CN116205271A - Method for generating opposite network deep learning direction finding under small sample condition - Google Patents
Method for generating opposite network deep learning direction finding under small sample condition Download PDFInfo
- Publication number
- CN116205271A CN116205271A CN202310130162.6A CN202310130162A CN116205271A CN 116205271 A CN116205271 A CN 116205271A CN 202310130162 A CN202310130162 A CN 202310130162A CN 116205271 A CN116205271 A CN 116205271A
- Authority
- CN
- China
- Prior art keywords
- sample
- data
- direction finding
- discriminator
- input
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000013135 deep learning Methods 0.000 title claims abstract description 17
- 238000012549 training Methods 0.000 claims abstract description 17
- 238000013507 mapping Methods 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims description 10
- 238000011176 pooling Methods 0.000 claims description 10
- 238000005070 sampling Methods 0.000 claims description 9
- 230000008485 antagonism Effects 0.000 claims description 8
- 230000005855 radiation Effects 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 5
- 230000003321 amplification Effects 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 4
- 238000013145 classification model Methods 0.000 claims description 3
- 238000013527 convolutional neural network Methods 0.000 abstract description 12
- 238000013528 artificial neural network Methods 0.000 abstract description 2
- 239000011159 matrix material Substances 0.000 description 5
- 239000013598 vector Substances 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013144 data compression Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000000528 statistical test Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for generating a countering network deep learning direction finding under a small sample condition, relates to the technical field of radio direction finding, and solves the technical problems that in the prior art, the difficulty of constructing a complete training sample library is high, and the number of samples which can be acquired under the condition of an actual application environment is smaller than the required training sample number; the method comprises the following steps: constructing and generating an countermeasure network; the generator and the discriminator mutually resist to generate a generated sample approaching to the real distribution; preprocessing data; extracting the relative amplitude and covariance characteristics of each array element to be used as input data of CNN; realizing nonlinear feature mapping of data through convolution operation; dividing the input feature map into a plurality of non-overlapping regions; small sample CNN direction finding; the method has the advantages that the small sample condition aiming at the practical application environment is realized, the generated sample which approximates to the real sample distribution is generated by adopting the generation countermeasure network, the training sample data set is effectively expanded, and the convolution neural network direction finding model is utilized, so that the high-precision direction finding is realized.
Description
Technical Field
The invention belongs to the field of radio direction finding, relates to a neural network technology, and in particular relates to a method for generating a counter network deep learning direction finding under a small sample condition.
Background
The direction finding of the radiation source target is an electronic reconnaissance device, and the arrival direction of the target signal is estimated by utilizing the received target signal through signal processing. The method mainly comprises amplitude method direction finding, phase method direction finding, space spectrum estimation direction finding and the like for the radiation source target direction finding, and the methods are all traditional non-intelligent direction finding methods. At present, an intelligent direction finding method has become a research hot spot. The patent 'random array angle of arrival estimation method based on deep learning' proposes an array signal direction finding method based on a Convolutional Neural Network (CNN), and the method aims at any given multi-element array, and realizes rapid and high-precision direction finding of a radiation source target on the basis of deep learning by extracting characteristic information such as phase differences among array elements of the array element sampling data.
However, the number of training samples required by the CNN direction finding method is huge, because, first, the method is a direction finding method based on classification recognition, and if the required direction finding range is 360 degrees, the angular resolution is 0.1 degree, and the target number is unknown, the angular classification number is The number of training samples is huge at this time; secondly, when the wide-band direction finding requirement is met, a sample set needs to be built for each carrier frequency to train, and training samples are obviously increased. Therefore, the method has great difficulty in constructing a complete training sample library, and the number of samples which can be acquired under the condition of the practical application environment is far smaller than the required training sample number.
Therefore, a method for generating a countering network deep learning direction finding under the condition of a small sample is provided.
Disclosure of Invention
The present invention aims to solve at least one of the technical problems existing in the prior art. Therefore, the invention provides the method for generating the opposite network deep learning direction finding under the condition of the small sample, and the method for generating the opposite network deep learning direction finding under the condition of the small sample solves the problems that the difficulty in constructing a complete training sample library in the prior art is extremely high, and the number of samples which can be acquired under the condition of the practical application environment is far smaller than the required training sample number.
To achieve the above object, an embodiment according to a first aspect of the present invention proposes a method for generating a countering network deep learning direction finding under a small sample condition, including the steps of:
step one: constructing and generating an countermeasure network by using a nonlinear mapping function; wherein the generation countermeasure network comprises a generator G and a discriminator D;
step two: the generator G and the discriminator D reach Nash equilibrium after mutual antagonism and mutual promotion, and generate a generated sample, which comprises the following steps:
inputting the noise sequence into a generator to generate a generated sample conforming to the real data distribution;
inputting the generated sample and the real sample into a discriminator;
the generator G is fixed, a method of minimizing cross entropy is adopted to train the discriminator D, so that the discriminator D can effectively distinguish real samples and generate samples, and the loss function of the discriminator D is as follows:
in the method, in the process of the invention,wherein x is a real sample, and D (x) is the probability of judging the input real sample x by the judging device D;
wherein z represents a noise sequence input to the generator G, G (z) represents a generated sample, and D (G (z)) represents a probability that the generated sample is discriminated as a true sample after passing through the discriminator D;
the generator G learns the formal data distribution P data Deception discriminant D, fixed discriminant D, rawThe generator G increases the probability that the generated sample is judged to be a true sample, and when the generated sample is inputted, the output of the discriminator D is as close to 1 as possible, and the loss function of the generator G is:
step three: preprocessing the radiation sources received by each array element to obtain sampling data; wherein the preprocessing comprises power amplification, down-conversion and data sampling processing;
step four: taking the relative amplitude and covariance characteristics of each array element as the input data of CNN;
step five: the convolution layer carries out convolution operation with input data through different convolution kernels to realize nonlinear feature mapping of the data, so that feature extraction of the data is realized;
step six: the pooling layer divides the input feature map into a plurality of non-overlapping areas;
step seven: based on generating the antagonism network, the training data set under the condition of the small sample is expanded.
Preferably, the discriminator D is a classification model.
Preferably, the output of the discriminator D is the probability that the input sample is a true sample;
when the input real sample is output as close to 1 as possible;
when the input-generated samples are outputs as close to 0 as possible.
Preferably, the convolution layer adopts a local connection mode.
Preferably, the pooling layer compresses the data through a downsampling operation.
Compared with the prior art, the invention has the beneficial effects that:
the invention creates an antagonism network by using a nonlinear mapping function construction; the generator G and the discriminator D achieve Nash equilibrium after mutual antagonism and mutual promotion, and generate a generated sample; the method has the advantages that the small sample condition aiming at the practical application environment is realized, and the generation of the generated sample approaching to the real sample distribution is generated by adopting the generation countermeasure network;
preprocessing the radiation sources received by each array element to obtain sampling data; taking the relative amplitude and covariance characteristics of each array element as the input data of CNN; the convolution layer carries out convolution operation with input data through different convolution kernels to realize nonlinear feature mapping of the data, so that feature extraction of the data is realized; the pooling layer divides the input feature map into a plurality of non-overlapping areas; expanding a training data set under a small sample condition based on the generated antagonism network; the training sample data set is effectively expanded, and then the convolutional neural network direction finding model is utilized, so that stable high-precision direction finding is realized on the basis of learning and training.
Drawings
FIG. 1 is a schematic diagram of a generation of an impedance network;
fig. 2 is a flow chart of the present invention.
Detailed Description
The technical solutions of the present invention will be clearly and completely described in connection with the embodiments, and it is obvious that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
As shown in fig. 1-2, the method for generating the opposite network deep learning direction finding under the condition of small sample comprises the following steps:
step one: building a spanning countermeasure network
The principle of creating an antagonism network using a nonlinear mapping function construction is shown in figure 1. The generating countermeasure network is composed of a generator G and a discriminator D, wherein the generator G is used for generating a generating sample G (z) according to an input noise sequence z, and the discriminator D is a classification model and is used for judging whether the input sample is a real sample x or the generating sample G (z).
Step two: generation and countermeasure
Firstly, inputting a noise sequence z into a generator G to generate a generated sample G (z) conforming to real data distribution, then inputting the generated sample G (z) and a real sample x into a discriminator D, wherein the discriminator D outputs the probability that the input sample is the real sample, the output is as close to 1 as possible when the real sample x is input, and the output is as close to 0 as possible when the generated sample G (z) is input.
The generator G is fixed, a method for minimizing cross entropy is adopted to train the discriminator D, so that the discriminator D can effectively distinguish a real sample x and generate a sample G (z), and the loss function of the discriminator D is as follows:
in the method, in the process of the invention,wherein x is a real sample, and D (x) is the probability of judging the input real sample x by the judging device D;
where z represents the noise sequence in the input generator G, G (z) represents the generated sample, and D (G (z)) represents the probability that the generated sample passes through the discriminator D and is discriminated as a true sample.
The generator G learns the formal data distribution P data To deceptively determine the discriminator D, the discriminator D is fixed, the generator G increases the probability that the generated sample G (z) is determined to be a true sample, and when the generated sample G (z) is input, the output of the discriminator D is as close to 1 as possible, and the loss function of the generator G is:
the generator G and the arbiter D are opposed to each other and promote each other, and in the continuous countermeasure, they reach nash equalization, and finally, a pseudo-spurious generated sample G (z) can be generated.
Step three: data preprocessing
The data preprocessing includes power amplification, down-conversion and data sampling processing.
Assuming that the number of receiving array antenna array elements is L, the number of incident signals is M, and the vector x (t) = [ x ] for array output 1 (t),x 2 (t),…,x L (t)] T The representation is:
wherein a= [ a ] 1 ,a 1 ,…,a M ]Is an array response matrix corresponding to M incident signals,is the i-th signal incident angle theta i Corresponding array response vectors.
And carrying out power amplification, down-conversion and data sampling processing on the radiation sources received by each array element to obtain sampling data.
Step four: generating CNN input data
In general, each signal is independent, and the covariance matrix of the array output is:
R=E[x(t)x H (t)]=AE[s(t)s H (t)]A H +E[v(t)v H (t)]=AR s A H +R v
wherein R is s And R is v Representing the signal covariance matrix and the noise covariance matrix, respectively.
The relative amplitude of each array element is related to the incidence direction of the signal, and the covariance matrix keeps the angle information of all targets, so that the invention takes the relative amplitude of each array element and the covariance characteristic as the input data of CNN.
Step five: convolution processing
The convolution layer primarily analyzes each small piece of data more deeply, thereby abstracting higher-level features. The convolution layer adopts a local connection mode, and carries out convolution operation on different convolution kernels and input data to realize nonlinear feature mapping on the data, so as to realize feature extraction.
Step six: pooling treatment
The pooling layer realizes data compression through downsampling operation, is used for reducing characteristic dimension after convolution and reducing the number of neurons required by a network. The pooling process divides the input feature map into a plurality of non-overlapping regions based on the translation invariant feature. And (3) carrying out pooling treatment to realize automatic extraction of the target angle feature vector.
Step seven: small sample CNN direction finding
Based on the generation of the countermeasure network, learning training under the condition of a small sample is realized. When the real data is input, the CNN direction finding model is utilized, and the distributed generation features are coupled to different space angles by the full-connection layer through pretreatment, convolution treatment and pooling treatment, so that stable high-precision direction finding is realized.
The above formulas are all formulas with dimensions removed and numerical values calculated, the formulas are formulas which are obtained by acquiring a large amount of data and performing software simulation to obtain the closest actual situation, and preset parameters and preset thresholds in the formulas are set by a person skilled in the art according to the actual situation or are obtained by simulating a large amount of data.
In order to verify the performance of the method, a 9-array element uniform linear array is sampled, under the condition of different angle sample numbers (the sample number of a single angle), a patent 'random array angle of arrival estimation method based on deep learning' (hereinafter referred to as method 1) and the method of the invention are respectively applied, a direction finding statistical test is carried out, and the statistical result of the direction finding accuracy is shown in table 1.
TABLE 1 statistical results of direction finding accuracy
Experimental results show that when the number of training samples is large, the direction-finding accuracy of the two methods is high, but under the condition of small samples, the direction-finding accuracy of the method is about 3% higher than that of the method 1, and the direction-finding stability is better.
The above embodiments are only for illustrating the technical method of the present invention and not for limiting the same, and it should be understood by those skilled in the art that the technical method of the present invention may be modified or substituted without departing from the spirit and scope of the technical method of the present invention.
Claims (5)
1. The method for generating the opposite network deep learning direction finding under the condition of a small sample is characterized by comprising the following steps of:
step one: constructing and generating an countermeasure network by using a nonlinear mapping function; wherein the generation countermeasure network comprises a generator G and a discriminator D;
step two: the generator G and the discriminator D reach Nash equilibrium after mutual antagonism and mutual promotion, and generate a generated sample, which comprises the following steps:
inputting the noise sequence into a generator to generate a generated sample conforming to the real data distribution;
inputting the generated sample and the real sample into a discriminator;
the generator G is fixed, a method of minimizing cross entropy is adopted to train the discriminator D, so that the discriminator D can effectively distinguish real samples and generate samples, and the loss function of the discriminator D is as follows:
in the method, in the process of the invention,wherein x is a real sample, and D (x) is the probability of judging the input real sample x by the judging device D;
wherein z represents a noise sequence input to the generator G, G (z) represents a generated sample, and D (G (z)) represents a probability that the generated sample is discriminated as a true sample after passing through the discriminator D;
the generator G learns the formal data distribution P data Deception of the arbiter D, fixation of the arbiter D, and enlargement of the generator GWhen the generated sample is judged to be the true sample, the output of the discriminator D is as close to 1 as possible when the generated sample is input, and the loss function of the generator G is as follows:
step three: preprocessing the radiation sources received by each array element to obtain sampling data; wherein the preprocessing comprises power amplification, down-conversion and data sampling processing;
step four: taking the relative amplitude and covariance characteristics of each array element as the input data of CNN;
step five: the convolution layer carries out convolution operation with input data through different convolution kernels to realize nonlinear feature mapping of the data, so that feature extraction of the data is realized;
step six: the pooling layer divides the input feature map into a plurality of non-overlapping areas;
step seven: based on generating the antagonism network, the training data set under the condition of the small sample is expanded.
2. The method for generating a countering network deep learning direction finding under a small sample condition according to claim 1, wherein the discriminator D is a classification model.
3. The method for generating the countering network deep learning direction finding under the condition of small samples according to claim 1, wherein the output of the discriminator D is the probability that the input sample is a real sample;
when the input real sample is output as close to 1 as possible;
when the input-generated samples are outputs as close to 0 as possible.
4. The method for generating the opposite network deep learning direction finding under the condition of small samples according to claim 1, wherein the convolution layer adopts a local connection mode.
5. The small sample condition generation opposing network deep learning direction finding method of claim 1 wherein the pooling layer compresses the data by a downsampling operation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310130162.6A CN116205271A (en) | 2023-02-17 | 2023-02-17 | Method for generating opposite network deep learning direction finding under small sample condition |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310130162.6A CN116205271A (en) | 2023-02-17 | 2023-02-17 | Method for generating opposite network deep learning direction finding under small sample condition |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116205271A true CN116205271A (en) | 2023-06-02 |
Family
ID=86516847
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310130162.6A Pending CN116205271A (en) | 2023-02-17 | 2023-02-17 | Method for generating opposite network deep learning direction finding under small sample condition |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116205271A (en) |
-
2023
- 2023-02-17 CN CN202310130162.6A patent/CN116205271A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109597043B (en) | Radar signal identification method based on quantum particle swarm convolutional neural network | |
CN109471074B (en) | Radar radiation source identification method based on singular value decomposition and one-dimensional CNN network | |
CN110133599B (en) | Intelligent radar radiation source signal classification method based on long-time and short-time memory model | |
CN112990334A (en) | Small sample SAR image target identification method based on improved prototype network | |
CN110188647A (en) | One kind being based on the feature extraction of variation mode decomposition Radar emitter and its classification method | |
CN111582236B (en) | LPI radar signal classification method based on dense convolutional neural network | |
CN112036239B (en) | Radar signal working mode identification method and system based on deep learning network | |
CN109948722B (en) | Method for identifying space target | |
CN111948622A (en) | Linear frequency modulation radar signal TOA estimation algorithm based on parallel CNN-LSTM | |
CN113673312A (en) | Radar signal intra-pulse modulation identification method based on deep learning | |
CN116797846A (en) | Method and device for identifying small sample radar radiation source based on RoAtten-PN network | |
Pan et al. | Specific radar emitter identification using 1D-CBAM-ResNet | |
Feng et al. | FCGCN: Feature Correlation Graph Convolution Network for Few-Shot Individual Identification | |
CN112766381A (en) | Attribute-guided SAR image generation method under limited sample | |
CN116842450A (en) | Depth measurement learning-based radar radiation source individual identification method | |
CN115809426A (en) | Radiation source individual identification method and system | |
CN116205271A (en) | Method for generating opposite network deep learning direction finding under small sample condition | |
Meng et al. | Real‐valued propagator method for fast DOA estimation via polynomial rooting | |
CN112946567B (en) | Moving target fingerprint indoor positioning method based on domain antagonism neural network | |
CN115329821A (en) | Ship noise identification method based on pairing coding network and comparison learning | |
Guo et al. | SAR target recognition with limited samples based on meta knowledge transferring using relation network | |
He et al. | A multi-scale radar HRRP target recognition method based on pyramid depthwise separable convolution network | |
CN107403136A (en) | The SAR target model recognition methods of dictionary learning is kept based on structure | |
Sui et al. | Frequency-hopping signal radio sorting based on stacked auto-encoder subtle feature extraction | |
Guo et al. | Deinterleaving of radar pulse based on implicit feature |
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 |