CN110532932A - A kind of multi -components radar emitter signal intra-pulse modulation mode recognition methods - Google Patents
A kind of multi -components radar emitter signal intra-pulse modulation mode recognition methods Download PDFInfo
- Publication number
- CN110532932A CN110532932A CN201910787759.1A CN201910787759A CN110532932A CN 110532932 A CN110532932 A CN 110532932A CN 201910787759 A CN201910787759 A CN 201910787759A CN 110532932 A CN110532932 A CN 110532932A
- Authority
- CN
- China
- Prior art keywords
- signal
- network
- time
- training
- radar
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/88—Radar or analogous systems specially adapted for specific applications
- G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
-
- 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
- G06F2218/04—Denoising
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- Physics & Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- General Physics & Mathematics (AREA)
- Electromagnetism (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The present invention relates to the automatic identification algorithm fields of deep learning, and in particular to a kind of multi -components radar emitter signal intra-pulse modulation mode recognition methods.Obtain the simple component of several different intra-pulse modulation modes or the time-frequency image of overlapping multi -components radar signal；Using image processing algorithm, radar signal time-frequency image is pre-processed, using the signal type for including in radar signal as label, makes training set and test set；It designs the pre-training network based on convolutional neural networks and extracts radar signal time-frequency image feature, design the multicomponent data processing sorter network based on intensified learning and obtain classification recognition result；It trains, test, improve network structure and parameter；Realize the Classification and Identification of multicomponent data processing.Multi -components Radar Signal Recognition algorithm of the present invention has extensive radar signal type adaptation range and higher recognition accuracy in low signal-to-noise ratio, realizes the intrapulse modulation recognition of random overlapping multi -components radar signal.
Description
Technical field
The present invention relates to the automatic identification algorithm fields of deep learning, and in particular to a kind of multi -components Radar Signal In-Pulse Characteristics are interior to be adjusted
Mode recognition methods processed.
Background technique
Radar emitter signal intra-pulse modulation mode identification is hyundai electronics intelligence reconnaissance, the important link in electronic support system.
Since radar signal density is continuously increased in modern electronic warfare environment, and when Modern Radar Signal mostly uses big greatly
Wide pulse compression signal, reconnaissance system for radar often intercept the overlapping pulse of time domain, form multi -components radar signal.It is existing
The most of radar signal Modulation Mode Recognition technologies having do not have adaptability to multicomponent data processing environment, cause signal identification wrong
Mistake or recognition failures.Therefore, for the analysis of multicomponent data processing and processing be in current radar reconnaissance system it is in the urgent need to address
The problem of.
Currently, the recognition methods of multi -components radar signal is there are two types of thinking: a kind of thinking is using signal in certain transformation
Separability in domain extracts feature of the signal in transform domain and carries out Classification and Identification to multicomponent data processing；Another thinking is benefit
Classification and Identification is carried out with the recognition methods of the separation method combination simple component signal of multicomponent data processing.Yu Zhibin was utilized in 2012
Separability of the phase shift keyed signal on cycle frequency axis proposes a kind of multicomponent data processing identification for phase shift keyed signal
Method, this method reach 97% to the average correct recognition rata of 3 class phase-shift keying (PSK) radar signals when signal-to-noise ratio is 0dB.Tong Chao
A kind of multi -components Radar Signal Recognition method using independent component analysis combination wavelet transformation, this method were proposed in 2016
When signal-to-noise ratio is 0dB, 90% or more is reached to the average correct recognition rata of 4 class radar signals.
There are some problems in the recognition methods of the multi -components radar signal proposed at present: being mentioned based on signal transform domain feature
The recognition methods taken has limitation, effective only for certain a kind of signal specific, it is difficult to adapt to extensive radar signal type；Base
The separation of multicomponent data processing separation algorithm is largely determined by the recognition effect of the recognition methods of multicomponent data processing separation
Effect, however the separation algorithm proposed at present there are noiseproof features it is poor, calculation amount is higher the problems such as, and be difficult to solve time-frequency domain friendship
Folded multicomponent data processing separation problem, these problems are by the recognition capability of limit algorithm.
Summary of the invention
The purpose of the present invention is to provide a kind of multi -components radar emitter signal intra-pulse modulation mode recognition methods, in low noise
Than in the case of, the intrapulse modulation recognition for the typical radar signal that 8 classes overlap at random is realized, and algorithm is to simple component radar
The intrapulse modulation recognition of signal is likewise supplied with adaptability.
The embodiment of the present invention provides a kind of multi -components radar emitter signal intra-pulse modulation mode recognition methods, comprising:
Step 1: the simple component of several different intra-pulse modulation modes or the time-frequency figure of overlapping multi -components radar signal are obtained
Picture, including LFM signal, MP signal, SFM signal, bpsk signal, 2FSK signal, 4FSK signal, EQFM signal and Frank signal,
Using these signals as sample signal, when being converted to radar signal using the signal that time-frequency distributions receive radar receiver
Frequency image；
Step 2: utilizing image processing algorithm, pre-process to the radar signal time-frequency image that above-mentioned steps one obtain,
Inhibit time-frequency image noise by two dimensional wiener filtering, simultaneously amplitude normalization then is adjusted to the size of time-frequency image, it will
By pretreated radar signal time-frequency image as sample, using the signal type for including in radar signal as label, production
Obtain training set and test set；
Step 3: according to algorithm requirements, overall network is extracted time-frequency image feature and is carried out according to the feature extracted more
Labeling identification designs the pre-training network based on convolutional neural networks and extracts radar signal time-frequency image feature, and designs
Multicomponent data processing sorter network based on intensified learning obtains classification recognition result；
Step 4: the network architecture designed in above-mentioned steps three is instructed according to training set obtained in above-mentioned steps two
Practice, and the recognition effect of network is tested using the test set obtained in above-mentioned steps two, according to test result, improves net
Network structure and parameter obtains the network of training completion；
Step 5: to random overlap signal any in above-mentioned steps one as to be identified after the processing of above-mentioned steps two
Radar signal time-frequency image sample, input above-mentioned steps four training complete network in, training complete network can basis
Input provides the radar signal set of types for including in current demand signal, realizes the Classification and Identification of multicomponent data processing；
The invention also includes structure features some in this way:
The step 1, comprising:
Wherein, the time-frequency distributions method particularly includes:
To the signal x (t) received, using Cohen class time-frequency distributions, mathematic(al) representation are as follows:
In above formula, t and ω represent independent variable time and the angular frequency of time-frequency distributions, and φ (τ, v) is known as kernel function, the present invention
It proposes to utilize two seed nucleus for the signal of different chirp rates in the characteristic distributions of fuzzy field according to radar signal and cross term
Function obtains the time-frequency distributions of radar signal respectively, and kernel function expression formula is respectively as follows:
In above formula, α, β, γ and ε are used to adjust the size of kernel function, obtain radar signal respectively under two kinds of kernel functions
Cohen class time-frequency distributions image, collectively as the time-frequency distributions information of current radar signal；
In the step 3, the planned network framework method particularly includes:
A) design convolutional neural networks framework is used for feature extraction, and convolutional neural networks framework is by 3 convolutional layers and 3 ponds
Change layer composition, carries out vectorized process by exporting to network the last layer, obtain the feature vector of current time-frequency image；
B) structure that full articulamentum is followed by Softmax layers constructs multiple sorter network units as sorter network unit
The label output for respectively corresponding multicomponent data processing, classifies to the feature vector obtained in step a), and with convolutional Neural net
Network framework is collectively as pre-training time-frequency image feature extraction network；
C) design deep layer intensified learning Recognition with Recurrent Neural Network is used for multicomponent data processing Classification and Identification, point in alternative steps b)
Class network unit records the control by kinds of the feature vector and previous cycle step that obtain in step a) collectively as input,
Multicomponent data processing recognition result is exported by multiple loop iteration assorting process；
In the step 4, the training and test method particularly includes:
D) after step 2, using the radar signal time-frequency image by image preprocessing as the defeated of convolutional neural networks
Enter, the tag along sort of each corresponding radar signal component of sorter network unit output, to time-frequency image characteristics extraction net
Network carries out pre-training, and the parameter of convolutional neural networks is retained after training；
E) convolutional neural networks and deep layer intensified learning Recognition with Recurrent Neural Network then obtained step d) training are as new
Sorter network keep convolutional neural networks model parameter constant to promote network training efficiency, using identical as step d)
Training set, individually training deep layer intensified learning Recognition with Recurrent Neural Network model parameter, and retains final convolutional neural networks and depth
All parameters of layer intensified learning Recognition with Recurrent Neural Network；
F) sample of test set is input to convolutional neural networks and deep layer intensified learning Recognition with Recurrent Neural Network composition one by one
In classifier, classifier can recycle the signal type collection for repeatedly exporting and including in current sample, and contrast sample corresponds to standard signal
Set of types eventually by the identification rate of precision that whole correct identification probability and each signal type is calculated and is recalled
Rate；
G) network structure and parameter are improved according to the recognition effect test result of step f), and new network structure is carried out
Training, recurrent network adjustment and training process, until the recognition effect test result of network reaches expected, completion is trained；
The beneficial effects of the present invention are:
1. Time-Frequency Analysis Method proposed by the present invention is directed to the characteristics of different radar signals and designs different kernel functions, take into account
The Energy distribution situation of different chirp rate radar signals obtains the time-frequency image with higher signal energy time-frequency locality；
2. the present invention carries out pre-training to convolutional neural networks, radar signal time-frequency image further feature is obtained, is mentioned simultaneously
The training effectiveness of high subsequent multicomponent data processing sorter network；
3. the present invention is trained multi -components sorter network using the training method of intensified learning, network is improved to time-frequency
The adaptability of image pattern；The connection between multiple classification recognition result is established using recirculating network framework, classification is improved and knows
The accuracy and reliability of other result.
Detailed description of the invention
Fig. 1 is a kind of flow chart of multi -components radar emitter signal intra-pulse modulation mode recognition methods；
Fig. 2 is that the present invention is based on the structure charts of the pre-training time-frequency image feature extraction network of convolutional neural networks；
Fig. 3 is that the present invention is based on the structures of convolutional neural networks and the multicomponent data processing Classification and Identification network of intensified learning
Figure；
Fig. 4 is the present invention to the average correct recognition rata of multi -components radar signal and the schematic diagram of Between Signal To Noise Ratio；
Fig. 5 is that the present invention overlaps the identification rate of precision of multi -components radar signal and the signal of Between Signal To Noise Ratio to 8 classes at random
Figure；
Fig. 6 is that the present invention overlaps the identification recall rate of multi -components radar signal and the signal of Between Signal To Noise Ratio to 8 classes at random
Figure；
Fig. 7 is the present invention to the average correct recognition rata of simple component radar signal and the schematic diagram of Between Signal To Noise Ratio.
Specific embodiment
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing to the present invention
It is described further:
The technical scheme of the present invention is realized as follows:
If Fig. 1 is the flow chart of multi -components radar emitter signal intra-pulse modulation mode recognizer of the invention, below with reference to Fig. 1
The step of to the algorithm and principle are described in detail.
Step 1: the simple component of several different intra-pulse modulation modes or the time-frequency image of overlapping multi -components radar signal are obtained,
It, will including LFM signal, MP signal, SFM signal, bpsk signal, 2FSK signal, 4FSK signal, EQFM signal and Frank signal
These signals are converted to time-frequency image as sample signal, using the signal that time-frequency distributions receive radar receiver.
In this step, the radar signal mathematical model after noise is added can be write as:
X (t)=s (t)+n (t) (1)
In above formula, x (t) is the signal that radar receives, and s (t) is signal, and n (t) indicates interchannel noise.
The present invention is using the Cohen class time-frequency distributions that radar signal is converted into time-frequency image
Wherein t and ω represents independent variable time and the angular frequency of time-frequency distributions, and φ (τ, v) is known as kernel function.Root of the present invention
It proposes to utilize two seed nucleus letters for the signal of different chirp rates in the characteristic distributions of fuzzy field according to radar signal and cross term
Number obtains the time-frequency distributions of radar signal respectively, and kernel function expression formula is respectively
Wherein, kernel functional parameter α, β, γ and ε is used to adjust the size of kernel function.For the lower radar letter of chirp rate
Number, the present invention uses double gauss kernel function φ1(τ, v) adjusts the parameter of kernel function, its long axis is made to be directed toward delay, τ axis, obtains low
The main energetic of chirp rate signal；For the higher radar signal of chirp rate, the present invention proposes a kind of new kernel function φ2
(τ, v) adjusts the parameter of kernel function, so that its long axis is directed toward frequency displacement v axis, obtains the main energetic of high frequency modulation slope signal.Respectively
Cohen class time-frequency distributions image of the radar signal under two kinds of kernel functions is obtained, collectively as the when frequency division of current radar signal
Cloth information.
Step 2: radar signal time-frequency image being pre-processed using image processing algorithm, is pressed down by two dimensional wiener filtering
Time-frequency image noise processed, adjustment image size and amplitude normalization, then according to treated radar signal time-frequency image and right
The signal class label collection production training set and test set that induction signal includes.
This algorithm inhibits the noise of time-frequency image using two dimensional wiener filtering, by calculating a certain picture of time-frequency image
Mean value and variance in vegetarian refreshments filtering neighborhood are adjusted the amplitude of the pixel, realize the two-dimentional wiener filter of time-frequency image
Wave；Time-frequency image size is adjusted using bilinear interpolation, the method by pressing linear interpolation by row, after calculating is sized
The amplitude of each pixel of image realizes the adjustment of time-frequency image size；In order to facilitate subsequent classifier work, by the width of time-frequency image
Degree divided by Amplitude maxima, realizes amplitude normalization；Data set is obtained after pretreated time-frequency image adds label.
Particularly, different with the output form of deep layer intensified learning Recognition with Recurrent Neural Network for Softmax layers, the mark of time-frequency image addition
Label form is also different: the corresponding label form of Softmax layers of output is the splicing of each signal component one-hot encoding label vector, deep layer
It is the signal component in current demand signal comprising each signal type that intensified learning Recognition with Recurrent Neural Network, which exports corresponding label form,
Number.
Step 3: design multi -components radar signal time-frequency image identifies network.
The pre-training feature extraction network of this algorithm design and each inside modules net of multicomponent data processing Classification and Identification network
Network framework is as shown in Figures 2 and 3.
(3.1) design convolutional neural networks are used for time-frequency image feature extraction, will input by pretreated time-frequency image
Convolutional neural networks pass sequentially through the network structure that 3 convolutional layers are followed by pond layer, by the output vector of the last one pond layer
Change obtains the feature vector of time-frequency image.The concrete operations of convolutional layer are: using convolution kernel as basic unit, traversal input picture into
The multiply-add operation of row obtains output characteristic image；The concrete operations of pond layer are: in a region, taking a specific value
As output valve, the pond mode in this algorithm is using average pond, i.e., using the average value in region as output valve；Vector quantization mistake
It is feature vector that journey, which is by multi-dimensional matrix characteristic expansion,.
(3.2) structure that full articulamentum is followed by Softmax layers is established special as sorter network unit by full articulamentum
Vector and the other mapping relations of class signal are levied, and are mapped as codomain in [0,1] using the Softmax layers of output by full articulamentum
The probability value of each signal type is exported the corresponding signal type of maximum probability as the label of sorter network unit, is constructed more
A sorter network unit respectively corresponds the label output of multicomponent data processing, divides the feature vector obtained in step (3.1)
Class, and with convolutional neural networks framework collectively as pre-training time-frequency image feature extraction network.
(3.3) design deep layer intensified learning Recognition with Recurrent Neural Network is used for multicomponent data processing Classification and Identification, alternative steps (3.2)
In sorter network unit, pass through multiple loop iteration assorting process export multicomponent data processing recognition result.In order to promote network
The accuracy rate of classification results, the input of network further include previous cycle step in addition to the feature vector that convolutional neural networks export
Recognition with Recurrent Neural Network exports the signal component number of each signal type, i.e. the control by kinds record of Recognition with Recurrent Neural Network before.
Step 4: training multi -components radar signal time-frequency image identifies network.
(4.1) after step 2, using the radar signal time-frequency image by image preprocessing as convolutional neural networks
Input, the tag along sort of each corresponding radar signal component of sorter network unit output, calculates output label and standard
The difference of two squares loss function of label, using error back propagation and stochastic gradient descent algorithm to time-frequency image characteristics extraction network
Pre-training is carried out, the parameter of convolutional neural networks is retained after training.
(4.2) convolutional neural networks and deep layer intensified learning Recognition with Recurrent Neural Network for then obtaining step (4.1) training
Keep convolutional neural networks model parameter constant, utilization and step to promote network training efficiency as new sorter network
(4.1) identical training set, individually training deep layer intensified learning Recognition with Recurrent Neural Network model parameter, training process use extensive chemical
The training method of habit.In order to promote network to the adaptability of sample, the classification results decision procedure based on rewards and punishments rule is introduced:
If recognition result belongs to current multi -components radar signal tag set, one positive reward value of network is fed back to, otherwise instead
It feeds one negative penalty value of network, decision process only considers that final multicomponent data processing determines the correctness of output result, does not examine
Consider the sequencing that each signal component determines result output；In order to avoid the overfitting problem that network is likely to occur, improve simultaneously
The training effectiveness of network introduces the memory playback training mode of intensified learning: by state vector input each time, Classification and Identification
Determine that result, rewards and punishments value are stored in the playback data base of certain memory space, it is next after playback data base space is full of
The data of secondary generation can cover first data in data base, and the data in data base are randomly selected in training process
It practises, upsets the correlation between training data.Finally retain convolutional neural networks and deep layer intensified learning Recognition with Recurrent Neural Network
All parameters.
(4.3) sample of test set is input to convolutional neural networks one by one and deep layer intensified learning Recognition with Recurrent Neural Network forms
Classifier in, classifier can recycle the signal type collection for repeatedly exporting and including in current sample, and contrast sample corresponds to standard letter
Number set of types eventually by the identification rate of precision that whole correct identification probability and each signal type is calculated and is recalled
Rate.
(4.4) network structure and parameter are improved according to the recognition effect test result of step (4.3), and to new network knot
Structure is trained, and recurrent network adjustment and training process, until the recognition effect test result of network reaches expected, completion is instructed
Practice.
Step 5: the network after training can realize that the intra-pulse modulation mode to multi -components radar signal time-frequency image is known
Not, simple component or overlapping multi -components radar signal is randomly generated, time frequency analysis and the pretreated time-frequency figure of time-frequency image will be passed through
As the input as network, final network exports the signal type collection for including in current radar signal by multiple loop iteration,
Realize the identification of multi -components radar emitter signal intra-pulse modulation mode.
Specifically, it is verified in the present embodiment by emulation:
8 kinds in total of radar modulated signal of emulation, type and parameter are as shown in table 1, and signal sampling points N=1024~
2048.Training set sample SNR ranges are that -6dB arrives 10dB, generate the simple component letter that 4000 parameters meet table 1 every 2dB
Number or overlapping multicomponent data processing sample at random, wherein the sample proportion of simple component signal and overlapping multicomponent data processing is 1:4, altogether
36000 samples are generated as training set.Test set sample SNR ranges are that -10dB arrives 10dB, the identical side with training set
Formula generates 44000 simple component sample of signal and overlapping multicomponent data processing sample respectively, and raw 88000 samples of common property are as test
Collection.
Table 1 emulates radar signal parameter list
Further, Fig. 4~Fig. 7 shows the neural network in the embodiment of the present invention after training under different signal-to-noise ratio
Recognition effect.When signal-to-noise ratio is -6dB, the average correct recognition rata of overlapping multi -components radar signal reaches 94.13%, and
The identification rate of precision and recall rate of each radar signal type reach 90% or more；And for simple component signal, be in signal-to-noise ratio-
When 6dB, the average correct recognition rata of simple component radar signal reaches 96.30%.
This shows that the algorithm is effectively, in low signal-to-noise ratio, it can be achieved that 8 classes overlap multi -components radar signal at random
Intrapulse modulation recognition, while the algorithm also has adaptability to the intrapulse modulation recognition of simple component radar signal.
Other step details and effect of the multi -components radar emitter signal intra-pulse modulation mode recognizer of the embodiment of the present invention
All be for a person skilled in the art it is known, in order to reduce redundancy, this will not be repeated here.
The above, specific implementation case only of the invention, scope of protection of the present invention is not limited thereto, any ripe
Those skilled in the art are known in technical specification of the present invention, modifications of the present invention or replacement all should be in the present invention
Within protection scope.
To sum up, the present invention is a kind of multi -components radar signal intra-pulse modulation side based on convolutional neural networks and intensified learning
Formula recognizer.The invention obtains the time-frequency image of radar signal first with Cohen class time-frequency distributions；It is tieed up followed by two dimension
Nanofiltration wave inhibits time-frequency image noise, and is adjusted to the size and amplitude of time-frequency image, then adds to time-frequency image
It tags, makes data set；It designs multi -components radar signal time-frequency image and identifies network, using training set to convolutional neural networks
Pre-training is carried out, extracts radar signal time-frequency image feature, and then carry out to multicomponent data processing sorter network using intensified learning
Training, is tested, and improve network structure and parameter according to test result using recognition effect of the test set to network；It will be to
The radar signal of identification inputs trained multi -components radar signal time-frequency image after time frequency analysis and image preprocessing and knows
Other network, network export the type for each signal component for including in radar signal automatically, complete identification.More points of the present invention
Radar Signal Recognition algorithm is measured in low signal-to-noise ratio, there is extensive radar signal type adaptation range and higher identification
Accuracy rate realizes the intrapulse modulation recognition of random overlapping multi -components radar signal, while the algorithm is to simple component radar
The intrapulse modulation recognition of signal also has adaptability.
Claims (4)
1. a kind of multi -components radar emitter signal intra-pulse modulation mode recognition methods characterized by comprising
Step 1: the simple component of several different intra-pulse modulation modes or the time-frequency image of overlapping multi -components radar signal, packet are obtained
LFM signal, MP signal, SFM signal, bpsk signal, 2FSK signal, 4FSK signal, EQFM signal and Frank signal are included, by this
A little signals are converted to radar signal time-frequency figure using the signal that time-frequency distributions receive radar receiver as sample signal
Picture；
Step 2: image processing algorithm is utilized, the radar signal time-frequency image that above-mentioned steps one obtain is pre-processed, is passed through
Two dimensional wiener filtering inhibits time-frequency image noise, is then adjusted simultaneously amplitude normalization to the size of time-frequency image, will pass through
Pretreated radar signal time-frequency image is as sample, and using the signal type for including in radar signal as label, production is obtained
Training set and test set；
Step 3: according to algorithm requirements, overall network extracts time-frequency image feature and carries out multi-tag according to the feature extracted
Classification and Identification designs the pre-training network based on convolutional neural networks and extracts radar signal time-frequency image feature, and designs and be based on
The multicomponent data processing sorter network of intensified learning obtains classification recognition result；
Step 4: being trained the network architecture designed in above-mentioned steps three according to training set obtained in above-mentioned steps two,
And the recognition effect of network is tested using the test set obtained in above-mentioned steps two, according to test result, improve network
Structure and parameter obtains the network of training completion；
Step 5: thunder to be identified is used as after the processing of above-mentioned steps two to random overlap signal any in above-mentioned steps one
Up to signal time-frequency image sample, input in the network that the training of above-mentioned steps four is completed, the network that training is completed can be according to input
The radar signal set of types for including in current demand signal is provided, realizes the Classification and Identification of multicomponent data processing.
2. a kind of multi -components radar emitter signal intra-pulse modulation mode recognition methods according to claim 1, which is characterized in that institute
State step 1, comprising:
Wherein, the time-frequency distributions method particularly includes:
To the signal x (t) received, using Cohen class time-frequency distributions, mathematic(al) representation are as follows:
In above formula, t and ω represent independent variable time and the angular frequency of time-frequency distributions, and φ (τ, v) is known as kernel function, the present invention according to
Radar signal and cross term propose to utilize two kinds of kernel functions in the characteristic distributions of fuzzy field for the signal of different chirp rates
The time-frequency distributions of radar signal are obtained respectively, and kernel function expression formula is respectively as follows:
In above formula, α, β, γ and ε are used to adjust the size of kernel function, obtain radar signal respectively under two kinds of kernel functions
Cohen class time-frequency distributions image, collectively as the time-frequency distributions information of current radar signal.
3. a kind of multi -components radar emitter signal intra-pulse modulation mode recognition methods according to claim 1, it is characterised in that: institute
It states in step 3, the planned network framework method particularly includes:
A) design convolutional neural networks framework is used for feature extraction, and convolutional neural networks framework is by 3 convolutional layers and 3 pond layers
Composition carries out vectorized process by exporting to network the last layer, obtains the feature vector of current time-frequency image；
B) structure that full articulamentum is followed by Softmax layers constructs multiple sorter network unit difference as sorter network unit
The label output of corresponding multicomponent data processing, classifies to the feature vector obtained in step a), and with convolutional neural networks frame
Structure is collectively as pre-training time-frequency image feature extraction network；
C) design deep layer intensified learning Recognition with Recurrent Neural Network is used for multicomponent data processing Classification and Identification, the classification net in alternative steps b)
The control by kinds of the feature vector and previous cycle step that obtain in step a) is recorded collectively as input, is passed through by network unit
Multiple loop iteration assorting process exports multicomponent data processing recognition result.
4. a kind of multi -components radar emitter signal intra-pulse modulation mode recognition methods according to claim 1, it is characterised in that: institute
It states in step 4, training and test method particularly includes:
D) after step 2, using the radar signal time-frequency image by image preprocessing as the input of convolutional neural networks,
The tag along sort of the corresponding radar signal component of each sorter network unit output, to time-frequency image characteristics extraction network into
Row pre-training retains the parameter of convolutional neural networks after training；
E) convolutional neural networks and deep layer intensified learning Recognition with Recurrent Neural Network then obtained step d) training are as new point
Class network keeps convolutional neural networks model parameter constant, utilizes instruction identical with step d) to promote network training efficiency
Practice collection, individually training deep layer intensified learning Recognition with Recurrent Neural Network model parameter, and it is strong with deep layer to retain final convolutional neural networks
Chemistry practises all parameters of Recognition with Recurrent Neural Network；
F) sample of test set is input to the classification of convolutional neural networks and deep layer intensified learning Recognition with Recurrent Neural Network composition one by one
In device, classifier can recycle the signal type collection for repeatedly exporting and including in current sample, and contrast sample corresponds to standard signal type
Collection, eventually by the identification rate of precision and recall rate that whole correct identification probability and each signal type is calculated；
G) network structure and parameter are improved according to the recognition effect test result of step f), and new network structure is instructed
Practice, recurrent network adjustment and training process, until the recognition effect test result of network reaches expected, completion is trained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910787759.1A CN110532932B (en) | 2019-08-26 | 2019-08-26 | Method for identifying multi-component radar signal intra-pulse modulation mode |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910787759.1A CN110532932B (en) | 2019-08-26 | 2019-08-26 | Method for identifying multi-component radar signal intra-pulse modulation mode |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110532932A true CN110532932A (en) | 2019-12-03 |
| CN110532932B CN110532932B (en) | 2022-12-13 |
Family
ID=68662779
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910787759.1A Active CN110532932B (en) | 2019-08-26 | 2019-08-26 | Method for identifying multi-component radar signal intra-pulse modulation mode |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110532932B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112103971A (en) * | 2020-09-01 | 2020-12-18 | 广西大学 | Vector reinforcement learning control method for power grid frequency modulation type flywheel energy storage system |
| CN112232120A (en) * | 2020-09-10 | 2021-01-15 | 中国人民解放军海军工程大学 | Radar radiation source signal classification system and method based on software radio |
| CN112329710A (en) * | 2020-11-25 | 2021-02-05 | 北京邮电大学 | Signal-to-noise ratio evaluation method and device based on unmanned aerial vehicle remote control signal |
| CN112491677A (en) * | 2020-11-06 | 2021-03-12 | 厦门大学 | CAN bus identification method and device based on physical layer characteristic fingerprint |
| CN113312996A (en) * | 2021-05-19 | 2021-08-27 | 哈尔滨工程大学 | Detection and identification method for aliasing short-wave communication signals |
| CN113376586A (en) * | 2021-06-03 | 2021-09-10 | 哈尔滨工程大学 | Method for constructing classification model of double-component radar signals |
| CN113962260A (en) * | 2021-10-21 | 2022-01-21 | 中国人民解放军空军航空大学 | Radar signal intelligent sorting method based on denoising depth residual error network |
| CN114021603A (en) * | 2021-10-25 | 2022-02-08 | 哈尔滨工程大学 | Radar signal modulation mode identification method based on model reparameterization |
| CN115712867A (en) * | 2022-11-03 | 2023-02-24 | 哈尔滨工程大学 | Multi-component radar signal modulation identification method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108388834A (en) * | 2017-01-24 | 2018-08-10 | 福特全球技术公司 | The object detection mapped using Recognition with Recurrent Neural Network and cascade nature |
| CN108805039A (en) * | 2018-04-17 | 2018-11-13 | 哈尔滨工程大学 | The Modulation Identification method of combination entropy and pre-training CNN extraction time-frequency image features |
| CN109086700A (en) * | 2018-07-20 | 2018-12-25 | 杭州电子科技大学 | Radar range profile's target identification method based on depth convolutional neural networks |
| US20190050729A1 (en) * | 2018-03-26 | 2019-02-14 | Intel Corporation | Deep learning solutions for safe, legal, and/or efficient autonomous driving |
| CN109726682A (en) * | 2018-12-29 | 2019-05-07 | 南京信息工程大学 | A kind of human motion recognition method towards weak label sensor data |
| CN109766791A (en) * | 2018-12-25 | 2019-05-17 | 哈尔滨工程大学 | A kind of Modulation Recognition of Communication Signal method based on self-encoding encoder |
| US10304208B1 (en) * | 2018-02-12 | 2019-05-28 | Avodah Labs, Inc. | Automated gesture identification using neural networks |
-
2019
- 2019-08-26 CN CN201910787759.1A patent/CN110532932B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108388834A (en) * | 2017-01-24 | 2018-08-10 | 福特全球技术公司 | The object detection mapped using Recognition with Recurrent Neural Network and cascade nature |
| US10304208B1 (en) * | 2018-02-12 | 2019-05-28 | Avodah Labs, Inc. | Automated gesture identification using neural networks |
| US20190050729A1 (en) * | 2018-03-26 | 2019-02-14 | Intel Corporation | Deep learning solutions for safe, legal, and/or efficient autonomous driving |
| CN108805039A (en) * | 2018-04-17 | 2018-11-13 | 哈尔滨工程大学 | The Modulation Identification method of combination entropy and pre-training CNN extraction time-frequency image features |
| CN109086700A (en) * | 2018-07-20 | 2018-12-25 | 杭州电子科技大学 | Radar range profile's target identification method based on depth convolutional neural networks |
| CN109766791A (en) * | 2018-12-25 | 2019-05-17 | 哈尔滨工程大学 | A kind of Modulation Recognition of Communication Signal method based on self-encoding encoder |
| CN109726682A (en) * | 2018-12-29 | 2019-05-07 | 南京信息工程大学 | A kind of human motion recognition method towards weak label sensor data |
Non-Patent Citations (4)
| Title |
|---|
| ZHIYU QU 等: "Radar Signal Intra-Pulse Modulation Recognition Based on Convolutional Neural Network and Deep Q-Learning Network", 《IEEE ACCESS》 * |
| ZHIYU QU 等: "Radar Signal Intra-Pulse Modulation Recognition Based on Convolutional Neural Network", 《IEEE ACCESS》 * |
| 刘全 等: "一种带探索噪音的深度循环Q网络", 《计算机学报》 * |
| 岳应娟 等: "变分模态分解与Cohen核结合的时频分布交叉项抑制方法", 《科学技术与工程》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112103971A (en) * | 2020-09-01 | 2020-12-18 | 广西大学 | Vector reinforcement learning control method for power grid frequency modulation type flywheel energy storage system |
| CN112103971B (en) * | 2020-09-01 | 2023-07-28 | 广西大学 | Vector reinforcement learning control method of power grid frequency modulation type flywheel energy storage system |
| CN112232120A (en) * | 2020-09-10 | 2021-01-15 | 中国人民解放军海军工程大学 | Radar radiation source signal classification system and method based on software radio |
| CN112491677A (en) * | 2020-11-06 | 2021-03-12 | 厦门大学 | CAN bus identification method and device based on physical layer characteristic fingerprint |
| CN112491677B (en) * | 2020-11-06 | 2022-10-14 | 厦门大学 | CAN bus identification method and device based on physical layer characteristic fingerprint |
| CN112329710A (en) * | 2020-11-25 | 2021-02-05 | 北京邮电大学 | Signal-to-noise ratio evaluation method and device based on unmanned aerial vehicle remote control signal |
| CN113312996A (en) * | 2021-05-19 | 2021-08-27 | 哈尔滨工程大学 | Detection and identification method for aliasing short-wave communication signals |
| CN113376586A (en) * | 2021-06-03 | 2021-09-10 | 哈尔滨工程大学 | Method for constructing classification model of double-component radar signals |
| CN113962260A (en) * | 2021-10-21 | 2022-01-21 | 中国人民解放军空军航空大学 | Radar signal intelligent sorting method based on denoising depth residual error network |
| CN114021603A (en) * | 2021-10-25 | 2022-02-08 | 哈尔滨工程大学 | Radar signal modulation mode identification method based on model reparameterization |
| CN114021603B (en) * | 2021-10-25 | 2024-05-31 | 哈尔滨工程大学 | Radar signal modulation mode identification method based on model re-parameterization |
| CN115712867A (en) * | 2022-11-03 | 2023-02-24 | 哈尔滨工程大学 | Multi-component radar signal modulation identification method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110532932B (en) | 2022-12-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110532932A (en) | A kind of multi -components radar emitter signal intra-pulse modulation mode recognition methods | |
| Daldal et al. | Automatic determination of digital modulation types with different noises using convolutional neural network based on time–frequency information | |
| CN107979554B (en) | Radio signal Modulation Identification method based on multiple dimensioned convolutional neural networks | |
| CN108282263B (en) | Coded modulation joint recognition methods based on one-dimensional depth residual error light weight network | |
| Wang et al. | Radar emitter recognition based on the short time Fourier transform and convolutional neural networks | |
| Li et al. | IncepTCN: A new deep temporal convolutional network combined with dictionary learning for strong cultural noise elimination of controlled-source electromagnetic data | |
| CN114564982B (en) | Automatic identification method for radar signal modulation type | |
| CN110378205A (en) | A kind of Complex Radar Radar recognition algorithm based on modified CNN network | |
| CN110555841A (en) | SAR image change detection method based on self-attention image fusion and DEC | |
| CN110309854A (en) | A kind of signal modulation mode recognition methods and device | |
| CN102279358A (en) | MCSKPCA based neural network fault diagnosis method for analog circuits | |
| CN109361635A (en) | Subsurface communication Modulation Mode Recognition method and system based on depth residual error network | |
| CN104103033A (en) | Image real-time processing method | |
| CN105117736A (en) | Polarized SAR image classification method based on sparse depth stack network | |
| CN115270872A (en) | Radar radiation source individual small sample learning and identifying method, system, device and medium | |
| CN114636975A (en) | LPI radar signal identification method based on spectrogram fusion and attention mechanism | |
| CN114943245A (en) | Automatic modulation recognition method and device based on data enhancement and feature embedding | |
| CN110048978A (en) | A kind of signal modulate method | |
| CN116797796A (en) | Signal identification method based on time-frequency analysis and deep learning under DRFM intermittent sampling | |
| CN111310719B (en) | Unknown radiation source individual identification and detection method | |
| CN114021458B (en) | Small sample radar radiation source signal identification method based on parallel prototype network | |
| CN106446804A (en) | ELM-based multi-granularity iris recognition method | |
| CN117420514A (en) | Electronic interference effect evaluation method for extracting radar parameter variation | |
| CN111797690A (en) | Optical fiber perimeter intrusion identification method and device based on wavelet neural network grating array | |
| CN115951315A (en) | Radar deception jamming identification method and system based on improved wavelet packet energy spectrum |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |