CN108090412A - A kind of radar emission source category recognition methods based on deep learning - Google Patents
A kind of radar emission source category recognition methods based on deep learning Download PDFInfo
- Publication number
- CN108090412A CN108090412A CN201711145195.9A CN201711145195A CN108090412A CN 108090412 A CN108090412 A CN 108090412A CN 201711145195 A CN201711145195 A CN 201711145195A CN 108090412 A CN108090412 A CN 108090412A
- Authority
- CN
- China
- Prior art keywords
- mrow
- feature
- vector
- radar
- classification
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Signal Processing (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a kind of radar emission source category recognition methods based on deep learning, the section of its ambiguity function is asked for as feature vector by the use of the emitter Signals by pretreatment;Using the feature vector for largely accomplishing fluently label as training sample, it is trained by depth convolutional neural networks, and the Classification and Identification of input feature value is carried out using the convolutional neural networks grader obtained;In order to realize the identification of the radiation source for unknown classification, the meta identifiers based on support vector machines are built to judge whether the classification results of convolutional neural networks grader are credible, obtain final recognition result.This method can effectively improve the accuracy of recognizing radar radiation source.
Description
【Technical field】
The invention belongs to recognizing radar radiation source technical fields, and in particular to a kind of radar emission based on deep learning
Source category recognition methods.
【Background technology】
Modern battlefield situation is fast changing, and effect of the information countermeasure in modern military is more and more important.Electronic warfare is also referred to as
Electronic countermeasure, including three electronic reconnaissance, electronic attack and electronic protection aspects.Electronic reconnaissance refer mainly to from enemy radar and
Its weapon system obtains useful information, can be to radiation source divides between ourselves and the enemy in battlefield surroundings by specific emitter identification
Cloth situation is implemented to scout, and provides more comprehensive, accurate electromagnetism struggle and the situation of weapon, carries out effective battleficld command
With decision-making.Radar recognition has become the research hotspot and difficult point in Current electronic war particularly electronic reconnaissance field.Radiation source is special
Levy unknown, increasingly sophisticated signal waveform, severe wartime electromagnetic environment is brought more and more sternly to the accurate identification of radiation source
High challenge.
In terms of emitter Signals feature mining, in the seventies in last century, foreign countries related researcher has begun to the portion
Divide research, many scholars have done numerous studies work, can be divided into two stages:
First stage is radiation source basic parameter properties study.Its carrier frequency, arteries and veins are directly asked for original signal characteristic
The information such as width, impulse amplitude, angle of arrival and arrival time are rushed, by the use of wherein one or more as feature vector.It is this
Situation is mainly applied to electromagnetic environment is relatively single, radiation source category is less, signal form is single, radar parameter is fixed
In early days.
Second stage since the 1990s, west military power begin one's study radar emission signal arteries and veins in it is special
Sign proposes a variety of methods for analyzing Features of Radar Signal In A Pulse of knowing clearly in succession.Representative method has:Time-domain waveform analysis method,
Instantaneous frequency feature in spectrum correlation method, time-frequency synthesis, wavelet analysis method, information theory criterion and clustering technique synthesis, arteries and veins
With accumulative etc..
The country starts from the beginning of the eighties in last century to the research of Radar emitter individual identification technology, although starting late,
Great attention is received, gives in " 95 ", " 15 " and Eleventh Five-Year Plan and subsidizes energetically.In terms of intrapulse feature excavation,
Bi great Ping proposes instantaneous frequency distilling technology in the arteries and veins for be easy to Project Realization;Zhang Gexiang proposes radar emitter signal
Wavelet packet character, resemblance Coefficient feature, entropy feature, Rough Set, information dimension and fractal box;Zhu Ming, which is proposed, to be based on
The feature of Atomic Decomposition, the feature based on Chirplet atoms, time-frequency atom feature;General fortune is big to propose instantaneous frequency derivation
Feature, ambiguity function backbone section feature;Old rice is big to propose symbolism intrapulse feature, contour integral bispectrum feature etc.;Yu Zhi
It is refined to propose Local Wave Decomposition, Wavelet Ridge frequency cascade nature.
On the other hand, recognizing radar radiation source is a typical classification problem, and main thought is to obtain radiation source
After the character representation of signal, conversion of the feature space to decision space is realized by effective sorting algorithm, so that it is determined that
The generic of signal.Substantial amounts of sorting algorithm is employed in recognizing radar radiation source, such as template matches, neutral net, branch
Hold vector machine etc..Generally be applied to the field there are three types of sorting technique, one kind is differentiation type grader, and needs are being learned
Certain object function is optimized during practising;Another kind is generation model classifiers, is mainly based upon prior probability and classification
Conditional probability density is estimated, such as linear discrimination classification device, K arest neighbors;The third is Decision Tree Algorithm, is passed through
The priori of human expert is classified, such as ID3, C4.5 algorithm.
The shortcomings that prior art, first was that sorting algorithm precision is low mainly at two aspects.It is more to have algorithm
Expert's priori or the feature artificially chosen are relied on, increasingly complicated electromagnetic environment can not be adapted to.Second shortcoming is nothing
Method identifies unknown classification, and existing algorithm can only identify existing classification, and more importantly for unknown point in electronic reconnaissance
The quick identification of class radiation source.
【The content of the invention】
The object of the present invention is to provide a kind of radar emission source category recognition methods based on deep learning, in solution
The defects of stating the prior art.
The present invention uses following technical scheme:A kind of radar emission source category recognition methods based on deep learning utilizes
Emitter Signals by pretreatment ask for the section of its ambiguity function as feature vector;The spy of label will largely be accomplished fluently
Sign vector is trained by depth convolutional neural networks as training sample, and is classified using the convolutional neural networks obtained
Device carries out the Classification and Identification of input feature value;In order to realize the identification of the radiation source for unknown classification, structure is based on support
The meta identifiers of vector machine judge whether the classification results of convolutional neural networks grader are credible, obtain final identification
As a result.
Further, specific implementation method is:
Step 1, emitter Signals pretreatment and feature extraction optimization:Useless and mistake in emitter Signals is rejected first
Data, then emitter Signals are sorted;It is vectorial that the slice of ambiguity function of each radar is extracted again, and as
Feature vector;
Step 2, structure depth convolutional neural networks structure:According to the feature of radiation source slice of ambiguity function, one is devised
A depth is 13 layers of convolutional neural networks, and convolutional layer has selected the convolution kernel of the 1*3 with 128 layers, used length as 1*
2 maximum pond method, and add dropout parameters, by the use of softmax as final output layer;
Step 3, training neural network model:On the basis of traditional gradient optimization algorithm, using first moment with
Second order moments estimation is adjusted learning rate;
Step 4, the identification of convolutional neural networks grader:The model obtained using step 3 carries out Classification and Identification, is somebody's turn to do
Radiation source belongs to the probability of each classification as preliminary output;
Step 5, structure support vector machines meta graders:What is obtained using in step 4 tentatively exports as the grader
Input carries out secondary judgement to classification, determines if it is unknown classification.
Further, the specific method of step 1 is:
Step 1.1 rejects data useless and wrong in emitter Signals, and then emitter Signals are sorted, point
Process is selected mainly to utilize the correlation of same portion's radar emitter signal parameter and the otherness of different radar signal parameters, from
The pulse signal of each radar is isolated in overlapping pulse signal stream and select useful emitter Signals at random;
Step 1.2, the instantaneous auto-correlation function of emitter Signals x (t) are defined as:
Wherein, τ is time delay;
The definition of radar ambiguity function is:
That is RxThe Fourier inversion of (t, τ) on time t, wherein v are frequency deviation, and j is negative unit, and π is pi, e
For natural constant;
Formula (2) can be equivalent to following formula (3) by conversion:
To emitter Signals x (t) uniform samplings, formula (3) becomes formula (4):
Formula (4) is the slice of ambiguity function vector of the radar extracted, wherein, τlWhen for time offset being l
Time delay, vmFrequency deviation when for frequency offset being m, N is the total frequency spectrum cycle, if fsFor working frequency, then have
Further, the specific method of step 2 is:
Step 2.1, convolution feature extraction:
If the sequence that the slice of ambiguity function vector for the emitter Signals that step 1.2 obtains is 1 × K, this is initial characteristics
Vector carries out convolution algorithm to it, obtains first convolutional layer, is represented with C1;Using 128 sizes be 1 × 3 convolution kernel,
Therefore in feature vector each neuron with input in 1 × 3 neighborhood be connected, it is C1 layer such in feature sizes just be 1 ×
(K-3);Again because each wave filter have 3 cell parameters and an offset parameter, altogether 128 wave filters, C1 it is common (1 ×
3+1) × 128=512 can training parameter, totally 512 × (K-3) a connection, will connection by ReLU active coatings, so far complete
Convolution feature extraction for initial characteristics vector;
Step 2.2, maximum pondization processing:
Pond length is used as 2 maximum pond method, it will be through adjacent in step 2.1 treated initial characteristics vector
Two elements in maximum feature as Chi Huahou feature for subsequent operation, realize the dimensionality reduction of feature vector;
Step 2.3, using the feature vector obtained by step 2.2 as new input, add continuous two convolutional layers,
A pond layer is added again, then adds two convolutional layers, then adds a pond layer, so by continuous several times convolution, Chi Hua
Operation, the information being fully extracted in initial characteristics vector obtain a new feature vector;
Step 2.4, structure output layer:
It evens up and handles the new feature vector obtained through step 2.3, in this, as a mistake of convolutional layer to full articulamentum
It crosses, dropout parameters 0.5 is added on the basis of first full articulamentum, then add the second layer and connect entirely, pass through activation
Function Softmax obtains the output layer of classification results.
Further, the specific method of step 3 is:
After neural network model is put up, further training is done to the model using training sample:
Step 3.1, initialization step-length ε=0.001, the expectation rate of disintegration of first moment and second order moments estimation is respectively ρ1=
0.9, ρ2=0.999, initial learning rate θ, first moment and second moment variable s=0, r=0, iterations k=0, greatest iteration time
Number Kmax;
Step 3.2 is obtained from training set corresponding to target y(i)The sampling { x with M sample(1)..., x(M),
Wherein M is the number of a collection of sample in batch processing;
Step 3.3 calculates gradientWherein L is logarithm loss function,
It is defined as being distributed X, Y has L (P (Y | X), Y)=-logP (Y | X), P represents probability;
Step 3.4, update iterations k ← k+1, there is inclined single order moments estimation s ← ρ1s+(1-ρ1) g, there is inclined second moment to estimate
Count r ← ρ2r+(1-ρ2) g ⊙ g, wherein g ⊙ g represent corresponding element product;
Step 3.5 corrects single order moments estimationSecond order moments estimation
Step 3.6, renewal learning rateWherein δ=10-8It is one to be used for keeping
The constant of stability, judges whether iterations k is more than maximum iteration KmaxIf iterations k changes no more than maximum
Generation number KmaxWith regard to repeating step 3.2-3.6, if iterations k is more than maximum iteration KmaxWith regard to stopping iteration.
Further, the specific method of step 4 is:
After the training of step 3, a depth convolutional neural networks grader, the mould of input radiation source signal are obtained
Function section vector is pasted, the output that the radiation source belongs to the probability of each classification can be obtained by the grader.
Further, the specific method of step 5 is:
It is special by the use of probability of all categories as it using the output of depth convolutional neural networks as the input of the grader
Sign is trained identification, chooses the recognition result that the consecutive numbers from same emitter Signals is clapped and carries out an average conduct
Finally enter, the input by support vector machine classifier can finally obtain its whether be unknown classification probability.
The beneficial effects of the invention are as follows:The present invention proposes a kind of method by using deep learning to improve radiation source
The method that classification identifies and tells unknown classification, weather radar radiation source entrained by bound fraction aircraft of the present invention return
Ripple signal analyzes the accuracy of the identification of known classification and the identification of unknown classification, its validity is verified, based on depth
The Radar recognition algorithm of habit can effectively improve the accuracy of recognizing radar radiation source.
【Description of the drawings】
Fig. 1 is general frame figure of the present invention;
Fig. 2 is slice of ambiguity function schematic diagram of the present invention;
Fig. 3 is convolutional neural networks general frame figure of the present invention;
Fig. 4 is the deep neural network structure chart that the present invention is built;
Fig. 5 is slice of ambiguity function convolution characteristic pattern of the present invention;
Fig. 6 is slice of ambiguity function pond of the present invention schematic diagram;
Fig. 7 is convolutional neural networks training flow chart of the present invention;
Fig. 8 is iterations of the present invention and classification accuracy curve;
Fig. 9 is the different classes of number data recognition result of the present invention.
【Specific embodiment】
Below by drawings and examples, technical scheme is described in further detail.
The present invention provides a kind of radar emission source category recognition methods based on deep learning, at Comprehensive Radar signal
The multidisciplinary theory such as reason, deep learning, emphasis are centered around the individual knowledge for having algorithm under complex electromagnetic environment to different radiation sources
Not can force difference the problems such as with challenge, propose rational radar pulse finger-print model, with reference to deep learning theory with side
Method solves the limitation of conventional radiation source discrimination.Therefore, the present invention combines the intrapulse feature of Radar emitter, carries for the first time
The accuracy of identification for going out a kind of method by using deep learning to improve radiation source and the calculation recognized to unknown classification
Method.The overall structure of the present invention is as shown in Figure 1, by deep learning grader and support vector machines meta graders, the two is mutual
It mutually feeds back, realizes for the identification of unknown classification and the identification of known classification.
Specifically implement according to following steps:
Step 1, emitter Signals processing:
Processing for emitter Signals, present invention primarily contemplates two aspects:Signal Pretreatment and feature extraction optimization.
In terms of Signal Pretreatment, it is necessary first to reject useless and wrong data.Then signal is sorted,
Mainly the pulse signal of each radar is isolated from pulse signal stream overlapping at random and select useful signal.Its essence
Be overlap, de-interlacing, what is utilized is the correlation of same radar signals parameter and the difference of different radar signal parameters
Property.
In terms of feature extraction optimization, reasonably it is characterized in the basis of Classification and Identification.The present invention is with the ambiguity function of radar
As input signal, its slice feature is extracted in analysis.
Ambiguity function can not only describe the resolution characteristic and fuzziness of radar signal, moreover it is possible to describe to be determined by radar signal
Fixed measurement accuracy and clutter recognition characteristic etc., this radar are not intended to modulate the signal pulse finger-print generated as classification
Required feature is widely used.In Practical Project, there are phase noises and all kinds of spuious defeated for Radar emitter itself
Go out, even so the signal of the identical radiation emission of model, parameter there are still nuances.By obscuring letter
Number in time delay and frequency deviation two-dimensional transform, can multi-angle depict the influence modulated unintentionally to emitting signal.Fig. 2 represents an allusion quotation
The slice of ambiguity function figure of type.
The instantaneous auto-correlation function of emitter Signals x (t) is:
Wherein, τ is time delay.
The definition of radar ambiguity function is:
That is RxThe Fourier inversion of (t, τ) on time t, wherein v are frequency deviation, and j is narration unit, and π is pi, e
For natural constant.
It uses in a digital signal for convenience, formula (2) can be equivalent to following formula (3) by conversion:
To emitter Signals uniform sampling, i.e., after the docking collection of letters number and reference signal discretization, formula (3) becomes formula
(4):
Formula (4) is the slice of ambiguity function vector of the radar extracted, wherein, τlWhen for time offset being l
Time delay, vmFrequency deviation when for frequency offset being m, N is the total frequency spectrum cycle, if fsFor working frequency, then have
Step 2, structure depth convolutional neural networks structure:
The present invention is referred to Fig. 3 for the global design of convolutional neural networks, and it is big with identification two can be divided into training
Part.In terms of classifier design, the present invention devises a grader using convolutional neural networks.Typical convolution god
The multiple and different layers being stacked through network by deep structure form:Input layer, multigroup convolution and pond layer, limited quantity
The hidden layer connected entirely and output layer.Wherein most important part is convolutional layer, this layer utilizes the part in input data
The entire input space is divided into the hidden unit of very little, and the convolution that the weight of each hidden unit is built by structure
Core acts on the entire input space, so as to obtain feature vector.Using this mechanism, while greatly reducing number of parameters
Improve the translation invariance of data.
The present invention constructs one 13 layers of convolutional neural networks according to emitter Signals characteristic, and basic structure is being schemed
It is provided in 4.Every layer has multiple feature vectors, and each feature vector has multiple neurons, and each feature vector comes from
A kind of feature of input is extracted in a kind of convolution kernel.Main process is to carry out multiple convolution, pond to the emitter Signals of input
Change operation, then feature extraction again is trained by backpropagation (Back Propagation) network.
Step 2.1, convolution feature extraction
If the sequence that the slice of ambiguity function vector for the emitter Signals that step 1.2 obtains is 1 × K, this is initial spy
Sign vector, carries out convolution algorithm to it, obtains first convolutional layer, represented with C1.The present invention is 1 × 3 using 128 sizes
Convolution kernel, therefore in feature vector each neuron with input in 1 × 3 neighborhood be connected, it is C1 layers such in feature sizes
It is just 1 × (K-3).Again because each wave filter has 3 cell parameters and an offset parameter, 128 wave filters are had altogether,
Altogether (1 × 3+1) × 128=512 parameter, i.e. C1 have 512 can training parameter, totally 512 × (K-3) a connection, by connect lead to
ReLU active coatings are crossed, so far complete the convolution feature extraction for feature vector.
Convolutional neural networks primary operational in characteristic extraction procedure is convolution, an input vector x and a power
The convolution operation of weight w is denoted as:
S (t)=(x*w) (t),
Wherein, t is current time.
For the discrete system of the present invention, above formula is equivalent to:
Fig. 5 gives the signal that a discrete vector carries out convolution algorithm.
Step 2.2, maximum pondization processing:
Pondization operation replaces adjacent multiple features using a feature, by reducing the length of feature vector, is subtracting
Over-fitting situation is also had modified to a certain extent while small calculation amount.
In next step classified using these features after feature is obtained by convolution.Theoretically, institute can be used
There is the feature training grader that extraction obtains, but so doing can cause calculation amount excessive, and excessive feature vector is also easy in addition
Cause over-fitting.
Since radiation source slice of ambiguity function data have the attribute of a kind of " nature static ", this is meant that in a number
It is very likely equally applicable in another region according to region useful feature, so the feature after convolution can be used.Therefore, it is
The more data of description data volume, method is exactly to carry out aggregate statistics to the feature of different position, for example, people can
To calculate the maximum or average value of some special characteristic on one region of image.These summary statistics features are compared to use
All obtained features of extracting not only have much lower dimension, while can also improve result, it is not easy to over-fitting.It is this poly-
The operation of conjunction is just called pond, and common pond method has average pondization and maximum pond.
If the successive range in slice of ambiguity function is selected as pond region, and simply pondization is identical (repetition)
The feature that generates of hidden unit, then, these pond units are just with translation invariance.Even if this means that fuzzy letter
After number section experienced a small translation, the feature in identical pond can be still generated.Pond length is used in the present invention
For 2 maximum pond method namely selection maximum feature subsequent operation is used for as the feature of Chi Huahou.It will be through step
After feature maximum in two adjacent elements is used for as the feature of Chi Huahou in rapid 2.1 treated initial characteristics vectors
The dimensionality reduction of feature vector is realized in continuous operation.Fig. 6 gives the signal in the maximum pond that a pond length is 2.
Step 2.3, using the feature vector obtained by step 2.2 as new input, add continuous two convolutional layers,
A pond layer is added again, then adds two convolutional layers, then adds a pond layer, so by continuous several times convolution, Chi Hua
Operation, the information being fully extracted in initial input feature vector obtain a new feature vector.
Step 2.4, structure output layer:
It evens up and handles the new feature vector obtained through step 2.3, in this, as a mistake of convolutional layer to full articulamentum
It crosses.Dropout parameters 0.5 are added on the basis of first full articulamentum, the second layer is then added and connects entirely, pass through activation
Function Softmax obtains the output layer of classification results.
Step 3, training neural network model
After neural network model is put up, further training, whole thinking are done to the model using training sample
As shown in fig. 7, the problem of present invention is set for optimization algorithm learning rate initial value, utilizes first moment and second order moments estimation pair
Learning rate has carried out ART network, and basic step is as follows:
Step 3.1, initialization step-length ε=0.001, the expectation rate of disintegration of first moment and second order moments estimation is respectively ρ1=
0.9, ρ2=0.999, initial learning rate θ, first moment and second moment variable s=0, r=0, iterations k=0, greatest iteration
Number Kmax。
Step 3.2 is obtained from training set corresponding to target y(i)The sampling (x with M sample(1)..., x(M),
Wherein M is the number of a collection of sample in batch processing.
Step 3.3 calculates gradientWherein L is logarithm loss function,
It is defined as being distributed X, Y has L (P (Y | X), Y) and=- logP (Y | X), P represents probability.
Step 3.4, update iterations k ← k+1, there is inclined single order moments estimation s ← ρ1s+(1-ρ1) g, there is inclined second moment to estimate
Count r ← ρ2r+(1-ρ2) g ⊙ g, wherein g ⊙ g represent corresponding element product.
Step 3.5 corrects single order moments estimationSecond order moments estimation
Step 3.6, renewal learning rateWherein δ=10-8It is one to be used for keeping
The constant of stability.Judge whether iterations k is more than maximum iteration KmaxIf iterations k is more than greatest iteration
Number KmaxWith regard to stopping iteration;If iterations k is not more than maximum iteration Kmax, then repeatedly step 3.2-3.6.
Fig. 8 gives the change curve of iterations and recognition accuracy, for different required precisions, Ke Yiju
This selects different iterationses.
Step 4, the identification of convolutional neural networks grader
After the training of step 3, a depth convolutional neural networks grader, input radiation source signal can be obtained
Slice of ambiguity function vector, can obtain the output that the radiation source belongs to the probability of each classification by the grader.
Step 5, structure support vector machines meta graders
Support vector machines is a kind of sorting technique of prevalence, can have been generated in the case where mass data is not required
As a result.The data of all target data and unknown object can be utilized to be used as training sample to the support vector cassification
Device is trained.This part we using the output of depth convolutional neural networks as the input of the grader, using of all categories
Probability is trained identification as its feature.Since in the identification process of classification, there are certain fluctuations, and whether influence
Belong to the resolution of unknown classification, we choose the recognition result that the consecutive numbers from same radiation source is clapped and carry out one averagely
As finally entering, specific umber of beats can carry out test selection according to real data here, such as can select 10 umber of beats
According to.The input by support vector machine classifier can finally obtain its whether be unknown classification probability.
Here is the design of support vector machine classifier.It is the selection of kernel function first.Kernel function maps the input space
To high-dimensional feature space, optimal separating hyperplane is finally constructed in high-dimensional feature space, so as to bad in itself in plane
The nonlinear data divided separates.Common kernel function is linear kernel function and gaussian kernel function.In terms of the selection of kernel function,
Due to Radar recognition problem characteristic small number, therefore select gaussian kernel function.
Support vector machines tool there are two key parameter, punishment parameter C's and nuclear parameter δ, the value of the two parameters is very
The quality of support vector machines performance is determined in big degree.It is empty in high dimensional feature that the parameter of kernel function mainly influences sample data
Between middle distribution complexity, i.e. dimension.The dimension of proper subspace is higher, and obtained optimal separating hyper plane will be more multiple
It is miscellaneous.Vice versa.Therefore suitable nuclear parameter is only selected to obtain suitable proper subspace, it is good to be just promoted ability
Good support vector machine classifier.Gauss nuclear parameter is used in the present invention.Lot of experimental data show if with sample point it
Between apart from very little, δ → 0;If the distance between sample point is very big, δ → ∞;When δ very littles, gaussian kernel function is supported
There is over-fitting close to being a constant in the discriminant function that vector machine obtains.When δ is very big, the correct classification rate of sample
It also can be than relatively low.
Punishment parameter is to influence another key factor of algorithm of support vector machine performance.Its main function is to adjust
The fiducial range of supporting vector machine model and the ratio of empiric risk in proper subspace make the generalization ability of support vector machines
Reach best.During proper subspace difference, optimal value of the parameter value also can be different.The punishment of punishment parameter and experience error and
The complexity of support vector machines is directly proportional, is inversely proportional with empiric risk value, vice versa.The present invention uses grid data service pair
Support vector machines parameter carries out tuning, and final choice parameter punishment parameter is 32, nuclear parameter 0.0312.
The method of the present invention is can be seen that with very high unknown classification identification precision and known classification from Fig. 9 results
Recognition accuracy.
The present invention disturbs big, radar signal ginseng for the electromagnetic signal that the identification of radiation source under complex electromagnetic environment faces
The problems such as number is close and challenge using the thought and method of deep learning, further investigate radiation source pulse finger-print, design
Suitable neural network structure, and verified based on actual airborne weather radar data.Main characteristics are with innovative point:(1)
Radar recognition is carried out using deep learning method.By analyzing existing emitter Signals, subtle spy in its arteries and veins is utilized
Sign improves recognition accuracy as training sample.Although have the engineerings such as research and utilization neutral net, support vector machines
It practises algorithm to be identified, however, there remains the basic parameter based on radar signal, the internal feature for not accounting for signal is joined
Number.(2) present invention has stronger antinoise, antijamming capability using method.Before conventional method carries out specific emitter identification
It is both needed to carry out the complicated Signal Pretreatment work such as noise reduction, multipaths restraint and sorting, these operations can weaken to a certain extent
The personal feature of radar.Deep learning method can pass through imparting by substantial amounts of sample, the weight of each feature of intelligent decision
Different weights avoids the influence of interference in the case where retaining radar personal feature.It can be seen that the present invention was used
Method has preferable robustness.
For the structure of convolutional neural networks in the present invention, although deep learning method is the prior art, for not
With the problem of its structure network model and centre parameter be different.The method of the present invention is also achieved for unknown
Classification and Identification, traditional sorting algorithm may only be identified for known classification, and the present invention is by adding support vector machines
Meta identifiers are recognized the output result of depth convolutional neural networks, realize the identification of unknown classification.
Claims (7)
1. a kind of radar emission source category recognition methods based on deep learning, which is characterized in that utilize the spoke by pretreatment
It penetrates source signal and asks for the section of its ambiguity function as feature vector;The feature vector of label will largely be accomplished fluently as training sample
This, be trained by depth convolutional neural networks, and using obtain convolutional neural networks grader progress input feature vector to
The Classification and Identification of amount;In order to realize the identification of the radiation source for unknown classification, the meta identifications based on support vector machines are built
Device judges whether the classification results of convolutional neural networks grader are credible, obtains final recognition result.
A kind of 2. radar emission source category recognition methods based on deep learning as described in claim 1, which is characterized in that tool
Body implementation is:
Step 1, emitter Signals pretreatment and feature extraction optimization:Number useless and wrong in emitter Signals is rejected first
According to then being sorted to emitter Signals;Extract again each radar slice of ambiguity function vector, and as feature to
Amount;
Step 2, structure depth convolutional neural networks structure:According to the feature of radiation source slice of ambiguity function, a depth is devised
Spend the convolutional neural networks for 13 layers, convolutional layer has selected the convolution kernel of the 1*3 with 128 layers, use length for 1*2 most
Great Chiization method, and add dropout parameters, by the use of softmax as final output layer;
Step 3, training neural network model:On the basis of traditional gradient optimization algorithm, first moment and second order are utilized
Moments estimation is adjusted learning rate;
Step 4, the identification of convolutional neural networks grader:The model obtained using step 3 carries out Classification and Identification, obtains the radiation source
Belong to the probability of each classification as preliminary output;
Step 5, structure support vector machines meta graders:The preliminary input exported as the grader obtained using in step 4,
Secondary judgement is carried out to classification, determines if it is unknown classification.
A kind of 3. radar emission source category recognition methods based on deep learning as claimed in claim 2, which is characterized in that institute
The specific method for stating step 1 is:
Step 1.1 rejects data useless and wrong in emitter Signals, and then emitter Signals are sorted, were sorted
Journey mainly using the correlation of same portion's radar emitter signal parameter and the otherness of different radar signal parameters, is handed over from random
The pulse signal of each radar is isolated in folded pulse signal stream and selects useful emitter Signals;
Step 1.2, the instantaneous auto-correlation function of emitter Signals x (t) are defined as:
Wherein, τ is time delay;
The definition of radar ambiguity function is:
<mrow>
<mi>A</mi>
<mrow>
<mo>(</mo>
<mi>&tau;</mi>
<mo>,</mo>
<mi>v</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msubsup>
<mo>&Integral;</mo>
<mrow>
<mo>-</mo>
<mi>&infin;</mi>
</mrow>
<mrow>
<mo>+</mo>
<mi>&infin;</mi>
</mrow>
</msubsup>
<msub>
<mi>R</mi>
<mi>x</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>,</mo>
<mi>&tau;</mi>
<mo>)</mo>
</mrow>
<msup>
<mi>e</mi>
<mrow>
<mi>j</mi>
<mn>2</mn>
<mi>&pi;</mi>
<mi>v</mi>
<mi>t</mi>
</mrow>
</msup>
<mi>d</mi>
<mi>t</mi>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
<mo>,</mo>
</mrow>
That is RxThe Fourier inversion of (t, τ) on time t, wherein υ are frequency deviation, and j is negative unit, and π is pi, and e is nature
Constant;
Formula (2) can be equivalent to following formula (3) by conversion:
<mrow>
<mi>A</mi>
<mrow>
<mo>(</mo>
<mi>&tau;</mi>
<mo>,</mo>
<mi>v</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msubsup>
<mo>&Integral;</mo>
<mn>0</mn>
<mi>&tau;</mi>
</msubsup>
<mi>x</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<msup>
<mi>x</mi>
<mo>*</mo>
</msup>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>+</mo>
<mi>&tau;</mi>
<mo>)</mo>
</mrow>
<msup>
<mi>e</mi>
<mrow>
<mi>j</mi>
<mn>2</mn>
<mi>&pi;</mi>
<mi>v</mi>
<mi>t</mi>
</mrow>
</msup>
<mi>d</mi>
<mi>t</mi>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>)</mo>
</mrow>
<mo>,</mo>
</mrow>
To emitter Signals x (t) uniform samplings, formula (3) becomes formula (4):
<mrow>
<mi>A</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>&tau;</mi>
<mi>l</mi>
</msub>
<mo>,</mo>
<msub>
<mi>v</mi>
<mi>m</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>A</mi>
<mrow>
<mo>(</mo>
<mi>l</mi>
<mo>,</mo>
<mi>m</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>n</mi>
<mo>=</mo>
<mn>0</mn>
</mrow>
<mrow>
<mi>N</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</munderover>
<mi>x</mi>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<msup>
<mi>x</mi>
<mo>*</mo>
</msup>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>+</mo>
<mi>l</mi>
<mo>)</mo>
</mrow>
<msup>
<mi>e</mi>
<mfrac>
<mrow>
<mi>j</mi>
<mn>2</mn>
<mi>&pi;</mi>
<mi>m</mi>
<mi>n</mi>
</mrow>
<mi>N</mi>
</mfrac>
</msup>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>4</mn>
<mo>)</mo>
</mrow>
<mo>,</mo>
</mrow>
Formula (4) is the slice of ambiguity function vector of the radar extracted, wherein, τlTime delay when for time offset being l,
vmFrequency deviation when for frequency offset being m, N is the total frequency spectrum cycle, if fsFor working frequency, then have
A kind of 4. radar emission source category recognition methods based on deep learning as claimed in claim 2, which is characterized in that institute
The specific method for stating step 2 is:
Step 2.1, convolution feature extraction:
If the sequence that the slice of ambiguity function vector for the emitter Signals that step 1.2 obtains is 1 × K, this is initial characteristics vector,
Convolution algorithm is carried out to it, first convolutional layer is obtained, is represented with C1;Use the convolution kernel that 128 sizes are 1 × 3, therefore feature
In vector each neuron with input in 1 × 3 neighborhood be connected, it is C1 layer such in feature sizes just be 1 × (K-3);Again
Because each wave filter has 3 cell parameters and an offset parameter, altogether 128 wave filters, common (1 × 3+1) × 128 of C1
=512 can training parameter, totally 512 × (K-3) a connection, will connection by ReLU active coatings, so far complete for initial
The convolution feature extraction of feature vector;
Step 2.2, maximum pondization processing:
Pond length is used as 2 maximum pond method, it will be through two adjacent in step 2.1 treated initial characteristics vector
Maximum feature is used for subsequent operation as the feature of Chi Huahou in element, realizes the dimensionality reduction of feature vector;
Step 2.3, as new input, is added by the feature vector that step 2.2 obtains continuous two convolutional layers, then is added
One pond layer, then two convolutional layers are added, then a pond layer is added, it so operates, fills by continuous several times convolution, pondization
Divide the information being extracted in initial characteristics vector, obtain a new feature vector;
Step 2.4, structure output layer:
It evens up and handles the new feature vector obtained through step 2.3, in this, as a transition of convolutional layer to full articulamentum,
Dropout parameters 0.5 are added on the basis of first full articulamentum, the second layer is then added and connects entirely, pass through activation primitive
Softmax obtains the output layer of classification results.
A kind of 5. radar emission source category recognition methods based on deep learning as claimed in claim 2, which is characterized in that institute
The specific method for stating step 3 is:
After neural network model is put up, further training is done to the model using training sample:
Step 3.1, initialization step-length ε=0.001, the expectation rate of disintegration of first moment and second order moments estimation is respectively ρ1=0.9, ρ2
=0.999, initial learning rate θ, first moment and second moment variable s=0, r=0, iterations k=0, maximum iteration Kmax;
Step 3.2 is obtained from training set corresponding to target y(i)The sampling { x with M sample(1)..., x(M), wherein M
For the number of a collection of sample in batch processing;
Step 3.3 calculates gradientWherein L is logarithm loss function, is defined
For for being distributed X, Y has L (P (Y | X), Y) and=- log P (Y | X), P represents probability;
Step 3.4, update iterations k ← k+1, there is inclined single order moments estimation s ← ρ1S+(1-ρ1) g, there is inclined second order moments estimationWhereinRepresent the product of corresponding element;
Step 3.5 corrects single order moments estimationSecond order moments estimation
Step 3.6, renewal learning rateWherein δ=10-8It is one to be used for keeping stability
Constant, judge iterations k whether be more than maximum iteration KmaxIf iterations k is not more than maximum iteration
KmaxWith regard to repeating step 3.2-3.6, if iterations k is more than maximum iteration KmaxWith regard to stopping iteration.
A kind of 6. radar emission source category recognition methods based on deep learning as claimed in claim 2, which is characterized in that institute
The specific method for stating step 4 is:
After the training of step 3, a depth convolutional neural networks grader, the ambiguity function of input radiation source signal are obtained
Section vector can obtain the output that the radiation source belongs to the probability of each classification by the grader.
A kind of 7. radar emission source category recognition methods based on deep learning as claimed in claim 2, which is characterized in that institute
The specific method for stating step 5 is:
Using the output of depth convolutional neural networks as the input of the grader, carried out using probability of all categories as its feature
Training identification chooses the recognition result that the consecutive numbers from same emitter Signals is clapped and carries out one averagely as final defeated
Enter, the input by support vector machine classifier can finally obtain its whether be unknown classification probability.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711145195.9A CN108090412A (en) | 2017-11-17 | 2017-11-17 | A kind of radar emission source category recognition methods based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711145195.9A CN108090412A (en) | 2017-11-17 | 2017-11-17 | A kind of radar emission source category recognition methods based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108090412A true CN108090412A (en) | 2018-05-29 |
Family
ID=62172540
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711145195.9A Pending CN108090412A (en) | 2017-11-17 | 2017-11-17 | A kind of radar emission source category recognition methods based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108090412A (en) |
Cited By (48)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108932480A (en) * | 2018-06-08 | 2018-12-04 | 电子科技大学 | The study of distributing optical fiber sensing signal characteristic and classification method based on 1D-CNN |
CN108985454A (en) * | 2018-06-28 | 2018-12-11 | 西南电子技术研究所(中国电子科技集团公司第十研究所) | Airline carriers of passengers individual goal recognition methods |
CN109271926A (en) * | 2018-09-14 | 2019-01-25 | 西安电子科技大学 | Intelligent Radiation source discrimination based on GRU depth convolutional network |
CN109272040A (en) * | 2018-09-20 | 2019-01-25 | 中国科学院电子学研究所苏州研究院 | A kind of radar operation mode generation method |
CN109272027A (en) * | 2018-08-30 | 2019-01-25 | 昆明理工大学 | A kind of pulsar candidate's body recognition methods based on artificial neural network ensembles |
CN109324527A (en) * | 2018-09-04 | 2019-02-12 | 西北工业大学 | Radar resource management method based on the support of empty world sea Information Network |
CN109407067A (en) * | 2018-10-13 | 2019-03-01 | 中国人民解放军海军航空大学 | Radar moving targets detection and classification integral method based on time-frequency figure convolutional neural networks |
CN109471074A (en) * | 2018-11-09 | 2019-03-15 | 西安电子科技大学 | Recognition Method of Radar Emitters based on singular value decomposition Yu one-dimensional CNN network |
CN109507648A (en) * | 2018-12-19 | 2019-03-22 | 西安电子科技大学 | Recognition Method of Radar Emitters based on VAE-ResNet network |
CN109614905A (en) * | 2018-12-03 | 2019-04-12 | 中国人民解放军空军工程大学 | A kind of radar emitter signal depth intrapulse feature extraction method |
CN109633588A (en) * | 2018-12-29 | 2019-04-16 | 杭州电子科技大学 | Recognition Method of Radar Emitters based on depth convolutional neural networks |
CN109711314A (en) * | 2018-12-21 | 2019-05-03 | 西安电子科技大学 | Radar emitter signal classification method based on Fusion Features and SAE |
CN110018447A (en) * | 2019-04-09 | 2019-07-16 | 电子科技大学 | Signal transmitter individual discrimination method based on double-spectrum analysis and convolutional neural networks |
CN110109059A (en) * | 2019-03-27 | 2019-08-09 | 西安电子科技大学 | A kind of radar emitter signal recognition methods based on deep learning network |
CN110187321A (en) * | 2019-05-30 | 2019-08-30 | 电子科技大学 | Radar emitter characteristic parameter extraction method under complex environment based on deep learning |
CN110427835A (en) * | 2019-07-11 | 2019-11-08 | 清华大学 | The electromagnet signal recognition method and device of figure convolutional network and transfer learning |
CN110596668A (en) * | 2019-09-20 | 2019-12-20 | 中国人民解放军国防科技大学 | Target external radiation source passive positioning method based on reciprocal deep neural network |
CN110632584A (en) * | 2019-09-20 | 2019-12-31 | 中国人民解放军国防科技大学 | Passive target external radiation source positioning method based on parallel radial basis network |
CN110751869A (en) * | 2019-10-12 | 2020-02-04 | 南京摄星智能科技有限公司 | Simulated environment and battlefield situation strategy transfer technology based on countermeasure discrimination migration method |
CN110988803A (en) * | 2019-11-11 | 2020-04-10 | 浙江大学 | Radar radiation source individual identification system with high confidence degree dynamic adjustment |
CN110988804A (en) * | 2019-11-11 | 2020-04-10 | 浙江大学 | Radar radiation source individual identification system based on radar pulse sequence |
CN110988802A (en) * | 2019-11-11 | 2020-04-10 | 浙江大学 | Radar radiation source identification system based on signal scale decomposition |
CN111123232A (en) * | 2019-11-11 | 2020-05-08 | 浙江大学 | Radar individual identification system with task adaptability |
CN111126134A (en) * | 2019-11-11 | 2020-05-08 | 浙江大学 | Radar radiation source deep learning identification method based on non-fingerprint signal eliminator |
CN111190146A (en) * | 2020-01-13 | 2020-05-22 | 中国船舶重工集团公司第七二四研究所 | Complex signal sorting method based on visual graphic features |
CN111239692A (en) * | 2020-01-10 | 2020-06-05 | 西安电子科技大学 | PRI (pulse repetition index) combined intra-pulse information radiation source signal identification method based on deep learning |
CN111310680A (en) * | 2020-02-24 | 2020-06-19 | 电子科技大学 | Radiation source individual identification method based on deep learning |
CN111310719A (en) * | 2020-03-10 | 2020-06-19 | 电子科技大学 | Unknown radiation source individual identification and detection method |
CN111401226A (en) * | 2020-03-13 | 2020-07-10 | 电子科技大学 | Rapid identification method for radiation source |
CN111582320A (en) * | 2020-04-17 | 2020-08-25 | 电子科技大学 | Dynamic individual identification method based on semi-supervised learning |
CN111913156A (en) * | 2020-08-05 | 2020-11-10 | 电子科技大学 | Radar radiation source individual identification method based on deep learning model and feature combination |
CN112115768A (en) * | 2020-08-03 | 2020-12-22 | 中国人民解放军63892部队 | Radar radiation source identification method facing complex electromagnetic environment |
CN112183279A (en) * | 2020-09-21 | 2021-01-05 | 中国人民解放军国防科技大学 | Communication radiation source individual identification method based on IQ graph characteristics |
CN112464713A (en) * | 2020-10-21 | 2021-03-09 | 安徽农业大学 | Communication radiation source radio frequency fingerprint identification method based on deep learning |
CN112633320A (en) * | 2020-11-26 | 2021-04-09 | 西安电子科技大学 | Radar radiation source data cleaning method based on phase image coefficient and DBSCAN |
CN112668498A (en) * | 2020-12-30 | 2021-04-16 | 西安电子科技大学 | Method, system, terminal and application for identifying individual intelligent increment of aerial radiation source |
CN112711978A (en) * | 2020-11-13 | 2021-04-27 | 浙江大学 | High-confidence-degree rapid radar radiation source individual recognition instrument |
CN112884059A (en) * | 2021-03-09 | 2021-06-01 | 电子科技大学 | Small sample radar working mode classification method fusing priori knowledge |
CN112949383A (en) * | 2021-01-22 | 2021-06-11 | 中国人民解放军63892部队 | Waveform agility radar radiation source identification method based on Hydeep-Att network |
CN113095354A (en) * | 2021-03-03 | 2021-07-09 | 电子科技大学 | Unknown radar target identification method based on radiation source characteristic subspace knowledge |
CN113158886A (en) * | 2021-04-19 | 2021-07-23 | 中国人民解放军63892部队 | Waveform agility radar radiation source identification method based on deep reinforcement learning |
CN113177520A (en) * | 2021-05-26 | 2021-07-27 | 电子科技大学 | Intelligent radiation source identification method based on ensemble learning |
CN113298138A (en) * | 2021-05-21 | 2021-08-24 | 西安建筑科技大学 | Radar radiation source individual identification method and system |
CN113359091A (en) * | 2021-05-28 | 2021-09-07 | 太原理工大学 | Deep learning-based multi-kernel function aliasing radar radiation source identification method and system |
CN113435245A (en) * | 2021-05-18 | 2021-09-24 | 西安电子科技大学 | Method, system and application for identifying individual airborne radiation source |
CN113780521A (en) * | 2021-08-24 | 2021-12-10 | 中国人民解放军93114部队 | Radiation source individual identification method based on deep learning |
CN113899948A (en) * | 2021-12-08 | 2022-01-07 | 成都中星世通电子科技有限公司 | System and method for quickly extracting electromagnetic spectrum target characteristic data |
CN115166679A (en) * | 2022-09-06 | 2022-10-11 | 中国人民解放军海军工程大学 | Task attribute-based unknown radar radiation source target characterization identification method |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107220606A (en) * | 2017-05-22 | 2017-09-29 | 西安电子科技大学 | The recognition methods of radar emitter signal based on one-dimensional convolutional neural networks |
CN107301381A (en) * | 2017-06-01 | 2017-10-27 | 西安电子科技大学昆山创新研究院 | Recognition Method of Radar Emitters based on deep learning and multi-task learning strategy |
-
2017
- 2017-11-17 CN CN201711145195.9A patent/CN108090412A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107220606A (en) * | 2017-05-22 | 2017-09-29 | 西安电子科技大学 | The recognition methods of radar emitter signal based on one-dimensional convolutional neural networks |
CN107301381A (en) * | 2017-06-01 | 2017-10-27 | 西安电子科技大学昆山创新研究院 | Recognition Method of Radar Emitters based on deep learning and multi-task learning strategy |
Non-Patent Citations (3)
Title |
---|
DIEDERIK P KINGMA ET AL: "ADAM:A METHOD FOR STOCHASTIC OPTIMIZATION", 《ARXIV:1412.6980V9》 * |
WEI-HAO LIN ET AL: "News Video Classification Using SVM-based Multimodal Classifiers and Combination Strategies", 《MULITMEDIA’02》 * |
周志文等: "一种深度学习的雷达辐射源识别算法", 《西安电子科技大学学报(自然科学版)》 * |
Cited By (74)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11562224B2 (en) * | 2018-06-08 | 2023-01-24 | University Of Electronic Science And Technology Of China | 1D-CNN-based distributed optical fiber sensing signal feature learning and classification method |
CN108932480A (en) * | 2018-06-08 | 2018-12-04 | 电子科技大学 | The study of distributing optical fiber sensing signal characteristic and classification method based on 1D-CNN |
CN108932480B (en) * | 2018-06-08 | 2022-03-15 | 电子科技大学 | Distributed optical fiber sensing signal feature learning and classifying method based on 1D-CNN |
CN108985454A (en) * | 2018-06-28 | 2018-12-11 | 西南电子技术研究所(中国电子科技集团公司第十研究所) | Airline carriers of passengers individual goal recognition methods |
CN108985454B (en) * | 2018-06-28 | 2022-04-01 | 西南电子技术研究所(中国电子科技集团公司第十研究所) | Individual target identification method for civil aviation airliner |
CN109272027A (en) * | 2018-08-30 | 2019-01-25 | 昆明理工大学 | A kind of pulsar candidate's body recognition methods based on artificial neural network ensembles |
CN109324527A (en) * | 2018-09-04 | 2019-02-12 | 西北工业大学 | Radar resource management method based on the support of empty world sea Information Network |
CN109324527B (en) * | 2018-09-04 | 2021-04-02 | 西北工业大学 | Radar resource management method based on aerospace, geodetic and marine information network support |
CN109271926A (en) * | 2018-09-14 | 2019-01-25 | 西安电子科技大学 | Intelligent Radiation source discrimination based on GRU depth convolutional network |
CN109272040A (en) * | 2018-09-20 | 2019-01-25 | 中国科学院电子学研究所苏州研究院 | A kind of radar operation mode generation method |
CN109407067B (en) * | 2018-10-13 | 2023-06-27 | 中国人民解放军海军航空大学 | Radar moving target detection and classification integrated method based on time-frequency graph convolution neural network |
CN109407067A (en) * | 2018-10-13 | 2019-03-01 | 中国人民解放军海军航空大学 | Radar moving targets detection and classification integral method based on time-frequency figure convolutional neural networks |
CN109471074A (en) * | 2018-11-09 | 2019-03-15 | 西安电子科技大学 | Recognition Method of Radar Emitters based on singular value decomposition Yu one-dimensional CNN network |
CN109614905B (en) * | 2018-12-03 | 2022-10-21 | 中国人民解放军空军工程大学 | Automatic extraction method for depth intra-pulse features of radar radiation source signals |
CN109614905A (en) * | 2018-12-03 | 2019-04-12 | 中国人民解放军空军工程大学 | A kind of radar emitter signal depth intrapulse feature extraction method |
CN109507648A (en) * | 2018-12-19 | 2019-03-22 | 西安电子科技大学 | Recognition Method of Radar Emitters based on VAE-ResNet network |
CN109711314A (en) * | 2018-12-21 | 2019-05-03 | 西安电子科技大学 | Radar emitter signal classification method based on Fusion Features and SAE |
CN109711314B (en) * | 2018-12-21 | 2021-04-02 | 西安电子科技大学 | Radar radiation source signal classification method based on feature fusion and SAE |
CN109633588A (en) * | 2018-12-29 | 2019-04-16 | 杭州电子科技大学 | Recognition Method of Radar Emitters based on depth convolutional neural networks |
CN110109059A (en) * | 2019-03-27 | 2019-08-09 | 西安电子科技大学 | A kind of radar emitter signal recognition methods based on deep learning network |
CN110018447A (en) * | 2019-04-09 | 2019-07-16 | 电子科技大学 | Signal transmitter individual discrimination method based on double-spectrum analysis and convolutional neural networks |
CN110187321A (en) * | 2019-05-30 | 2019-08-30 | 电子科技大学 | Radar emitter characteristic parameter extraction method under complex environment based on deep learning |
CN110187321B (en) * | 2019-05-30 | 2022-07-22 | 电子科技大学 | Radar radiation source characteristic parameter extraction method based on deep learning in complex environment |
CN110427835B (en) * | 2019-07-11 | 2022-03-11 | 清华大学 | Electromagnetic signal identification method and device for graph convolution network and transfer learning |
CN110427835A (en) * | 2019-07-11 | 2019-11-08 | 清华大学 | The electromagnet signal recognition method and device of figure convolutional network and transfer learning |
CN110632584A (en) * | 2019-09-20 | 2019-12-31 | 中国人民解放军国防科技大学 | Passive target external radiation source positioning method based on parallel radial basis network |
CN110596668A (en) * | 2019-09-20 | 2019-12-20 | 中国人民解放军国防科技大学 | Target external radiation source passive positioning method based on reciprocal deep neural network |
CN110751869B (en) * | 2019-10-12 | 2021-11-05 | 南京星耀智能科技有限公司 | Simulated environment and battlefield situation strategy transfer technology based on countermeasure discrimination migration method |
CN110751869A (en) * | 2019-10-12 | 2020-02-04 | 南京摄星智能科技有限公司 | Simulated environment and battlefield situation strategy transfer technology based on countermeasure discrimination migration method |
CN111123232B (en) * | 2019-11-11 | 2022-03-18 | 浙江大学 | Radar individual identification system with task adaptability |
CN111126134B (en) * | 2019-11-11 | 2022-03-22 | 浙江大学 | Radar radiation source deep learning identification method based on non-fingerprint signal eliminator |
CN110988804A (en) * | 2019-11-11 | 2020-04-10 | 浙江大学 | Radar radiation source individual identification system based on radar pulse sequence |
CN110988802A (en) * | 2019-11-11 | 2020-04-10 | 浙江大学 | Radar radiation source identification system based on signal scale decomposition |
CN110988804B (en) * | 2019-11-11 | 2022-01-25 | 浙江大学 | Radar radiation source individual identification system based on radar pulse sequence |
CN110988803A (en) * | 2019-11-11 | 2020-04-10 | 浙江大学 | Radar radiation source individual identification system with high confidence degree dynamic adjustment |
CN111126134A (en) * | 2019-11-11 | 2020-05-08 | 浙江大学 | Radar radiation source deep learning identification method based on non-fingerprint signal eliminator |
CN111123232A (en) * | 2019-11-11 | 2020-05-08 | 浙江大学 | Radar individual identification system with task adaptability |
CN111239692A (en) * | 2020-01-10 | 2020-06-05 | 西安电子科技大学 | PRI (pulse repetition index) combined intra-pulse information radiation source signal identification method based on deep learning |
CN111239692B (en) * | 2020-01-10 | 2023-01-24 | 西安电子科技大学 | PRI (pulse repetition index) combined intra-pulse information radiation source signal identification method based on deep learning |
CN111190146B (en) * | 2020-01-13 | 2021-02-09 | 中国船舶重工集团公司第七二四研究所 | Complex signal sorting method based on visual graphic features |
CN111190146A (en) * | 2020-01-13 | 2020-05-22 | 中国船舶重工集团公司第七二四研究所 | Complex signal sorting method based on visual graphic features |
CN111310680A (en) * | 2020-02-24 | 2020-06-19 | 电子科技大学 | Radiation source individual identification method based on deep learning |
CN111310680B (en) * | 2020-02-24 | 2022-11-04 | 电子科技大学 | Radiation source individual identification method based on deep learning |
CN111310719A (en) * | 2020-03-10 | 2020-06-19 | 电子科技大学 | Unknown radiation source individual identification and detection method |
CN111401226B (en) * | 2020-03-13 | 2022-11-01 | 电子科技大学 | Rapid identification method for radiation source |
CN111401226A (en) * | 2020-03-13 | 2020-07-10 | 电子科技大学 | Rapid identification method for radiation source |
CN111582320B (en) * | 2020-04-17 | 2022-10-14 | 电子科技大学 | Dynamic individual identification method based on semi-supervised learning |
CN111582320A (en) * | 2020-04-17 | 2020-08-25 | 电子科技大学 | Dynamic individual identification method based on semi-supervised learning |
CN112115768A (en) * | 2020-08-03 | 2020-12-22 | 中国人民解放军63892部队 | Radar radiation source identification method facing complex electromagnetic environment |
CN111913156A (en) * | 2020-08-05 | 2020-11-10 | 电子科技大学 | Radar radiation source individual identification method based on deep learning model and feature combination |
CN112183279A (en) * | 2020-09-21 | 2021-01-05 | 中国人民解放军国防科技大学 | Communication radiation source individual identification method based on IQ graph characteristics |
CN112464713B (en) * | 2020-10-21 | 2022-11-08 | 安徽农业大学 | Communication radiation source radio frequency fingerprint identification method based on deep learning |
CN112464713A (en) * | 2020-10-21 | 2021-03-09 | 安徽农业大学 | Communication radiation source radio frequency fingerprint identification method based on deep learning |
CN112711978A (en) * | 2020-11-13 | 2021-04-27 | 浙江大学 | High-confidence-degree rapid radar radiation source individual recognition instrument |
CN112633320A (en) * | 2020-11-26 | 2021-04-09 | 西安电子科技大学 | Radar radiation source data cleaning method based on phase image coefficient and DBSCAN |
CN112668498A (en) * | 2020-12-30 | 2021-04-16 | 西安电子科技大学 | Method, system, terminal and application for identifying individual intelligent increment of aerial radiation source |
CN112668498B (en) * | 2020-12-30 | 2024-02-06 | 西安电子科技大学 | Method, system, terminal and application for identifying individual intelligent increment of aerial radiation source |
CN112949383A (en) * | 2021-01-22 | 2021-06-11 | 中国人民解放军63892部队 | Waveform agility radar radiation source identification method based on Hydeep-Att network |
CN113095354A (en) * | 2021-03-03 | 2021-07-09 | 电子科技大学 | Unknown radar target identification method based on radiation source characteristic subspace knowledge |
CN112884059A (en) * | 2021-03-09 | 2021-06-01 | 电子科技大学 | Small sample radar working mode classification method fusing priori knowledge |
CN113158886A (en) * | 2021-04-19 | 2021-07-23 | 中国人民解放军63892部队 | Waveform agility radar radiation source identification method based on deep reinforcement learning |
CN113435245A (en) * | 2021-05-18 | 2021-09-24 | 西安电子科技大学 | Method, system and application for identifying individual airborne radiation source |
CN113435245B (en) * | 2021-05-18 | 2023-06-30 | 西安电子科技大学 | Method, system and application for identifying individual aerial radiation source |
CN113298138A (en) * | 2021-05-21 | 2021-08-24 | 西安建筑科技大学 | Radar radiation source individual identification method and system |
CN113298138B (en) * | 2021-05-21 | 2024-04-23 | 西安建筑科技大学 | Individual identification method and system for radar radiation source |
CN113177520B (en) * | 2021-05-26 | 2022-06-28 | 电子科技大学 | Intelligent radiation source identification method based on ensemble learning |
CN113177520A (en) * | 2021-05-26 | 2021-07-27 | 电子科技大学 | Intelligent radiation source identification method based on ensemble learning |
CN113359091B (en) * | 2021-05-28 | 2022-05-17 | 太原理工大学 | Deep learning-based multi-kernel function aliasing radar radiation source identification method and system |
CN113359091A (en) * | 2021-05-28 | 2021-09-07 | 太原理工大学 | Deep learning-based multi-kernel function aliasing radar radiation source identification method and system |
CN113780521B (en) * | 2021-08-24 | 2022-03-04 | 中国人民解放军93114部队 | Radiation source individual identification method based on deep learning |
CN113780521A (en) * | 2021-08-24 | 2021-12-10 | 中国人民解放军93114部队 | Radiation source individual identification method based on deep learning |
CN113899948A (en) * | 2021-12-08 | 2022-01-07 | 成都中星世通电子科技有限公司 | System and method for quickly extracting electromagnetic spectrum target characteristic data |
CN115166679A (en) * | 2022-09-06 | 2022-10-11 | 中国人民解放军海军工程大学 | Task attribute-based unknown radar radiation source target characterization identification method |
CN115166679B (en) * | 2022-09-06 | 2022-12-16 | 中国人民解放军海军工程大学 | Task attribute-based unknown radar radiation source target characterization identification method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108090412A (en) | A kind of radar emission source category recognition methods based on deep learning | |
CN107464210B (en) | Image style migration method based on generating type countermeasure network | |
CN111178260B (en) | Modulation signal time-frequency diagram classification system based on generation countermeasure network and operation method thereof | |
CN107301381A (en) | Recognition Method of Radar Emitters based on deep learning and multi-task learning strategy | |
Zongker et al. | Algorithms for feature selection: An evaluation | |
CN109271926A (en) | Intelligent Radiation source discrimination based on GRU depth convolutional network | |
CN109886464B (en) | Low-information-loss short-term wind speed prediction method based on optimized singular value decomposition generated feature set | |
CN104794368A (en) | Rolling bearing fault classifying method based on FOA-MKSVM (fruit fly optimization algorithm-multiple kernel support vector machine) | |
CN110751121B (en) | Unsupervised radar signal sorting method based on clustering and SOFM | |
CN113452408B (en) | Network station frequency hopping signal sorting method | |
CN112884059B (en) | Small sample radar working mode classification method fusing priori knowledge | |
CN108256436A (en) | A kind of radar HRRP target identification methods based on joint classification | |
CN105701503B (en) | Interference SAR image classification method based on DBN model | |
CN109086886A (en) | A kind of convolutional neural networks learning algorithm based on extreme learning machine | |
CN112036239B (en) | Radar signal working mode identification method and system based on deep learning network | |
CN114564982A (en) | Automatic identification method for radar signal modulation type | |
Li et al. | Automatic modulation classification based on bispectrum and CNN | |
CN109284662A (en) | A kind of transfer learning method towards the classification of underwater voice signal | |
Li et al. | Carrier-free UWB sensor small-sample terrain recognition based on improved ACGAN with self-attention | |
CN107330447B (en) | Silhouette recognition system combining feedback type ICM neural network and FPF | |
CN113159218A (en) | Radar HRRP multi-target identification method and system based on improved CNN | |
Su et al. | Unsupervised k-means combined with SOFM structure adaptive radar signal sorting algorithm | |
Wang et al. | Kernel-based deep learning for intelligent data analysis | |
CN109784311A (en) | Target identification method based on chirplet networks of atoms | |
CN115982349A (en) | Improved TF-IDF-based intelligent classification method for client electricity utilization safety |
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: 20180529 |
|
RJ01 | Rejection of invention patent application after publication |