CN108805039A - The Modulation Identification method of combination entropy and pre-training CNN extraction time-frequency image features - Google Patents
The Modulation Identification method of combination entropy and pre-training CNN extraction time-frequency image features Download PDFInfo
- Publication number
- CN108805039A CN108805039A CN201810506714.8A CN201810506714A CN108805039A CN 108805039 A CN108805039 A CN 108805039A CN 201810506714 A CN201810506714 A CN 201810506714A CN 108805039 A CN108805039 A CN 108805039A
- Authority
- CN
- China
- Prior art keywords
- time
- frequency
- image
- frequency image
- radar signal
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- 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
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to radar emitter signal Modulation identification technology fields, and in particular to the Modulation Identification method of combination entropy and pre-training CNN extraction time-frequency image features.9 class radar signal collection to be identified are subjected to time-frequency conversion first and obtain time-frequency image；It is then based on the pre-training convolutional neural networks model imagenet-vgg-verydeep-19 of the official websites MatConvNet offer, constituting FT-VGGNet-fc6 features by its Input input layer to the full articulamentums of fc6 migrates extraction module；Then image is sent into feature and migrates extraction module after adjusting, and exports radar signal time-frequency image feature；Gray processing is carried out to image after adjustment again, the Renyi entropys of image after artificial extraction process；Next training set and test set are divided according to a certain percentage, and choose training set and SVM classifier is trained；Finally, the training set of time-frequency image is identified using the SVM classifier after training, the discrimination of FT-VGGNET-fc6-SVM graders is verified using the data set that 9 class radar signals under more signal-to-noise ratio form.
Description
Technical field
The invention belongs to radar emitter signal Modulation identification technology fields, and in particular to combination entropy is carried with pre-training CNN
Take the Modulation Identification method of time-frequency image feature.
Background technology
Radar emitter signal Modulation Identification is the important link in electronic countermeasure and electronic reconnaissance, is had in electronic warfare
Highly important status and effect.There is the letter based on five kinds of parameters using more universal radar emitter signal Modulation Identification method
Number Modulation Identification method, the intra-pulse modulation recognition methods based on time-frequency image and the signal modulate method based on CNN.
It is traditional based on five kinds of parameters (carrier frequency, pulse arrival time, impulse amplitude, pulse width and pulse arrival direction)
Modulation Identification method in other characteristic parameters that can get pulse after repeatedly measuring, but do not account for new system radar
Intrapulse modulation characteristics, can not achieve effective identification.Meanwhile the method excessively relies on original information in database, in library not
Existing radar cannot be identified and be unable to self study, therefore produce little effect in face of new system radar.
Intra-pulse modulation recognition methods based on time-frequency image is widely used, when it can pass through radar signal
Frequency division cloth is converted to time-frequency image, to extracting feature after image preprocessing, accesses support vector machines (SVM) in its back-end and classifies
Device can obtain more considerable discrimination under low signal-to-noise ratio.But the method needs artificial extraction characteristics of image, people
It is slow that work extracts characteristic velocity.If identification will be caused deviation occur in addition, manually extraction characteristic information is improper, its knowledge is eventually led to
Not rate is low.
Compared to above two method, the signal modulate method based on CNN is adjusted to radar emitter signal
When system identification, CNN can be automatically performed the extraction of time-frequency image feature, realize the increasingly automated of feature extraction.But it is sharp
Output characteristic quantity is excessive after extracting feature with CNN, and redundancy occur can cause the validity of system to decrease, and it is extracted
Feature under Low SNR recognition effect it is poor, corresponding system anti-noise ability is weak.Meanwhile when in face of Small Sample Database,
CNN connects interlayer neural unit quantity and drops suddenly entirely can lead to system recognition rate entire lowering.
Invention content
The purpose of the present invention is to propose to the Modulation Identification sides that a kind of combination entropy and pre-training CNN extract time-frequency image feature
Method automatically extracts characteristics of image to realize extraction feature automation, using principal component analysis (PCA) to exporting feature using CNN
Dimensionality reduction improves system effectiveness, and feature after dimensionality reduction is improved in conjunction with the Renyi entropys manually extracted under system Low SNR
Discrimination, and solves the problems, such as that deep layer network small sample training precision is not high using SVM, essence of the final realization to radar signal
Really quickly identification.
The object of the present invention is achieved like this：
The Modulation Identification method of combination entropy and pre-training CNN extraction time-frequency image features, which is characterized in that including following step
Suddenly：
Step 1 according to nine class radar signal parameters, generate by CW, LFM, BPSK, COSTAS, FRANK, P1, P2, P3,
The radar signal collection of P4 compositions；The nine class radar signal parameters include：Sample frequency, sampling number, CW modulation systems carry
Frequently the carrier frequency and FRANK, P1, P2, modulation that, Barker code, bandwidth, frequency sequence, initial frequency, P3 and P4 modulation systems generate
The carrier frequency that mode generates；
Step 2 carries out time frequency analysis to radar signal, and radar signal collection to be identified is carried out time-frequency conversion using CWD
Obtain time-frequency image；Corresponding CWD formula are as follows：
The local correlation function of time-varying can be obtained by pair correlation function as slide window processing：
It is without restriction when window function takes time impulse function, and take instantaneous value in time domain：
Clock synchronization becomes local correlation function and makees Fourier transformation, you can obtains Wigner Ville distributions (WVD)：
WVD can be by adding kernel functionObtain CWD：
The imagenet-vgg- that the official websites step 3 pre-training convolutional neural networks model selection MatConvNet provide
Picture is converted to the image of 224 × 224 × 3 sizes by verydeep-19；Make imagenet-vgg-verydeep-19 networks
Parameter remains unchanged, and constituting FT-VGGNet-fc6 features by its Input input layer to the full articulamentums of fc6 migrates extraction module；
Image set after step 4 adjusts step 3 is sent into feature migration extraction module, generates radar signal time-frequency figure
As feature, it can obtain 8 × 512 radars using the FT-VGGNet-fc6 features migration extraction module retained to the full articulamentums of fc6 and believe
Number time-frequency image feature；
The feature that step 5 FT-VGGNet-fc6 features migrate extraction module output is excessive, has redundancy and leads
It causes training speed and system effectiveness to decline, retains 95 features with notable discrimination with PCA dimensionality reductions；
Time-frequency image is constituted a data matrix by PCA：
Its covariance matrix is R=XXT, Eigenvalues Decomposition can be made to the covariance matrix：
RM×M=U ∧ UT
In formula, T indicates that transposition, ∧ are the characteristic value diagonal matrix of covariance matrix, and U is corresponding eigenmatrix, to time-frequency
Image makees such as down conversion：
PM×N=UTX=[p1,p2,…,pM]T
In formula, P is the principal component of time-frequency image two values matrix, p1It is first principal component, pjFor jth principal component, k before choosing
A principal component constitutes the eigenmatrix of time-frequency image；
Step 6 carries out image gray processing, the color image pixel of three-component R, G and B to the time-frequency image after adjustment
The brightness of the corresponding point can be calculated as with gray processing formula：
I=0.3B+0.59G+0.11R
The Renyi of discrimination under system recognition rate especially Low SNR can be improved in step 7 extraction gray level image
Entropy, and be allowed to constitute 100 new radar signal time-frequency image features with common normalized combine of 95 features after dimensionality reduction；When
The Renyi entropys of frequency image are expressed as：
In formula, Pα(t, f) indicates the time-frequency distributions of signal；Selection for Renyi entropy exponent numbers α, does not consider non-integral order α
The case where generating the entropy of plural number chooses identification feature of the Renyi entropys that exponent number is 3,5,7,9,11 as signal；
Step 8 is directed to nine kinds of radar signals, chooses every class radar signal time-frequency image each 300 when signal-to-noise ratio is 0dB
, radar signal time-frequency image feature is pressed 7:3 ratio is randomly divided into training set and test set；
Step 9 divides training set and test set, and particle cluster algorithm (PSO) and artificial bee colony algorithm (ABC) are combined, connection
Close and calculate adaptive value using two algorithms, per it is independent calculate once relatively take it is optimal, until arrival maximum iteration takes optimal value,
It realizes that two algorithms are combined to SVM parameter optimizations, and chooses training set and SVM classifier is trained；
The search behavior of particle cluster algorithm PSO particles is influenced by the search behavior of other particles in group and is easily absorbed in office
Portion's optimal solution, and artificial bee colony algorithm ABC local search abilities are weaker and convergence rate is slow；ABC algorithms can overcome the disadvantages that PSO is calculated
Method is absorbed in local solution, and the combination of the two improves the optimizing ability of PSO algorithms, and correlation formula is as follows：
Particle position and speed is updated in population：
Equation is searched in artificial bee colony algorithm is：
vij=xij+φij(xij-xkj)
It first has to (C, the σ) of population scale, PSO velocity intervals, the nectar sources ABC number and parameter, ABC algorithms and PSO algorithms
And initial adaptive value is initialized；Then 2 sub- populations of independence are divided into and use in conjunction ABC algorithms and PSO algorithms calculate
Adaptive value once relatively takes optimal per independent calculating；Adaptive optimal control value when the two finally to reach to maximum iteration as
SVM optimized parameters realize the parameter optimization combined to SVM using two algorithms；
Training set feature combination respective classes label is inputted into above-mentioned SVM parameter optimizations algorithm and carries out cross validation, is determined
The σ and C of SVM is trained SVM using training set according to optimal parameter；
Step 10：9 class radar signal time-frequency images totally 32400 pictures in the case of signal-to-noise ratio is -3~8dB are chosen,
In, per each 300 of time-frequency image under the single signal-to-noise ratio of class radar signal；From every single signal-to-noise ratio of class radar signal of data set
Select 210 at random down, totally 22680 are used as training set, data set per class radar signal single signal-to-noise ratio under remaining 90, altogether
9720 are used as test set；The training set of time-frequency image is identified using the SVM classifier after training, and verifies FT-
The discrimination of VGGNET-fc6-SVM graders.
Compared with the prior art, the advantages of the present invention are as follows：
1, it proposes and Renyi entropys and CNN is extracted into time-frequency image characteristic binding, access SVM and combine transfer learning theoretical
Identification is modulated to radar signal, overcomes CNN and the respective limitations of SVM, compared to its identification speed of current recognizer
Degree is faster, discrimination higher (especially under low signal-to-noise ratio), validity is more preferable and identifying system robustness higher；
2, CNN can obtain more representative information as deep layer network, to make Modulation Identification feature have more comprehensively
Effect, characteristics of image is automatically extracted using CNN, solves the problems, such as manually to extract characteristic velocity slowly and extraction feature is incomplete, and
System stability can be improved；
3, the FT-VGGNet-fc6 feature transferring modules of image input structure, CNN extraction output feature magnitudes are big, application
PCA dimensionality reductions can effectively find out most important 95 features, remove redundancy and invalid information, keep the validity of identifying system remote
It is remote to surmount the simple system for applying CNN Modulation Identifications；
4, the strong anti-noise ability that Renyi entropys have entropy feature publicly-owned, discrimination, which is higher than, especially under low signal-to-noise ratio is based on it
Discrimination obtained by his feature is returned jointly so manually extracting Renyi entropys in pretreatment image again after CNN extracts Feature Dimension Reduction
One changes the anti-noise ability for making the two combination that can improve identifying system；
5, the characteristics of needing very small amount of training sample to can be obtained by category patterns using SVM, effectively prevents rolling up
Product neural network causes to connect the problem of interlayer neural unit quantity drops suddenly entirely since sample class number is very few, and applies SVM energy
System Generalization Capability is set effectively to be promoted with robustness；
6, the FT-VGGNET-fc6-SVM set can apply SVM classifier training when in face of small sample, and face
When to mass data can fixed character extract parameter carry out feature migration so that radar signal data set can utilize large data
Whole network is trained in library, is more there is discrimination and the higher feature of robustness with this.
Description of the drawings
Fig. 1 is imagenet-vgg-verydeep-19 network structures of the present invention；
Fig. 2 is the Modulation Identification flow chart of combination entropy of the present invention and pre-training CNN extraction time-frequency image features；
Fig. 3 is the test result figure of the FT-VGGNET-fc6-SVM graders of the present invention.
Specific implementation mode
The present invention is described further below in conjunction with the accompanying drawings：
9 class radar signal collection to be identified are subjected to Choi-Williams distributions (CWD) time-frequency conversion first and obtain time-frequency
Image；The pre-training convolutional neural networks model imagenet-vgg-verydeep- of the official websites MatConvNet offer is provided
19, time-frequency image size is adjusted to 224 × 224 × 3, the parameter of pre-training network model is made to remain unchanged, it is defeated by its Input
Enter layer to the full articulamentums of fc6 and constitutes FT-VGGNet-fc6 features migration extraction module；Then image after adjustment feature is sent into move
Extraction module is moved, radar signal time-frequency image feature is exported, using PCA to Feature Dimension Reduction to 95 features；Again to scheming after adjustment
As carrying out gray processing, the Renyi entropys of image after artificial extraction process, by dimensionality reduction feature it is common normalized with Renyi entropys after combine
Constitute new radar signal time-frequency image feature；Next training set and test set are divided according to a certain percentage, and population is calculated
Method (PSO) is combined with artificial bee colony algorithm (ABC), and two algorithm of use in conjunction calculates adaptive value, is once relatively taken per independent calculating
It is optimal, optimal value is taken until reaching maximum iteration, realizes that two algorithms are combined to SVM parameter optimizations, and choose training set pair
SVM classifier is trained；Finally, the training set of time-frequency image is identified using the SVM classifier after training, using more
The discrimination for the data set verification FT-VGGNET-fc6-SVM graders that 9 class radar signals form under signal-to-noise ratio.
Specifically include following steps：
Step 1：According to table 1 provide 9 class radar signal parameter values, generate by CW, LFM, BPSK, COSTAS, FRANK,
The radar signal collection of P1, P2, P3, P4 composition.
Table 1
Step 2：Time domain or frequency domain cannot be only confined to the analyzing processing of this kind of non-stationary signal of radar signal,
Time frequency analysis is carried out to radar signal, and CWD effectively can inhibit to intersect while obtaining preferably time, frequency resolution
, radar signal collection to be identified is subjected to time-frequency conversion using CWD and obtains time-frequency image, corresponding CWD formula are as follows：
The local correlation function of time-varying can be obtained by pair correlation function as slide window processing：
It is without restriction when window function takes time impulse function, and take instantaneous value in time domain：
Clock synchronization becomes local correlation function and makees Fourier transformation, you can obtains Wigner Ville distributions (WVD)：
WVD can be by adding kernel functionObtain CWD：
Step 3：The imagenet-vgg- that the official websites pre-training convolutional neural networks model selection MatConvNet provide
Verydeep-19, since imagenet-vgg-verydeep-19 models need three primary colors (RGB) figure of input fixed size
Picture, it is therefore desirable to which picture is converted to the image of 224 × 224 × 3 sizes.Imagenet-vgg-verydeep-19 networks are made to join
Number remains unchanged, and constituting FT-VGGNet-fc6 features by its Input input layer to the full articulamentums of fc6 migrates extraction module.
Step 4：Image set after step 3 is adjusted is sent into feature and migrates extraction module, generates radar signal time-frequency figure
As feature, can be obtained using the FT-VGGNet-fc6 features migration extraction module retained to the full articulamentums of fc6
8 × 512 radar signal time-frequency image features.
Step 5：The feature that FT-VGGNet-fc6 features migrate extraction module output is excessive, has redundancy and leads
It causes training speed and system effectiveness to decline, PCA dimensionality reductions can be used to retain 95 features with notable discrimination.Specifically,
Time-frequency image is constituted a data matrix by PCA：
Its covariance matrix is R=XXT, Eigenvalues Decomposition can be made to the covariance matrix：
RM×M=U ∧ UT (6)
In formula, T indicates that transposition, ∧ are the characteristic value diagonal matrix of covariance matrix, and U is corresponding eigenmatrix, to time-frequency
Image makees such as down conversion：
PM×N=UTX=[p1,p2,…,pM]T (7)
In formula, P is the principal component of time-frequency image two values matrix, p1It is first principal component, pjFor jth principal component, k before choosing
A principal component constitutes the eigenmatrix of time-frequency image.
Step 6：Image gray processing, the color image pixel of three-component R, G and B are carried out to the time-frequency image after adjustment
The brightness of the corresponding point can be calculated as with gray processing formula：
I=0.3B+0.59G+0.11R (8)
Step 7：The Renyi of discrimination under system recognition rate especially Low SNR can be improved in extraction gray level image
Entropy, and be allowed to constitute 100 new radar signal time-frequency image features with common normalized combine of 95 features after dimensionality reduction.When
The Renyi entropys of frequency image are expressed as：
In formula, Pα(t, f) indicates the time-frequency distributions of signal.Selection for Renyi entropy exponent numbers α, the present embodiment do not consider
Non-integral order α generates the case where entropy of plural number, the main identification for choosing the Renyi entropys that exponent number is 3,5,7,9,11 as signal
Feature.
Step 8：For 9 kinds of radar signals, every class radar signal time-frequency image each 300 when signal-to-noise ratio is 0dB is chosen
, radar signal time-frequency image feature is pressed 7:3 ratio is randomly divided into training set and test set.
Step 9：The search behavior of particle cluster algorithm (PSO) particle is influenced by the search behavior of other particles in group
And it is easily absorbed in locally optimal solution, and artificial bee colony algorithm (ABC) local search ability is weaker and convergence rate is slow.ABC is calculated
Method can overcome the disadvantages that PSO algorithms are absorbed in local solution, and the combination of the two can improve the optimizing ability of PSO algorithms, and correlation formula is as follows：
Particle position and speed is updated in population：
Equation is searched in artificial bee colony algorithm is：
vij=xij+φij(xij-xkj) (11)
It first has to (C, the σ) of population scale, PSO velocity intervals, the nectar sources ABC number and parameter, ABC algorithms and PSO algorithms
And initial adaptive value is initialized；Then 2 sub- populations of independence are divided into and use in conjunction ABC algorithms and PSO algorithms calculate
Adaptive value once relatively takes optimal per independent calculating；Adaptive optimal control value when the two finally to reach to maximum iteration as
SVM optimized parameters realize the parameter optimization combined to SVM using two algorithms.
Training set feature combination respective classes label is inputted into above-mentioned SVM parameter optimizations algorithm and carries out cross validation, is determined
The σ and C of SVM is trained SVM using training set according to optimal parameter.
Step 10：9 class radar signal time-frequency images totally 32400 pictures in the case of signal-to-noise ratio is -3~8dB are chosen,
In, per each 300 of time-frequency image under the single signal-to-noise ratio of class radar signal.From every single signal-to-noise ratio of class radar signal of data set
Select 210 at random down, totally 22680 are used as training set, data set per class radar signal single signal-to-noise ratio under remaining 90, altogether
9720 are used as test set.The training set of time-frequency image is identified using the SVM classifier after training, and verifies FT-
The discrimination of VGGNET-fc6-SVM graders.
Fig. 3 provides the test result of FT-VGGNet-fc6-SVM graders, and discrimination is presented with the increase of signal-to-noise ratio
Rise trend, to 5dB after it is smooth-out.87% discrimination is achieved in -3dB, is in later the trend that grows steadily, and in 7dB
When reach 99% or more, show the present invention it is higher (especially under low signal-to-noise ratio) to radar emitter signal Modulation Identification rate.This
Outside, significantly fluctuating do not occur in stable testing of the present invention, discrimination, and it is preferable with robustness to show stability of the present invention, tool
There is good practicability.
The present invention provides a kind of Modulation Identification methods that combination entropy and pre-training CNN extract time-frequency image feature, specifically
Realize that there are many method and the approach of the technical solution, the above is only a preferred embodiment of the present invention.In the present embodiment not
The available prior art of specific each component part is realized.
Claims (1)
1. the Modulation Identification method of combination entropy and pre-training CNN extraction time-frequency image features, which is characterized in that including following step
Suddenly：
Step 1 is made of according to nine class radar signal parameters, generation CW, LFM, BPSK, COSTAS, FRANK, P1, P2, P3, P4
Radar signal collection；The nine class radar signal parameters include：Sample frequency, sampling number, CW modulation systems carrier frequency, Bark
Code, bandwidth, frequency sequence, initial frequency, the carrier frequency of P3 and P4 modulation systems generation and FRANK, P1, P2, modulation system production
Raw carrier frequency；
Step 2 carries out time frequency analysis to radar signal, and radar signal collection to be identified, which is carried out time-frequency conversion, using CWD obtains
Time-frequency image；Corresponding CWD formula are as follows：
The local correlation function of time-varying is obtained by pair correlation function as slide window processing：
It is without restriction when window function takes time impulse function, and take instantaneous value in time domain：
Clock synchronization becomes local correlation function and makees Fourier transformation, obtains Wigner Ville distributions (WVD)：
WVD is by adding kernel functionObtain CWD：
The imagenet-vgg- that the official websites step 3 pre-training convolutional neural networks model selection MatConvNet provide
Picture is converted to the image of 224 × 224 × 3 sizes by verydeep-19；Make imagenet-vgg-verydeep-19 networks
Parameter remains unchanged, and constituting FT-VGGNet-fc6 features by its Input input layer to the full articulamentums of fc6 migrates extraction module；
Image set after step 4 adjusts step 3 is sent into feature migration extraction module, and it is special to generate radar signal time-frequency image
Sign obtains 8 × 512 radar signal time-frequencies using the FT-VGGNet-fc6 features migration extraction module retained to the full articulamentums of fc6
Characteristics of image；
The feature that step 5 FT-VGGNet-fc6 features migrate extraction module output is excessive, has redundancy and occurs causing to instruct
Practice speed and system effectiveness declines, retains 95 features with notable discrimination with PCA dimensionality reductions；
Time-frequency image is constituted a data matrix by PCA：
Its covariance matrix is R=XXT, Eigenvalues Decomposition is made to the covariance matrix：
RM×M=U ∧ UT
In formula, T indicates that transposition, ∧ are the characteristic value diagonal matrix of covariance matrix, and U is corresponding eigenmatrix, to time-frequency image
Make such as down conversion：
PM×N=UTX=[p1,p2,…,pM]T
In formula, P is the principal component of time-frequency image two values matrix, p1It is first principal component, pjFor jth principal component, k master before choosing
Ingredient constitutes the eigenmatrix of time-frequency image；
Step 6 carries out image gray processing to the time-frequency image after adjustment, and the color image pixel of three-component R, G and B correspond to should
The brightness of point is calculated as with gray processing formula：
I=0.3B+0.59G+0.11R；
The Renyi entropys of discrimination under system recognition rate especially Low SNR can be improved in step 7 extraction gray level image, and
It is allowed to constitute 100 new radar signal time-frequency image features with common normalized combine of 95 features after dimensionality reduction；Time-frequency figure
The Renyi entropys of picture are expressed as：
In formula, Pα(t, f) indicates the time-frequency distributions of signal；Selection for Renyi entropy exponent numbers α does not consider that non-integral order α is generated
The case where entropy of plural number, chooses identification feature of the Renyi entropys that exponent number is 3,5,7,9,11 as signal；
Step 8 is directed to nine kinds of radar signals, chooses every class radar signal each 300 of time-frequency image when signal-to-noise ratio is 0dB, will
Radar signal time-frequency image feature presses 7:3 ratio is randomly divided into training set and test set；
Step 9 divides training set and test set, and particle cluster algorithm (PSO) and artificial bee colony algorithm (ABC) are combined, combines and answers
Adaptive value is calculated with two algorithms, once relatively takes optimal per independent calculate, takes optimal value until reaching maximum iteration, realization
Two algorithms are combined to SVM parameter optimizations, and are chosen training set and be trained to SVM classifier；
The search behavior of particle cluster algorithm PSO particles is influenced by the search behavior of other particles in group and is easily absorbed in part most
Excellent solution, and artificial bee colony algorithm ABC local search abilities are weaker and convergence rate is slow；ABC algorithms can overcome the disadvantages that PSO algorithms are fallen into
Enter local solution, the combination of the two improves the optimizing ability of PSO algorithms, and correlation formula is as follows：
Particle position and speed is updated in population：
Equation is searched in artificial bee colony algorithm is：
vij=xij+φij(xij-xkj)
First having to will be (C, the σ) of population scale, PSO velocity intervals, the nectar sources ABC number and parameter, ABC algorithms and PSO algorithms and first
Beginning adaptive value is initialized；Then 2 sub- populations of independence are divided into and use in conjunction ABC algorithms and PSO algorithms are calculated and adapted to
Value once relatively takes optimal per independent calculating；Adaptive optimal control value when the two finally to reach to maximum iteration as SVM most
Excellent parameter realizes the parameter optimization combined to SVM using two algorithms；
Training set feature combination respective classes label is inputted into above-mentioned SVM parameter optimizations algorithm and carries out cross validation, determines SVM's
σ and C is trained SVM using training set according to optimal parameter；
Step 10：Choose 9 class radar signal time-frequency images totally 32400 pictures in the case of signal-to-noise ratio is -3~8dB, wherein
Per each 300 of time-frequency image under the single signal-to-noise ratio of class radar signal；Under every single signal-to-noise ratio of class radar signal of data set
Select 210 at random, totally 22680 are used as training set, data set per class radar signal single signal-to-noise ratio under remaining 90, altogether
9720 are used as test set；The training set of time-frequency image is identified using the SVM classifier after training, and verifies FT-
The discrimination of VGGNET-fc6-SVM graders.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2018103403394 | 2018-04-17 | ||
| CN201810340339 | 2018-04-17 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108805039A true CN108805039A (en) | 2018-11-13 |
Family
ID=64092840
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810506714.8A Pending CN108805039A (en) | 2018-04-17 | 2018-05-24 | The Modulation Identification method of combination entropy and pre-training CNN extraction time-frequency image features |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108805039A (en) |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109597043A (en) * | 2018-11-16 | 2019-04-09 | 江苏科技大学 | Radar Signal Recognition method based on quantum particle swarm convolutional neural networks |
| CN109871757A (en) * | 2019-01-11 | 2019-06-11 | 中国船舶重工集团公司第七二四研究所 | A kind of radar signal intra-pulse modulation kind identification method based on joint time-frequency feature |
| CN110008899A (en) * | 2019-04-02 | 2019-07-12 | 北京市遥感信息研究所 | A kind of visible remote sensing image candidate target extracts and classification method |
| CN110008900A (en) * | 2019-04-02 | 2019-07-12 | 北京市遥感信息研究所 | A kind of visible remote sensing image candidate target extracting method by region to target |
| CN110048980A (en) * | 2019-04-19 | 2019-07-23 | 中国电子科技集团公司第三十六研究所 | A kind of blind demodulation method of digital communication and device |
| CN110243885A (en) * | 2019-07-09 | 2019-09-17 | 东营智图数据科技有限公司 | A kind of low yield gas well mouth of oil well hydrated comples ion method based on time-frequency characteristics |
| CN110417694A (en) * | 2019-06-07 | 2019-11-05 | 吉林大学 | A kind of modulation mode of communication signal recognition methods |
| CN110532932A (en) * | 2019-08-26 | 2019-12-03 | 哈尔滨工程大学 | A kind of multi -components radar emitter signal intra-pulse modulation mode recognition methods |
| CN110632132A (en) * | 2019-07-09 | 2019-12-31 | 东营智图数据科技有限公司 | High-yield gas-oil well wellhead liquid water content prediction method based on multi-sensor measurement and deep convolutional neural network |
| CN112084906A (en) * | 2020-08-27 | 2020-12-15 | 上海朱光亚战略科技研究院 | Radar pulse signal classification method and device, computer equipment and storage medium |
| CN112115821A (en) * | 2020-09-04 | 2020-12-22 | 西北工业大学 | Multi-signal intelligent modulation mode identification method based on wavelet approximate coefficient entropy |
| CN112308008A (en) * | 2020-11-12 | 2021-02-02 | 电子科技大学 | Radar radiation source individual identification method based on working mode open set of transfer learning |
| CN112733811A (en) * | 2020-09-23 | 2021-04-30 | 青岛科技大学 | Underwater sound signal modulation mode inter-class identification method based on improved dense neural network |
| CN112801003A (en) * | 2021-02-05 | 2021-05-14 | 江苏方天电力技术有限公司 | Unmanned aerial vehicle radiation source modulation pattern recognition method and device |
| CN112950922A (en) * | 2021-01-26 | 2021-06-11 | 浙江得图网络有限公司 | Fixed-point returning method for sharing electric vehicle |
| CN113076925A (en) * | 2021-04-23 | 2021-07-06 | 扬州大学 | M-QAM signal modulation mode identification method based on CNN and ELM |
| CN113376608A (en) * | 2021-05-27 | 2021-09-10 | 哈尔滨工程大学 | Convolutional neural network-based radar radiation source structure identification method |
| CN114330441A (en) * | 2021-12-28 | 2022-04-12 | 厦门大学 | Underwater sound JANUS signal identification method and system based on time-frequency spectrum and transfer learning |
| US11341598B2 (en) | 2020-06-05 | 2022-05-24 | International Business Machines Corporation | Interpretation maps with guaranteed robustness |
| CN115720184A (en) * | 2022-10-08 | 2023-02-28 | 西安电子科技大学 | Small sample signal modulation type identification method based on characteristic distribution |
| US11687777B2 (en) | 2020-08-27 | 2023-06-27 | International Business Machines Corporation | Certifiably robust interpretation |
| CN116684233A (en) * | 2023-06-15 | 2023-09-01 | 哈尔滨工程大学 | Communication signal modulation identification method based on image significance detection |
| CN116776230A (en) * | 2023-08-22 | 2023-09-19 | 北京海格神舟通信科技有限公司 | Method and system for identifying signal based on feature imprinting and feature migration |
| CN117129947A (en) * | 2023-10-26 | 2023-11-28 | 成都金支点科技有限公司 | Planar transformation method radar signal identification method based on mininet |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104809358A (en) * | 2015-05-21 | 2015-07-29 | 哈尔滨工业大学 | Radar radiation source identification method based on phase noise unintentional modulation characteristic |
| CN106778610A (en) * | 2016-12-16 | 2017-05-31 | 哈尔滨工程大学 | A kind of intra-pulse modulation recognition methods based on time-frequency image feature |
| CN107301432A (en) * | 2017-07-11 | 2017-10-27 | 哈尔滨工程大学 | Adaptive radiation source Modulation Identification method based on time frequency analysis |
-
2018
- 2018-05-24 CN CN201810506714.8A patent/CN108805039A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104809358A (en) * | 2015-05-21 | 2015-07-29 | 哈尔滨工业大学 | Radar radiation source identification method based on phase noise unintentional modulation characteristic |
| CN106778610A (en) * | 2016-12-16 | 2017-05-31 | 哈尔滨工程大学 | A kind of intra-pulse modulation recognition methods based on time-frequency image feature |
| CN107301432A (en) * | 2017-07-11 | 2017-10-27 | 哈尔滨工程大学 | Adaptive radiation source Modulation Identification method based on time frequency analysis |
Non-Patent Citations (4)
| Title |
|---|
| FAN HU 等: ""Transferring Deep Convolutional Neural Networks for the Scene Classification of High-Resolution Remote Sensing Imagery"", 《REMOTE SENS》 * |
| 刘康宁: ""基于时频分析的雷达信号调制方式识别技术研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 宁爱平 等: ""ABC-PSO算法优化混合核SVM参数及应用"", 《数学的实践与认识》 * |
| 高敬鹏 等: ""基于时频分析的自适应PCA辐射源调制识别"", 《应用科技》 * |
Cited By (36)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109597043A (en) * | 2018-11-16 | 2019-04-09 | 江苏科技大学 | Radar Signal Recognition method based on quantum particle swarm convolutional neural networks |
| CN109597043B (en) * | 2018-11-16 | 2023-05-26 | 江苏科技大学 | Radar signal identification method based on quantum particle swarm convolutional neural network |
| CN109871757A (en) * | 2019-01-11 | 2019-06-11 | 中国船舶重工集团公司第七二四研究所 | A kind of radar signal intra-pulse modulation kind identification method based on joint time-frequency feature |
| CN110008900B (en) * | 2019-04-02 | 2023-12-12 | 北京市遥感信息研究所 | Method for extracting candidate target from visible light remote sensing image from region to target |
| CN110008899A (en) * | 2019-04-02 | 2019-07-12 | 北京市遥感信息研究所 | A kind of visible remote sensing image candidate target extracts and classification method |
| CN110008900A (en) * | 2019-04-02 | 2019-07-12 | 北京市遥感信息研究所 | A kind of visible remote sensing image candidate target extracting method by region to target |
| CN110008899B (en) * | 2019-04-02 | 2021-02-26 | 北京市遥感信息研究所 | Method for extracting and classifying candidate targets of visible light remote sensing image |
| CN110048980A (en) * | 2019-04-19 | 2019-07-23 | 中国电子科技集团公司第三十六研究所 | A kind of blind demodulation method of digital communication and device |
| CN110417694A (en) * | 2019-06-07 | 2019-11-05 | 吉林大学 | A kind of modulation mode of communication signal recognition methods |
| CN110632132A (en) * | 2019-07-09 | 2019-12-31 | 东营智图数据科技有限公司 | High-yield gas-oil well wellhead liquid water content prediction method based on multi-sensor measurement and deep convolutional neural network |
| CN110243885A (en) * | 2019-07-09 | 2019-09-17 | 东营智图数据科技有限公司 | A kind of low yield gas well mouth of oil well hydrated comples ion method based on time-frequency characteristics |
| CN110532932A (en) * | 2019-08-26 | 2019-12-03 | 哈尔滨工程大学 | A kind of multi -components radar emitter signal intra-pulse modulation mode recognition methods |
| CN110532932B (en) * | 2019-08-26 | 2022-12-13 | 哈尔滨工程大学 | Method for identifying multi-component radar signal intra-pulse modulation mode |
| US11341598B2 (en) | 2020-06-05 | 2022-05-24 | International Business Machines Corporation | Interpretation maps with guaranteed robustness |
| CN112084906A (en) * | 2020-08-27 | 2020-12-15 | 上海朱光亚战略科技研究院 | Radar pulse signal classification method and device, computer equipment and storage medium |
| US11687777B2 (en) | 2020-08-27 | 2023-06-27 | International Business Machines Corporation | Certifiably robust interpretation |
| CN112115821B (en) * | 2020-09-04 | 2022-03-11 | 西北工业大学 | Multi-signal intelligent modulation mode identification method based on wavelet approximate coefficient entropy |
| CN112115821A (en) * | 2020-09-04 | 2020-12-22 | 西北工业大学 | Multi-signal intelligent modulation mode identification method based on wavelet approximate coefficient entropy |
| CN112733811A (en) * | 2020-09-23 | 2021-04-30 | 青岛科技大学 | Underwater sound signal modulation mode inter-class identification method based on improved dense neural network |
| CN112308008A (en) * | 2020-11-12 | 2021-02-02 | 电子科技大学 | Radar radiation source individual identification method based on working mode open set of transfer learning |
| CN112950922A (en) * | 2021-01-26 | 2021-06-11 | 浙江得图网络有限公司 | Fixed-point returning method for sharing electric vehicle |
| CN112801003B (en) * | 2021-02-05 | 2022-06-17 | 江苏方天电力技术有限公司 | Unmanned aerial vehicle radiation source modulation pattern recognition method |
| CN112801003A (en) * | 2021-02-05 | 2021-05-14 | 江苏方天电力技术有限公司 | Unmanned aerial vehicle radiation source modulation pattern recognition method and device |
| CN113076925B (en) * | 2021-04-23 | 2023-09-05 | 扬州大学 | M-QAM signal modulation mode identification method based on CNN and ELM |
| CN113076925A (en) * | 2021-04-23 | 2021-07-06 | 扬州大学 | M-QAM signal modulation mode identification method based on CNN and ELM |
| CN113376608A (en) * | 2021-05-27 | 2021-09-10 | 哈尔滨工程大学 | Convolutional neural network-based radar radiation source structure identification method |
| CN114330441A (en) * | 2021-12-28 | 2022-04-12 | 厦门大学 | Underwater sound JANUS signal identification method and system based on time-frequency spectrum and transfer learning |
| CN114330441B (en) * | 2021-12-28 | 2024-09-13 | 厦门大学 | Water sound JANUS signal identification method and system based on time spectrum and transfer learning |
| CN115720184A (en) * | 2022-10-08 | 2023-02-28 | 西安电子科技大学 | Small sample signal modulation type identification method based on characteristic distribution |
| CN115720184B (en) * | 2022-10-08 | 2024-04-19 | 西安电子科技大学 | Small sample signal modulation type identification method based on characteristic distribution |
| CN116684233A (en) * | 2023-06-15 | 2023-09-01 | 哈尔滨工程大学 | Communication signal modulation identification method based on image significance detection |
| CN116684233B (en) * | 2023-06-15 | 2024-10-18 | 哈尔滨工程大学 | Communication signal modulation identification method based on image significance detection |
| CN116776230A (en) * | 2023-08-22 | 2023-09-19 | 北京海格神舟通信科技有限公司 | Method and system for identifying signal based on feature imprinting and feature migration |
| CN116776230B (en) * | 2023-08-22 | 2023-11-14 | 北京海格神舟通信科技有限公司 | Method and system for identifying signal based on feature imprinting and feature migration |
| CN117129947A (en) * | 2023-10-26 | 2023-11-28 | 成都金支点科技有限公司 | Planar transformation method radar signal identification method based on mininet |
| CN117129947B (en) * | 2023-10-26 | 2023-12-26 | 成都金支点科技有限公司 | Planar transformation method radar signal identification method based on mininet |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108805039A (en) | The Modulation Identification method of combination entropy and pre-training CNN extraction time-frequency image features | |
| CN109254274B (en) | Radar radiation source identification method based on feature fusion | |
| CN109446877B (en) | Radar radiation source signal modulation identification method combined with multi-dimensional feature migration fusion | |
| WO2021134871A1 (en) | Forensics method for synthesized face image based on local binary pattern and deep learning | |
| CN109597043B (en) | Radar signal identification method based on quantum particle swarm convolutional neural network | |
| CN110175560A (en) | A kind of radar signal intra-pulse modulation recognition methods | |
| CN108416290A (en) | Radar signal feature method based on residual error deep learning | |
| CN114692741B (en) | Generalized face counterfeiting detection method based on domain invariant features | |
| CN102073880A (en) | Integration method for face recognition by using sparse representation | |
| CN104200237A (en) | High speed automatic multi-target tracking method based on coring relevant filtering | |
| CN110532932A (en) | A kind of multi -components radar emitter signal intra-pulse modulation mode recognition methods | |
| CN110136162B (en) | Unmanned aerial vehicle visual angle remote sensing target tracking method and device | |
| CN107315996A (en) | A kind of noise characteristic extracting method of ships under water based on IMF Energy-Entropies and PCA | |
| CN113420775A (en) | Image classification method under extremely small quantity of training samples based on adaptive subdomain field adaptation of non-linearity | |
| Zhang et al. | Few-shot learning of signal modulation recognition based on attention relation network | |
| CN113111786B (en) | Underwater target identification method based on small sample training diagram convolutional network | |
| CN111010356A (en) | Underwater acoustic communication signal modulation mode identification method based on support vector machine | |
| CN109766791A (en) | A kind of Modulation Recognition of Communication Signal method based on self-encoding encoder | |
| Wei et al. | Universal deep network for steganalysis of color image based on channel representation | |
| CN115314287B (en) | Counter anomaly detection system based on deep clustering | |
| CN112749633A (en) | Separate and reconstructed individual radiation source identification method | |
| CN114972904B (en) | Zero sample knowledge distillation method and system based on fighting against triplet loss | |
| CN113505810A (en) | Pooling vision-based method for detecting weed growth cycle by using Transformer | |
| CN103400370A (en) | Adaptive fuzzy C-means image segmentation method based on potential function | |
| CN103793720B (en) | A kind of eye locating method and 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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181113 |
|
| RJ01 | Rejection of invention patent application after publication |