linear FM signal, NLFM signal, Coded Signals, four phase encoded signals
Carrier frequency configuration up to signal is 150MHz；
Coded Signals use 13 Barker codes；
Four phase encoded signal signals use 16 Frank codes；
Two carrier frequency of two frequency encoded signals are respectively set to 200MHz and 400MHz, and use 13 Barker codes；
Four carrier frequency of four frequency encoded signals are respectively set to 100MHz, 300MHz, 500MHz and 700MHz.
3. according to the method described in claim 1, it is characterized in that 3a) in LSTM network cellular, be expressed as follows:
ht=σ (Wo·[ht-1,xt]+bo)*tanh(ft*Ct-1+Rt)
In formula, t represents time, htFor the output of t moment LSTM network cellular, σ is Sigmoid activation primitive, WOIndicate out gate
Weight, ht-1Indicate the output of a cellular, xtIndicate the input of t moment LSTM network cellular, boIndicate the inclined of out gate
It sets, ftIndicate the output that door is forgotten in t moment LSTM network cellular, Ct-1Indicate the cellular state at t-1 moment, RtIndicate t moment
Tanh operation is done in the output of Memory-Gate in LSTM network cellular, tanh expression.
4. according to the method described in claim 1, it is characterized in that 3a) in index linear unit function, be expressed as follows:
In formula, x indicates input value, and α is a constant coefficient, and f (x) indicates the output of index linear unit function.
5. according to the method described in claim 1, it is characterized in that 3b) in intensive block, be expressed as follows:
xl=Hl([x0,x1,...xl-1]),
In formula, l indicates the number of plies of intensive block, and H indicates a nonlinear transformation, HlIndicate to l layers of intensive block do one it is non-linear
Transformation, [x0,x1,...,xl-1] indicate x0To xl-1Output characteristic pattern do and link, xlIt is the output of intensive block.
6. according to the method described in claim 1, it is characterized in that 3c) in softmax classifier, be expressed as follows:
In formula, yiIndicate the scoring vector of i-th of element,Expression seeks index to the scoring vector of i-th of element,Table
Show the index summation to all elements scoring vector, LiIndicate the score value of i-th of element output.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811554579.0A CN109726653A (en) | 2018-12-19 | 2018-12-19 | Radar Signal Recognition method based on RNN-DenseNet network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811554579.0A CN109726653A (en) | 2018-12-19 | 2018-12-19 | Radar Signal Recognition method based on RNN-DenseNet network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109726653A true CN109726653A (en) | 2019-05-07 |
Family
ID=66296825
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811554579.0A Pending CN109726653A (en) | 2018-12-19 | 2018-12-19 | Radar Signal Recognition method based on RNN-DenseNet network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109726653A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110954872A (en) * | 2019-12-17 | 2020-04-03 | 西安电子科技大学 | Multi-layer perceptron MLP-based phased array radar working mode identification method |
| CN111062322A (en) * | 2019-12-17 | 2020-04-24 | 西安电子科技大学 | Phased array radar behavior recognition method based on Support Vector Machine (SVM) |
| CN111580058A (en) * | 2020-04-02 | 2020-08-25 | 杭州电子科技大学 | Radar HRRP target identification method based on multi-scale convolution neural network |
| CN111915437A (en) * | 2020-06-30 | 2020-11-10 | 深圳前海微众银行股份有限公司 | RNN-based anti-money laundering model training method, device, equipment and medium |
| CN111967309A (en) * | 2020-07-03 | 2020-11-20 | 西安电子科技大学 | Intelligent cooperative identification method and system for electromagnetic signals |
| CN113671493A (en) * | 2021-08-09 | 2021-11-19 | 黑龙江工程学院 | Sea surface small target detection method and system based on feature fusion |
| CN113962261A (en) * | 2021-10-21 | 2022-01-21 | 中国人民解放军空军航空大学 | Depth network model for radar signal sorting |
| CN114509736A (en) * | 2022-01-19 | 2022-05-17 | 电子科技大学 | Radar target identification method based on ultra-wideband electromagnetic scattering characteristics |
-
2018
- 2018-12-19 CN CN201811554579.0A patent/CN109726653A/en active Pending
Non-Patent Citations (3)
| Title |
|---|
| PHILIPPE WEINZAEPFEL ET AL.: "Learning to Track for Spatio-Temporal Action Localization", 《2015 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV)》 * |
| 井博军: "基于深度学习的雷达辐射源识别技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 李达 等: "基于密集卷积网络（DenseNets）的遥感图像分类研究", 《计算机时代》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110954872B (en) * | 2019-12-17 | 2023-03-31 | 西安电子科技大学 | Multi-layer perceptron MLP-based phased array radar working mode identification method |
| CN111062322A (en) * | 2019-12-17 | 2020-04-24 | 西安电子科技大学 | Phased array radar behavior recognition method based on Support Vector Machine (SVM) |
| CN110954872A (en) * | 2019-12-17 | 2020-04-03 | 西安电子科技大学 | Multi-layer perceptron MLP-based phased array radar working mode identification method |
| CN111580058A (en) * | 2020-04-02 | 2020-08-25 | 杭州电子科技大学 | Radar HRRP target identification method based on multi-scale convolution neural network |
| CN111915437A (en) * | 2020-06-30 | 2020-11-10 | 深圳前海微众银行股份有限公司 | RNN-based anti-money laundering model training method, device, equipment and medium |
| WO2021037280A3 (en) * | 2020-06-30 | 2021-05-27 | 深圳前海微众银行股份有限公司 | Rnn-based anti-money laundering model training method, apparatus and device, and medium |
| CN111915437B (en) * | 2020-06-30 | 2024-06-07 | 深圳前海微众银行股份有限公司 | Training method, device, equipment and medium of money backwashing model based on RNN |
| CN111967309A (en) * | 2020-07-03 | 2020-11-20 | 西安电子科技大学 | Intelligent cooperative identification method and system for electromagnetic signals |
| CN111967309B (en) * | 2020-07-03 | 2024-02-06 | 西安电子科技大学 | Intelligent cooperative identification method and system for electromagnetic signals |
| CN113671493A (en) * | 2021-08-09 | 2021-11-19 | 黑龙江工程学院 | Sea surface small target detection method and system based on feature fusion |
| CN113671493B (en) * | 2021-08-09 | 2023-08-11 | 黑龙江工程学院 | Sea surface small target detection method and system based on feature fusion |
| CN113962261A (en) * | 2021-10-21 | 2022-01-21 | 中国人民解放军空军航空大学 | Depth network model for radar signal sorting |
| CN113962261B (en) * | 2021-10-21 | 2024-05-14 | 中国人民解放军空军航空大学 | Deep network model construction method for radar signal sorting |
| CN114509736A (en) * | 2022-01-19 | 2022-05-17 | 电子科技大学 | Radar target identification method based on ultra-wideband electromagnetic scattering characteristics |
| CN114509736B (en) * | 2022-01-19 | 2023-08-15 | 电子科技大学 | Radar target identification method based on ultra-wide band electromagnetic scattering characteristics |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109726653A (en) | Radar Signal Recognition method based on RNN-DenseNet network | |
| CN108182450A (en) | A kind of airborne Ground Penetrating Radar target identification method based on depth convolutional network | |
| CN105975931B (en) | A kind of convolutional neural networks face identification method based on multiple dimensioned pond | |
| CN110378205A (en) | A kind of Complex Radar Radar recognition algorithm based on modified CNN network | |
| CN112308008B (en) | Radar radiation source individual identification method based on working mode open set of transfer learning | |
| CN110120926A (en) | Modulation mode of communication signal recognition methods based on evolution BP neural network | |
| CN107229904A (en) | A kind of object detection and recognition method based on deep learning | |
| CN108509911A (en) | Interference signal recognition methods based on convolutional neural networks | |
| CN110133599A (en) | Intelligent radar emitter Signals classification method based on long memory models in short-term | |
| CN109033780B (en) | A kind of edge calculations access authentication method based on wavelet transformation and neural network | |
| CN108282262B (en) | Intelligent clock signal classification method based on gating cycle unit depth network | |
| CN106529570B (en) | Image classification method based on depth ridge ripple neural network | |
| CN108446766A (en) | A kind of method of quick trained storehouse own coding deep neural network | |
| CN113156376B (en) | SACNN-based radar radiation source signal identification method | |
| Zhang et al. | VGM-RNN: HRRP sequence extrapolation and recognition based on a novel optimized RNN | |
| CN110210330A (en) | Electromagnetic signal recognition methods and device based on Tacit Knowledge structure figures convolutional network | |
| CN107516128A (en) | A kind of flowers recognition methods of the convolutional neural networks based on ReLU activation primitives | |
| CN113761805B (en) | Controllable source electromagnetic data denoising method, system, terminal and readable storage medium based on time domain convolution network | |
| CN110187321A (en) | Radar emitter characteristic parameter extraction method under complex environment based on deep learning | |
| Yang et al. | Thinning of convolutional neural network with mixed pruning | |
| CN108182316A (en) | A kind of Electromagnetic Simulation method and its electromagnetism brain based on artificial intelligence | |
| CN107491793A (en) | A kind of Classification of Polarimetric SAR Image method based on the sparse full convolution of scattering | |
| CN108549832A (en) | LPI radar signal sorting technique based on full Connection Neural Network | |
| Khafaga et al. | Optimized weighted ensemble using dipper throated optimization algorithm in metamaterial antenna | |
| Park et al. | Accuracy enhancement of hand gesture recognition using CNN |
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: 20190507 |