CN110045336A - Radar chaff recognition methods and device based on convolutional neural networks - Google Patents
Radar chaff recognition methods and device based on convolutional neural networks Download PDFInfo
- Publication number
- CN110045336A CN110045336A CN201910152766.4A CN201910152766A CN110045336A CN 110045336 A CN110045336 A CN 110045336A CN 201910152766 A CN201910152766 A CN 201910152766A CN 110045336 A CN110045336 A CN 110045336A
- Authority
- CN
- China
- Prior art keywords
- neural networks
- convolutional neural
- value
- networks model
- interference
- 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
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/36—Means for anti-jamming, e.g. ECCM, i.e. electronic counter-counter measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—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
Abstract
The radar chaff recognition methods and device that the present invention relates to a kind of based on convolutional neural networks, the method provides the Time And Space Parameters (such as orientation, distance, time) of target object using radar, the identification to various types interference is realized by intelligent algorithm based on convolutional neural networks, and operational commanding person is helped to identify interference.Have the characteristics that adaptable, speed of decision is fast, also, robustness and fault-tolerant ability with height.
Description
Technical field
The present embodiments relate to radar electronic warfare fields, and in particular to a kind of radar chaff knowledge based on convolutional neural networks
Other method and apparatus.
Background technique
Important military equipment of the radar as battlefield investigation monitoring and information gathering is that information and accurate system are obtained on battlefield
The Core equipment led in time, accurately and comprehensively obtains various target informations especially in vast area of operations, radar
Effect is not replaced.The normal work of radar is destroyed, the important information source of entire weapon system is also just destroyed.
Meanwhile in IT-based warfare, expansion, battle field information magnanimity, battlefield surroundings complicate, make before the space empty of battlefield
War object diversification, operation intensity increased dramatically, and the mass data that radar provides brings huge choose to the accurate commander of commander
War.
The existing method about radar chaff identification is the principle and mechanism angle generated from interference mostly at present, is led to
It crosses and the signal description word (such as frequency, pulsewidth, energy) of electromagnetic wave signal is received to radar to analyze, identify certain certain kinds
The interference of type.
However, inventors have found that the prior art has the following deficiencies: during carrying out innovation and creation
Experience and cognition of the tradition based on people, non intelligentization interference identification method can not successfully manage it is fast changing
Battlefield and mass data quickly and accurately obtain target information, Discrimination Radar disturbed condition from the radar detection data of magnanimity
Become problem.
Summary of the invention
The radar chaff recognition methods and device that the embodiment of the invention provides a kind of based on convolutional neural networks, to solve
At least one certainly above-mentioned technical problem.
In a first aspect, the embodiment of the present invention provides a kind of radar chaff recognition methods based on convolutional neural networks, comprising:
It obtains or simulation generates radar detection data, the radar detection data includes target object in a period
Interior multiple track points, each track points includes multiple attribute values of the target object in the multiple track points, described
Attribute value includes: orientation, distance, time, height, pitch angle；
Logarithm process is carried out to each attribute value in the multiple attribute value, and constructs a plurality of input data, every defeated
Entering data is the row vector that a data value and label value by higher-dimension after logarithm process forms；
Building includes the convolutional neural networks model of convolutional layer, pond layer and full articulamentum, and the convolutional layer is for mentioning
The feature of the input data is taken, the pond layer is used for the Feature Dimension Reduction extracted to the convolutional layer, and the full articulamentum is used
In classification of the realization to interference type；
It take a plurality of input data constructed as the training set of the convolutional neural networks model, the training convolutional Neural
Network model, and the parameter of the convolutional neural networks model is constantly adjusted, until the training of the convolutional neural networks model
Effect terminates to train when reaching default effect；
The parameter of convolutional neural networks model when terminating training is saved, and application has the convolutional Neural of the parameter
Network model identifies interference type.
Optionally, the method also includes:
When obtaining or simulating generation radar detection data, different label values is marked according to different types of interference,
The label value is any real number value, and the label value of same type interference is consistent, and the label value of different type interference is different.
Optionally, logarithm process is carried out to each attribute value in the multiple attribute value, comprising:
According to following formula, logarithm process is carried out to each attribute value in the multiple attribute value,
Y=10*lg (X/1000)
Wherein, X is untreated attribute value, and Y is the attribute value after logarithm process.
Optionally, the convolutional layer using one-dimensional convolution or uses two-dimensional convolution, the method also includes:
The size of convolution kernel, sliding step, port number, weight, biasing, fill method in the convolutional layer are initialized,
And tune ginseng is carried out according to the training effect of the convolutional neural networks model.
Optionally, the pond layer uses maximum value pond or average value pond, the method also includes:
The pond size of the pond layer, sliding step, weight and bias are initialized.
Optionally, the method also includes: it is initial to the number of plies of the full articulamentum, neuron number, weight, bias
Change.
Optionally, the method also includes:
To in the convolutional neural networks model batching data amount size, learning rate, the number of iterations, activation primitive,
Loss function, optimizer initialization, and constantly adjusted according to the training effect of the convolutional neural networks model.
Optionally, the method also includes:
Using new data set, the learning effect with the convolutional neural networks model of the parameter is tested.
Optionally, the interference type includes: noiseless, cheating interference, compacting interference.
Second aspect, the embodiment of the present invention provide a kind of radar chaff identification device based on convolutional neural networks, comprising:
Module is obtained, for obtaining or simulating generation radar detection data, the radar detection data includes object
The multiple track points of body in a period of time, each track points includes the more of the target object in the multiple track points
A attribute value, the attribute value include: orientation, distance, time, height, pitch angle；
Processing module for carrying out logarithm process to each attribute value in the multiple attribute value, and constructs a plurality of defeated
Enter data, every input data be the row that forms of a data value and a label value from higher-dimension after logarithm process to
Amount；
Module is constructed, it is described for constructing the convolutional neural networks model including convolutional layer, pond layer and full articulamentum
Convolutional layer is used to extract the feature of the input data, and the pond layer is used for the Feature Dimension Reduction extracted to the convolutional layer, institute
Full articulamentum is stated for realizing the classification to interference type；
Training module, for taking a plurality of input data constructed as the training set of the convolutional neural networks model, instruction
Practice the convolutional neural networks model, and constantly adjust the parameter of the convolutional neural networks model, until the convolutional Neural
The training effect of network model terminates to train when reaching default effect；
Application module terminates the parameter of convolutional neural networks model when training for saving, and application is with described
The convolutional neural networks model of parameter identifies interference type.
Beneficial effects of the present invention are as follows:
1, the present invention provides the Time And Space Parameters (such as orientation, distance, time) of target object using radar, based on convolution mind
The identification to various types interference is realized by intelligent algorithm through network, and operational commanding person is helped to identify interference.With adaptation
The feature that ability is strong, speed of decision is fast, also, robustness and fault-tolerant ability with height.
2, the present invention proposes to carry out logarithm process to the radar detection data with temporal characteristics, and logarithm process is a kind of
Nonlinear data processing method, also, the Singular variance between attribute value can reduce by logarithm process, keep variable more flat
Surely, so improve convolutional neural networks model study precision.
Detailed description of the invention
By reading the following detailed description of the preferred embodiment, various other advantages and benefits are common for this field
Technical staff will become clear.The drawings are only for the purpose of illustrating a preferred embodiment, and is not considered as to the present invention
Limitation.And throughout the drawings, the same reference numbers will be used to refer to the same parts.In the accompanying drawings:
Fig. 1 is a kind of process of radar chaff recognition methods based on convolutional neural networks provided in an embodiment of the present invention
Figure；
Fig. 2 is radar P aobvious figure when interference type is noiseless in the embodiment of the present invention；
Fig. 3 is radar P aobvious figure when interference type is compacting interference in the embodiment of the present invention；
Fig. 4 is radar P aobvious figure when interference type is cheating interference in the embodiment of the present invention；
Fig. 5 is the schematic diagram of convolutional neural networks model in the embodiment of the present invention；
Fig. 6 is a kind of signal of radar chaff identification device based on convolutional neural networks provided in an embodiment of the present invention
Figure.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
The radar chaff recognition methods based on convolutional neural networks that the embodiment of the invention discloses a kind of, can apply in electricity
Radar chaff identification is realized in radar electronic warfare field in the understanding of magnetic states gesture.It can be applied in aerospace field, equipment manufacturing neck
Domain.Using the embodiment of the invention discloses a kind of radar chaff recognition methods based on convolutional neural networks, convolutional Neural is constructed
Network model realizes the identification to radar chaff by the study to a large amount of radar detection datas.To fast and accurately be
Operational commanding provides decision assistant.
Based on this, An embodiment provides a kind of radar chaff identification side based on convolutional neural networks
Method.It is a kind of stream of radar chaff recognition methods based on convolutional neural networks provided in an embodiment of the present invention with reference to Fig. 1, Fig. 1
Cheng Tu.As shown in Figure 1, method includes the following steps:
S1, acquisition or simulation generate radar detection data, and the radar detection data includes target object at one
Between multiple track points in section, each track points includes multiple attribute values of the target object in the multiple track points,
The attribute value includes: orientation, distance, time, height, pitch angle；
S2, logarithm process is carried out to each attribute value in the multiple attribute value, and constructs a plurality of input data, every
Input data is the row vector that a data value and label value by higher-dimension after logarithm process forms；
S3, building include the convolutional neural networks model of convolutional layer, pond layer and full articulamentum, and the convolutional layer is used for
The feature of the input data is extracted, the pond layer is used for the Feature Dimension Reduction extracted to the convolutional layer, the full articulamentum
For realizing the classification to interference type；
S4, a plurality of input data to construct train the convolution for the training set of the convolutional neural networks model
Neural network model, and the parameter of the convolutional neural networks model is constantly adjusted, until the convolutional neural networks model
Training effect terminates to train when reaching default effect；
S5, the parameter for saving convolutional neural networks model when terminating training, and application has the convolution of the parameter
Neural network model identifies interference type.
The targeted data of the present invention are detection data provided by gadget (i.e. radar detection data), detections of radar
Data generally include the attribute values such as the orientation, distance, time of target object.When gadget is interfered, gadget institute
The detection data of offer is interference data.It is this that the P for being reflected in gadget is interfered to show on figure, it is usually expressed as compressed dry
It disturbs, cheating interference etc..And different interference types, the data characteristics presented are also different.Wherein, interference type packet
It includes: noiseless, cheating interference, compacting interference.
It is radar P aobvious figure when interference type is noiseless in the embodiment of the present invention with reference to Fig. 2, Fig. 2.Radar is in noiseless feelings
Under condition, its motion profile can be showed on the aobvious figure of P when target object by detecting.This track is according to temporal order, with certain
Time interval, it is discrete, show in dots, while radar can export the specific object value of each point, including side
Position, distance, time etc., as shown in table 1.
1 radar detection data of table (noiseless)
| Lot number | Distance (m) | Orientation (degree) | Pitching (degree) | Highly (m) | Time (ms) |
| 9 | 277299 | 266.82 | 0.00 | 7020 | 37942471 |
| 9 | 275315 | 266.74 | 0.00 | 6510 | 37952471 |
| 9 | 273351 | 266.83 | 0.00 | 6730 | 37962471 |
| 10 | 135050 | 169.51 | 0.00 | 0 | 37949767 |
| 10 | 134100 | 169.58 | 0.00 | 0 | 37959767 |
| … | … | … | … | … | … |
When radar is interfered, according to the difference of interference type, with reference to Fig. 3 and Fig. 4, target object, which shows figure in P, to be in
Reveal different forms, with reference to table 2 and table 3, the data that radar provides also can be different.Fig. 3 is done in the embodiment of the present invention
Disturb radar P aobvious figure when type is compacting interference.Fig. 4 is radar P aobvious figure when interference type is cheating interference in the embodiment of the present invention.
2 radar detection data of table (compacting interference)
| Lot number | Distance (m) | Orientation (degree) | Pitching (degree) | Highly (m) | Time (ms) |
| 97 | 108709 | 172.40 | 0.00 | 600 | 38826419 |
| 78 | 103342 | 192.74 | 0.00 | 7760 | 38827014 |
| 111 | 99723 | 202.20 | 0.00 | 510 | 38827245 |
| 98 | 103204 | 245.91 | 0.00 | 720 | 38828454 |
| 99 | 99649 | 277.50 | 0.00 | 670 | 38829318 |
| … | … | … | … | … | … |
3 radar detection data of table (cheating interference)
| Lot number | Distance (m) | Orientation (degree) | Pitching (degree) | Highly (m) | Time (ms) |
| 307 | 6353.57 | 9.73 | 0 | 5549 | 38001022 |
| 307 | 5909.41 | 352.41 | 0 | 6012 | 38014189 |
| 315 | 4126.25 | 79.48 | 0 | 5966 | 38018560 |
| 313 | 9550.73 | 42.78 | 0 | 6190 | 37972256 |
| 315 | 36.783 | 45 | 0 | 7870 | 37988563 |
| … | … | … | … | … | … |
In view of many times can not be artificial direct go to identify our radar whether by dry from a lot of numbers
Which kind of disturb, by interference, the present invention proposes a kind of radar chaff recognition methods based on convolutional neural networks, by mentioning to radar
The data (interfering containing noiseless, different type) of confession are learnt, and achieve the purpose that identification interference.This method comprises:
(1) it obtains or simulates and generate radar detection data.Radar detection data should be comprising target object sometime
Multiple track points in section.Each track points should include the time-space attributes value such as orientation, distance, time of target object.
Assuming that a track is made of 10 track points, the attribute value that each track points include has orientation, distance, time
(as shown in table 4).
4 one track datas of table
[10.771,12740.4638,12701,17.3892,11705.7059,12711,28.7081,8720.678,
12721,51.8093,5329.7534,12731,93.7079,3505.7183,12741,123.4339,2853.129,
12751,100.1716,1283.7975,12761,9.8649,3344.7384,12771,4.2846,7670.1518,12781,
6.4538,11241.9583,12791]
(2) when obtaining or simulating generation radar detection data, different labels is marked according to different types of interference
Value.The label value is any real number value, and the label value of same type interference is consistent, and the label value of different type interference is different.
Convolutional neural networks model of the present invention is a kind of learning method for having supervised, therefore for convolution mind
It has to through training set required in network model with label value, i.e. every input data requires to indicate interference
Type.When obtaining or simulating generation radar detection data, different label values, label are marked according to different types of interference
Value can be any real number value, but guarantee that the label value of same type interference is consistent, and the label value of different type interference is different
(as shown in table 5).
5 label value list of table
| Interference type | Label value |
| It is noiseless | 1 |
| Cheating interference | 2 |
| Compacting interference | 3 |
| … | … |
(3) logarithm process is carried out to radar detection data.In the specific implementation, to each category in the multiple attribute value
Property value carry out logarithm process, comprising: according to following formula, each attribute value in the multiple attribute value is carried out at logarithm
Reason, wherein, X is untreated attribute value to Y=10*lg (X/1000), and Y is the attribute value after logarithm process.
In view of dimension difference is big between different attribute value and the timing of radar detection data, the present invention is to attribute
The method that value uses logarithm process, specifically: Y=10*lg (X/1000), wherein X is untreated attribute value, and Y is warp pair
Number treated data value.Reduce the Singular variance between attribute value by logarithm process, keep variable more stable, and then improves convolution
The study precision of neural network model.
(4) a plurality of input data is constructed.The length of convolutional neural networks model needs input data is consistent, examines for radar
The track measured can choose a suitable value (such as 10 track points) as convolutional neural networks mode input data
Length.When certain track data is less than 10 track points, can be filled by the methods of zero padding；When certain track data
When greater than 10 track points, it can be rejected by taking the methods of median.Then whole attribute values of 10 track points are spliced
At a vector, and using this vector as the input data of convolutional neural networks model.Such as: assuming that each track points include
Attribute value has orientation, away from discrete time, then the track data being made of for one 10 track points, by each track points packet
Three attribute values included are stitched together, and just constitute the row vector after logarithm process of one 30 dimension.Furthermore an institute is added again
Belong to the label value of interference type, thus, every input data is a data value and one after logarithm process by 30 dimensions
The row vector (as shown in table 6) of a label value composition.
6 one input datas of table
[33.2372,9.0881,11.9997,44.0452,8.6259,12.0025,77.5632,7.4687,
12.0052,124.7794,8.2203,12.008,147.0685,9.6941,12.0107,154.6168,10.1555,
12.0134,153.9756,9.3977,12.0162,137.7946,7.0856,12.0189,70.7024,5.609,
12.0216,36.0962,8.1177,12.0243,3]
(5) building includes the convolutional Neural of convolution (convolution) layer, pond (pooling) layer and full articulamentum
Network model.The convolutional layer is used to extract the feature of the input data, and the pond layer is used to extract the convolutional layer
Feature Dimension Reduction, the full articulamentum is for realizing the classification to interference type.It with reference to Fig. 5, Fig. 5 is rolled up in the embodiment of the present invention
The schematic diagram of product neural network model.
(6) convolutional layer that a plurality of input data constructed inputs (input) convolutional neural networks model one by one is carried out
Feature extraction.
In the specific implementation, the convolutional layer using one-dimensional convolution or uses two-dimensional convolution, to convolution in the convolutional layer
The size of core, sliding step, port number, weight, biasing, fill method initialization, and according to the convolutional neural networks model
Training effect carry out tune ginseng.
Convolutional layer is used to extract the feature of input data.One-dimensional convolution can be used when convolution, two-dimensional convolution can also be used,
If using two-dimensional convolution, it is necessary to input vector be carried out two-dimensional map, row vector duplication, self-organizing feature map can be used
The methods of neural network (SOM) is mapped.
Assuming that needing the row vector that obtained above 30 tie up being mapped to two dimension, mapping side using the method for two-dimensional convolution
Method is replicated using row vector, finally obtains the vector matrix of 30*30.The size of convolution kernel, sliding step, channel in convolutional layer
Number, weight, biasing, fill method etc. first can be initialized voluntarily, subsequent to be carried out according to the training effect of convolutional neural networks model
Adjust ginseng.Assuming that convolution kernel size is 3*3, sliding step 1*1, port number 32, weight and bias using one layer of convolutional layer
It is all made of random initializtion method, fill method use ' SAME ' filling obtains the spy of new 30*30*32 after convolution operation
Levy data.
(6) the pond layer of the feature input convolutional neural networks model extracted convolutional layer carries out dimensionality reduction.
In the specific implementation, the pond layer uses maximum value pond or average value pond, the method also includes: to institute
State pond size, sliding step, weight and the bias initialization of pond layer.
Pond layer is used to the Feature Dimension Reduction extracted to convolutional layer, and the purpose of dimensionality reduction is to retain main feature, reduce convolution mind
Parameter through network model, and then improve the generalization ability of convolutional neural networks model.Can be used in the layer of pond maximum value pond,
The methods of average value pond, pond size, sliding step, weight and bias can be initialized voluntarily, and fill method can be used '
SAME ' method or ' VALID ' method.Assuming that using maximum value pond method, pond size and step-length are 2*2, weight and partially
It sets value and uses random initializtion method, using ' SAME ' fill method.After pondization operation, the characteristic of 15*15 is obtained
(feature data)。
(7) feature after the layer dimensionality reduction of pond is inputted full articulamentum to learn, realizes disturbance ecology, output
(output) interference type.
In the specific implementation, the method also includes:
To the number of plies of the full articulamentum, neuron number, weight, bias initialization.
Full articulamentum realizes the classification to interference type.Full articulamentum passes through to the spy extracted after convolutional layer and pond layer
Sign is learnt, and realizes classification of disturbance, final output interference type.The full articulamentum number of plies, neuron number, weight, biasing
The parameters such as value can be initialized voluntarily.Assuming that setting nerve cell layer is 2 layers, neuron number is 128, weight and bias
Using random initializtion.
(8) various parameters in convolutional neural networks model are constantly adjusted, until the instruction of convolutional neural networks model
Practicing when effect reaches ideal effect terminates to train.
In the specific implementation, the method also includes:
To in the convolutional neural networks model batching data amount size, learning rate, the number of iterations, activation primitive,
Loss function, optimizer initialization, and constantly adjusted according to the training effect of the convolutional neural networks model.
(9) the convolutional neural networks model measurement to terminating when training.
In the specific implementation, the method also includes:
Using new data set, the learning effect with the convolutional neural networks model of the parameter is tested.
(10) parameter of convolutional neural networks model when terminating training is saved, and application has the convolutional Neural net of the parameter
Network model identifies interference type.
Beneficial effects of the present invention are as follows:
1, the present invention provides the Time And Space Parameters (such as orientation, distance, time) of target object using radar, based on convolution mind
The identification to various types interference is realized by intelligent algorithm through network, and operational commanding person is helped to identify interference.With adaptation
The feature that ability is strong, speed of decision is fast, also, robustness and fault-tolerant ability with height.
2, the present invention proposes to carry out logarithm process to the radar detection data with temporal characteristics, and logarithm process is a kind of
Nonlinear data processing method can reduce the Singular variance between attribute value by logarithm process, keep variable more stable, in turn
Improve the study precision of convolutional neural networks model.
Based on the same inventive concept, it is dry that An embodiment provides a kind of radars based on convolutional neural networks
Disturb identification device.As shown in fig. 6, the device includes:
Module is obtained, for obtaining or simulating generation radar detection data, the radar detection data includes object
The multiple track points of body in a period of time, each track points includes the more of the target object in the multiple track points
A attribute value, the attribute value include: orientation, distance, time, height, pitch angle；
Processing module for carrying out logarithm process to each attribute value in the multiple attribute value, and constructs a plurality of defeated
Enter data, every input data be the row that forms of a data value and a label value from higher-dimension after logarithm process to
Amount；
Module is constructed, it is described for constructing the convolutional neural networks model including convolutional layer, pond layer and full articulamentum
Convolutional layer is used to extract the feature of the input data, and the pond layer is used for the Feature Dimension Reduction extracted to the convolutional layer, institute
Full articulamentum is stated for realizing the classification to interference type；
Training module, for taking a plurality of input data constructed as the training set of the convolutional neural networks model, instruction
Practice the convolutional neural networks model, and constantly adjust the parameter of the convolutional neural networks model, until the convolutional Neural
The training effect of network model terminates to train when reaching default effect；
Application module terminates the parameter of convolutional neural networks model when training for saving, and application is with described
The convolutional neural networks model of parameter identifies interference type.
Since the radar chaff identification device based on convolutional neural networks that the present embodiment is introduced is that can execute this hair
The device of the radar chaff recognition methods based on convolutional neural networks in bright embodiment, so based on institute in the embodiment of the present invention
The radar chaff recognition methods based on convolutional neural networks introduced, those skilled in the art can understand the present embodiment
The specific embodiment of radar chaff identification device based on convolutional neural networks and its various change form, so right at this
In the radar chaff identification device based on convolutional neural networks how to realize in the embodiment of the present invention based on convolutional Neural net
The radar chaff recognition methods of network is no longer discussed in detail.As long as those skilled in the art implement to be based in the embodiment of the present invention
Device used by the radar chaff recognition methods of convolutional neural networks belongs to the range to be protected of the application.
In the instructions provided here, numerous specific details are set forth.It is to be appreciated, however, that the implementation of the disclosure
Example can be practiced without these specific details.In some instances, well known method, structure is not been shown in detail
And technology, so as not to obscure the understanding of this specification.
Similarly, it should be understood that in order to simplify the disclosure and help to understand one or more of the various inventive aspects,
Above in the description of the exemplary embodiment of the disclosure, each feature of the disclosure is grouped together into single implementation sometimes
In example, figure or descriptions thereof.However, the disclosed method should not be interpreted as reflecting the following intention: i.e. required to protect
The disclosure of shield requires features more more than feature expressly recited in each claim.More precisely, as following
Claims reflect as, inventive aspect is all features less than single embodiment disclosed above.Therefore,
Thus the claims for following specific embodiment are expressly incorporated in the specific embodiment, wherein each claim itself
All as the separate embodiments of the disclosure.
Claims (9)
1. a kind of radar chaff recognition methods based on convolutional neural networks characterized by comprising
Obtain or simulation generate radar detection data, the radar detection data include target object in a period of time
Multiple track points, each track points includes multiple attribute values of the target object, the attribute in the multiple track points
Value includes: orientation, distance, time, height, pitch angle；
Logarithm process is carried out to each attribute value in the multiple attribute value, and constructs a plurality of input data, every input number
According to being row vector that a data value and label value by higher-dimension after logarithm process forms；
Building includes the convolutional neural networks model of convolutional layer, pond layer and full articulamentum, and the convolutional layer is for extracting institute
The feature of input data is stated, the pond layer is used for the Feature Dimension Reduction extracted to the convolutional layer, and the full articulamentum is for real
Now to the classification of interference type；
It take a plurality of input data constructed as the training set of the convolutional neural networks model, the training convolutional neural networks
Model, and the parameter of the convolutional neural networks model is constantly adjusted, until the training effect of the convolutional neural networks model
Terminate to train when reaching default effect；
The parameter of convolutional neural networks model when terminating training is saved, and application has the convolutional neural networks of the parameter
Model identifies interference type.
2. the method according to claim 1, wherein the method also includes:
When obtaining or simulating generation radar detection data, different label values is marked according to different types of interference, it is described
Label value is any real number value, and the label value of same type interference is consistent, and the label value of different type interference is different.
3. the method according to claim 1, wherein being carried out pair to each attribute value in the multiple attribute value
Number processing, comprising:
According to following formula, logarithm process, Y=10*lg (X/1000) are carried out to each attribute value in the multiple attribute value
Wherein, X is untreated attribute value, and Y is the attribute value after logarithm process.
4. the method according to claim 1, wherein the convolutional layer is using one-dimensional convolution or using two dimension volume
Product, the method also includes:
To the size of convolution kernel, sliding step, port number, weight, biasing, fill method initialization, and root in the convolutional layer
Tune ginseng is carried out according to the training effect of the convolutional neural networks model.
5. the method according to claim 1, wherein the pond layer uses maximum value pond or average value pond
Change, the method also includes:
The pond size of the pond layer, sliding step, weight and bias are initialized.
6. the method according to claim 1, wherein the method also includes: to the number of plies of the full articulamentum,
Neuron number, weight, bias initialization.
7. the method according to claim 1, wherein the method also includes:
To the batching data amount size in the convolutional neural networks model, learning rate, the number of iterations, activation primitive, loss
Function, optimizer initialization, and constantly adjusted according to the training effect of the convolutional neural networks model.
8. the method according to claim 1, wherein the interference type includes: noiseless, cheating interference, pressure
System interference.
9. a kind of radar chaff identification device based on convolutional neural networks characterized by comprising
Module is obtained, for obtaining or simulating generation radar detection data, the radar detection data includes that target object exists
Multiple track points in one period, each track points includes multiple categories of the target object in the multiple track points
Property value, the attribute value includes: orientation, distance, time, height, pitch angle；
Processing module for carrying out logarithm process to each attribute value in the multiple attribute value, and constructs a plurality of input number
It is the row vector that a data value and a label value by higher-dimension after logarithm process forms according to, every input data；
Module is constructed, for constructing the convolutional neural networks model including convolutional layer, pond layer and full articulamentum, the convolution
For layer for extracting the feature of the input data, the pond layer is used for the Feature Dimension Reduction extracted to the convolutional layer, described complete
Articulamentum is for realizing the classification to interference type；
Training module, for taking a plurality of input data constructed as the training set of the convolutional neural networks model, training institute
Convolutional neural networks model is stated, and constantly adjusts the parameter of the convolutional neural networks model, until the convolutional neural networks
The training effect of model terminates to train when reaching default effect；
Application module terminates the parameter of convolutional neural networks model when training for saving, and application has the parameter
Convolutional neural networks model identify interference type.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910152766.4A CN110045336B (en) | 2019-02-28 | 2019-02-28 | Convolutional neural network-based radar interference identification method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910152766.4A CN110045336B (en) | 2019-02-28 | 2019-02-28 | Convolutional neural network-based radar interference identification method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110045336A true CN110045336A (en) | 2019-07-23 |
| CN110045336B CN110045336B (en) | 2021-05-25 |
Family
ID=67274360
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910152766.4A Active CN110045336B (en) | 2019-02-28 | 2019-02-28 | Convolutional neural network-based radar interference identification method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110045336B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN110927706A (en) * | 2019-12-10 | 2020-03-27 | 电子科技大学 | Convolutional neural network-based radar interference detection and identification method |
| CN113311397A (en) * | 2021-05-25 | 2021-08-27 | 西安电子科技大学 | Large array rapid self-adaptive anti-interference method based on convolutional neural network |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0845685A3 (en) * | 1996-11-29 | 1999-05-26 | Alcatel | Device and method for automatic target classification |
| CN106443598A (en) * | 2016-12-08 | 2017-02-22 | 中国人民解放军海军航空工程学院 | Convolutional neural network based cooperative radar network track deception jamming discrimination method |
| CN107748356A (en) * | 2017-11-06 | 2018-03-02 | 西安电子科技大学 | Phase-coded signal deception jamming recognition methods based on multiple resemblance Coefficient |
| CN108229404A (en) * | 2018-01-09 | 2018-06-29 | 东南大学 | A kind of radar echo signal target identification method based on deep learning |
| CN108919203A (en) * | 2018-05-21 | 2018-11-30 | 西安电子科技大学 | A kind of recognition methods and system of radar active interference |
| CN109031221A (en) * | 2018-08-20 | 2018-12-18 | 中国人民解放军海军航空大学 | Data Layer resists intensive false target jamming profile method based on the radar of various dimensions feature extraction |
| CN109061577A (en) * | 2018-07-05 | 2018-12-21 | 西安电子科技大学 | A kind of recognition methods of different types of covering type interference and Deceiving interference |
-
2019
- 2019-02-28 CN CN201910152766.4A patent/CN110045336B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0845685A3 (en) * | 1996-11-29 | 1999-05-26 | Alcatel | Device and method for automatic target classification |
| CN106443598A (en) * | 2016-12-08 | 2017-02-22 | 中国人民解放军海军航空工程学院 | Convolutional neural network based cooperative radar network track deception jamming discrimination method |
| CN107748356A (en) * | 2017-11-06 | 2018-03-02 | 西安电子科技大学 | Phase-coded signal deception jamming recognition methods based on multiple resemblance Coefficient |
| CN108229404A (en) * | 2018-01-09 | 2018-06-29 | 东南大学 | A kind of radar echo signal target identification method based on deep learning |
| CN108919203A (en) * | 2018-05-21 | 2018-11-30 | 西安电子科技大学 | A kind of recognition methods and system of radar active interference |
| CN109061577A (en) * | 2018-07-05 | 2018-12-21 | 西安电子科技大学 | A kind of recognition methods of different types of covering type interference and Deceiving interference |
| CN109031221A (en) * | 2018-08-20 | 2018-12-18 | 中国人民解放军海军航空大学 | Data Layer resists intensive false target jamming profile method based on the radar of various dimensions feature extraction |
Non-Patent Citations (4)
| Title |
|---|
| 刘安长: "《我国财政政策对宏观调控目标的传导效应研究》", 31 January 2017, 光明日报出版社 * |
| 刘建洋: "基于时频域分析的电子干扰识别方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 尹鹏: "雷达有源压制干扰自适应感知方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 陆晓明: "基于神经网络的雷达杂波抑制", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| 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 |
| CN110927706A (en) * | 2019-12-10 | 2020-03-27 | 电子科技大学 | Convolutional neural network-based radar interference detection and identification method |
| CN113311397A (en) * | 2021-05-25 | 2021-08-27 | 西安电子科技大学 | Large array rapid self-adaptive anti-interference method based on convolutional neural network |
| CN113311397B (en) * | 2021-05-25 | 2023-03-10 | 西安电子科技大学 | Large array rapid self-adaptive anti-interference method based on convolutional neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110045336B (en) | 2021-05-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110045336A (en) | Radar chaff recognition methods and device based on convolutional neural networks | |
| CN110097103A (en) | Based on the semi-supervision image classification method for generating confrontation network | |
| CN103955702B (en) | SAR image terrain classification method based on depth RBF network | |
| CN110348399B (en) | Hyperspectral intelligent classification method based on prototype learning mechanism and multidimensional residual error network | |
| CN110363151A (en) | Based on the controllable radar target detection method of binary channels convolutional neural networks false-alarm | |
| CN107886123A (en) | A kind of synthetic aperture radar target identification method based on auxiliary judgement renewal learning | |
| Duan et al. | Multichannel pulse-coupled neural network-based hyperspectral image visualization | |
| CN104459668A (en) | Radar target recognition method based on deep learning network | |
| CN104298999B (en) | EO-1 hyperion feature learning method based on recurrence autocoding | |
| CN110378205A (en) | A kind of Complex Radar Radar recognition algorithm based on modified CNN network | |
| CN108520202A (en) | Confrontation robustness image characteristic extracting method based on variation spherical projection | |
| CN105023043A (en) | AOD-based PM2.5 inversion model for Hangzhou region | |
| CN110045335A (en) | Based on the Radar Target Track recognition methods and device for generating confrontation network | |
| CN108664994A (en) | A kind of remote sensing image processing model construction system and method | |
| CN109754079A (en) | A method of catastrophic forgetting is overcome based on parameter importance | |
| CN106980831A (en) | Based on self-encoding encoder from affiliation recognition methods | |
| Zhang et al. | VGM-RNN: HRRP sequence extrapolation and recognition based on a novel optimized RNN | |
| CN103456017B (en) | Image partition method based on the semi-supervised weight Kernel fuzzy clustering of subset | |
| CN107169188A (en) | A kind of multidimensional multivariable non-gaussian spatial random field analogy method | |
| CN109285168A (en) | A kind of SAR image lake boundary extraction method based on deep learning | |
| CN111539488B (en) | Imaging distortion resisting target classification method for ultra-narrow pulse radar under complex dynamic track | |
| CN114332545A (en) | Image data classification method and device based on low-bit pulse neural network | |
| Xie et al. | High-efficiency and low-energy ship recognition strategy based on spiking neural network in SAR images | |
| CN104699781A (en) | Specific absorption rate image retrieval method based on double-layer anchor chart hash | |
| CN107516069A (en) | Target identification method based on geometry reconstruction and multiscale analysis |
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 |