CN113205140A - Semi-supervised specific radiation source individual identification method based on generative countermeasure network - Google Patents
Semi-supervised specific radiation source individual identification method based on generative countermeasure network Download PDFInfo
- Publication number
- CN113205140A CN113205140A CN202110490858.0A CN202110490858A CN113205140A CN 113205140 A CN113205140 A CN 113205140A CN 202110490858 A CN202110490858 A CN 202110490858A CN 113205140 A CN113205140 A CN 113205140A
- Authority
- CN
- China
- Prior art keywords
- sample
- network
- radiation source
- real
- training
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06K—GRAPHICAL DATA READING; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
- G06K7/00—Methods or arrangements for sensing record carriers, e.g. for reading patterns
- G06K7/10—Methods or arrangements for sensing record carriers, e.g. for reading patterns by electromagnetic radiation, e.g. optical sensing; by corpuscular radiation
- G06K7/10009—Methods or arrangements for sensing record carriers, e.g. for reading patterns by electromagnetic radiation, e.g. optical sensing; by corpuscular radiation sensing by radiation using wavelengths larger than 0.1 mm, e.g. radio-waves or microwaves
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Toxicology (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Electromagnetism (AREA)
- General Health & Medical Sciences (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses a semi-supervised individual identification method of a specific radiation source based on a generative countermeasure network. The method comprises the following steps: acquiring a real sample and generating a sample; the real samples comprise labeled real samples and unlabeled real samples; judging the generated sample and the labeled real sample; training the auxiliary classification network through the obtained labeled real sample and the generated sample; judging the generated sample and the label-free real sample; training the trained auxiliary classification network again through the generated sample obtained by discrimination; and identifying the specific radiation source individuals through the retrained auxiliary classification network. The method can be well adapted to the specific radiation source individual identification task under the conditions of low signal-to-noise ratio and small sample weak labeling, and the efficiency and accuracy of specific radiation source individual identification are improved.
Description
Technical Field
The invention relates to the technical field of radio frequency signal processing, in particular to a semi-supervised individual identification method of a specific radiation source based on a generative countermeasure network.
Background
The specific radiation source refers to a technology for identifying the communication radiation source individual to which the specific radiation source belongs according to the fingerprint characteristics of the received radio frequency communication signal. The internal hardware of different radiation source individuals has slight difference, and the difference is reflected in the radio frequency signals emitted by the radiation source individuals and shows the characteristic of unique identification, namely radio frequency fingerprint characteristics, so that the different radiation source individuals can be identified. The technology is widely applied to the civil and military fields, particularly the military electronic reconnaissance field, at present, military troops of various countries are equipped with electronic equipment with large quantities of models and consistent signal patterns, and therefore the traditional electronic reconnaissance based on signal pattern recognition faces great difficulty in tasks such as distinguishing different target individuals. Furthermore, this difficulty is further exacerbated by the nature of non-cooperative communications in the battlefield environment. Therefore, the communication reconnaissance based on the specific radiation source individual identification technology is developed, the number and the scale of the targets in the battle mission are further judged by distinguishing the radiation sources of the same production model and signal style and different individuals, and finally the information such as the force deployment, the battlefield situation and the like of the enemy is obtained, so that the method has wide application prospect and higher military value.
The core of the individual identification technology of the specific radiation source is the extraction of radio frequency fingerprint characteristics. In the traditional method, a manual predefining mode is mostly adopted, and an extraction method of the radio frequency fingerprint features is designed. The usual methods include: high-order signal spectrum, time-frequency domain feature extraction, empirical mode decomposition and the like. However, the rf fingerprint features are generated by the combined action of various hardware differences inside the radiation source, and have high complexity. Therefore, when a specific radiation source individual identification task is met, the traditional method usually needs to adopt a 'one-by-one try' method to select a proper and effective fingerprint feature extraction algorithm. The feature selection thought with extremely low efficiency seriously influences the development speed of the communication radiation source identification system, and is not beneficial to the popularization and application of the technology.
Disclosure of Invention
The invention aims to provide a semi-supervised specific radiation source individual identification method based on a generative countermeasure network, which improves the efficiency and accuracy of specific radiation source individual identification.
In order to achieve the purpose, the invention provides the following scheme:
a semi-supervised specific radiation source individual identification method based on a generative countermeasure network comprises the following steps:
acquiring a real sample and generating a sample; the real samples comprise labeled real samples and unlabeled real samples; the generated sample is obtained by coding an implicit code and an implicit vector; the implicit code is obtained by sampling uniform distribution containing radiation source individual category information, and the implicit vector is obtained by sampling multimode Nakagami-m distribution containing wireless channel prior information;
acquiring an auxiliary classification network;
judging the generated sample and the labeled real sample;
training the auxiliary classification network through the obtained labeled real sample and the generated sample;
judging the generated sample and the label-free real sample;
training the trained auxiliary classification network again through the generated sample obtained by discrimination;
and identifying the specific radiation source individuals through the retrained auxiliary classification network.
Further, the distinguishing the generated sample and the labeled real sample specifically includes:
acquiring a discrimination network;
training the discrimination network by generating a sample and a labeled real sample;
and judging the generated sample and the labeled real sample through the trained judging network.
Further, the acquiring process of the real sample and the generated sample comprises:
obtaining a characterization vector; the characterization vector is obtained by encoding sample data; the sample data comprises labeled sample data and unlabeled sample data;
sampling the uniform distribution containing the individual category information of the radiation source to obtain implicit codes;
sampling multi-modal Nakagami-m distribution containing wireless channel prior information to obtain an implicit vector;
acquiring a generation network;
training the generation network through the characterization vector, the implicit codes and the implicit vectors, wherein the generation network outputs generation samples and real samples.
Further, the process of obtaining the characterization vector specifically includes:
acquiring sample data; the sample data comprises labeled sample data and unlabeled sample data;
acquiring a coding network;
training the coding network through the sample data; the coded network output is a characterization vector.
The invention also provides a semi-supervised specific radiation source individual identification system based on the generative countermeasure network, which comprises the following steps:
the sample acquisition module is used for acquiring a real sample and generating a sample; the real samples comprise labeled real samples and unlabeled real samples; the generated sample is obtained by coding an implicit code and an implicit vector; the implicit code is obtained by sampling uniform distribution containing radiation source individual category information, and the implicit vector is obtained by sampling multimode Nakagami-m distribution containing wireless channel prior information;
the auxiliary classification network acquisition module is used for acquiring an auxiliary classification network;
the first judging module is used for judging the generated sample and the labeled real sample;
the auxiliary classification network training module is used for training an auxiliary classification network through the labeled real sample and the generated sample obtained through discrimination;
the second judging module is used for judging the generated sample and the label-free real sample;
the retraining module is used for retraining the trained auxiliary classification network through the generated sample obtained by discrimination;
and the recognition module is used for recognizing the specific radiation source individuals through the retrained auxiliary classification network.
Further, the first determination module specifically includes:
a discrimination network acquisition unit for acquiring a discrimination network;
the judgment network training unit is used for training the judgment network through generating samples and labeled real samples;
and the judging unit is used for judging the generated sample and the labeled real sample through the trained judging network.
Further, the sample acquisition module comprises:
the characterization vector acquisition unit is used for acquiring a characterization vector; the characterization vector is obtained by encoding sample data; the sample data comprises labeled sample data and unlabeled sample data;
the first sampling unit is used for sampling the uniform distribution containing the individual category information of the radiation source to obtain implicit codes;
the second sampling unit is used for sampling the multi-modal Nakagami-m distribution containing the prior information of the wireless channel to obtain an implicit vector;
a generation network acquisition unit for acquiring a generation network;
and the generating network training unit is used for training the generating network through the characterization vector, the implicit codes and the implicit vectors, and the generating network outputs a generating sample and a real sample.
Further, the token vector obtaining unit specifically includes:
the data acquisition subunit is used for acquiring sample data; the sample data comprises labeled sample data and unlabeled sample data;
the coding network acquiring subunit is used for acquiring a coding network;
the coding network training unit is used for training the coding network through the sample data; the coded network output is a characterization vector.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
(1) the individual identification task of a specific radiation source can be realized under the condition of low signal-to-noise ratio, and the identification rate can be maintained at a higher level. Meanwhile, the robustness to noise interference is good;
(2) the ratio of the number of the signal data with the label to the number of the signal data without the label participating in the training is in a smaller range, and the network model can realize higher recognition rate;
(3) the invention has faster convergence speed and can reduce the time and difficulty of network training.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of a semi-supervised individual identification method of a specific radiation source based on a generative countermeasure network according to an embodiment of the present invention;
FIG. 2 is an overall block diagram of a network model according to an embodiment of the present invention;
FIG. 3 is a graph of identification performance testing at different signal-to-noise ratios;
FIG. 4 is a graph of identification performance testing at different ratios of the amount of labeled to unlabeled data;
fig. 5 is a test of the convergence rate of the network model.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a semi-supervised specific radiation source individual identification method based on a generative countermeasure network, which improves the efficiency and accuracy of specific radiation source individual identification.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
As shown in fig. 1, the semi-supervised specific radiation source individual identification method based on the generative countermeasure network disclosed by the invention comprises the following steps:
step 101: acquiring a real sample and generating a sample; the real samples comprise labeled real samples and unlabeled real samples; the generated sample is obtained by coding an implicit code and an implicit vector; the implicit codes are obtained by sampling the uniform distribution containing the individual category information of the radiation source, and the implicit vectors are obtained by sampling the multimode Nakagami-m distribution containing the prior information of the wireless channel.
Step 102: and acquiring the auxiliary classification network.
Step 103: and judging the generated sample and the labeled real sample.
Step 104: and training the auxiliary classification network through distinguishing the obtained labeled real sample and the generated sample.
Step 105: and judging the generated sample and the label-free real sample.
Step 106: and training the trained auxiliary classification network again through the generated sample obtained by discrimination.
Step 101: and identifying the specific radiation source individuals through the retrained auxiliary classification network.
The distinguishing between the generated sample and the labeled real sample specifically includes: acquiring a discrimination network; training the discrimination network by generating a sample and a labeled real sample; and judging the generated sample and the labeled real sample through the trained judging network.
The process of distinguishing the generated sample from the non-labeled real sample is the same as the process of distinguishing the generated sample from the labeled real sample.
Wherein the acquiring process of the real sample and the generated sample comprises the following steps: obtaining a characterization vector; the characterization vector is obtained by encoding sample data; the sample data comprises labeled sample data and unlabeled sample data; sampling the uniform distribution containing the individual category information of the radiation source to obtain implicit codes; sampling multi-modal Nakagami-m distribution containing wireless channel prior information to obtain an implicit vector; acquiring a generation network; training the generation network through the characterization vector, the implicit codes and the implicit vectors, wherein the generation network outputs generation samples and real samples.
The obtaining process of the characterization vector specifically includes: acquiring sample data; the sample data comprises labeled sample data and unlabeled sample data; acquiring a coding network; training the coding network through the sample data; the coded network output is a characterization vector.
The following describes the training procedures of the generation network g (generatornenetworks), the discrimination network d (discriminator networks), the encoding network e (encoder), and the auxiliary classification network q (automatic classification networks) in detail:
generally, the radio frequency fingerprint features are generated by a plurality of different hardware parts inside the radiation source acting together in different working processes, but the influence generated by a Power Amplifier (PA) is most significant. The non-linear components of the power amplifier and the memory storage component produce significant non-linear distortion to the communication signal, which can be used as a fingerprint of the rf signal. The frequency response of the power amplifier may be represented by a form of taylor polynomial. Assuming an input signal
Wherein s is0(t) denotes a baseband signal, fcRepresenting the carrier frequency. The output signal of the power amplifier is then:
wherein λiAnd the coefficients representing the Taylor polynomials comprise the radio frequency fingerprint characteristics.
Fig. 2 is a block diagram of a network architecture according to the present invention. The overall network model contains 4 sub-networks: generation networks g (generatornetworks), discriminant networks d (discriminant networks), encoding networks e (encoder), and auxiliary classification networks q (auxiary classifiernetworks).
The inputs to the network include initial signal data (x, y), implicit vector z (late vector), and implicit code c (late code). In the initial signal data, x ═ xl,xulDenotes signal samples, comprising M labeled signal samples xl＝{xl1,xl2,…,xlMAnd N unlabeled signal samples xul＝{xu1,xu2,…,xuN}。yl＝{yl1,yl2,…,ylMIndicates the label information corresponding to the labeled signal sample. The implicit vector z is obtained by sampling random noise, and the implicit code c is obtained by sampling a uniform distribution (K, p is 1/K) containing the individual category information of the radiation sources, wherein K represents the number of the radiation sources. For the generic generative confrontation network model, the implicit vector z follows a uniform distribution or a standard normal distribution. However, the received communication signal is affected by multipath fading during transmission, so that a multi-modal probability distribution can be constructed and sampled to obtain the implicit vector z. Therefore, the generated samples obtained by generating the network can better match the signal classes with the signal distribution, and the classification performance of the network is improved. In particular, where the multimodal probability distribution is chosen to be a Nakagami-m distribution, the probability distribution of the implicit vector z can be expressed as:
wherein, { ciIt is noted that (i ═ 1,2, … K) represents the implicit code c, used to determine which sub-Nakagami-m distribution the implicit vector z should be sampled from. p is a radical ofi(z) represents a Nakagami-m distribution group, which can be represented as follows:
wherein m represents the fading parameter of Nakagami-m distribution, and the value range
And is specific to each daughter Nakagami-m distribution. Γ (·) represents a gamma function. p is a radical ofriRepresenting the average energy of the signal of the i-th class.
Training samples (x, y) are firstly sent into an encoding network E, the encoding network performs dimension compression processing on original data to obtain a mean vector mu and a standard deviation vector sigma, so that a sample vector space Z ═ mu + epsilon ∑ sigma is constructed, wherein epsilon-N (0, I), in order to avoid the problem that gradients cannot be reversely propagated because mu and sigma are random values, a random vector epsilon which is subjected to normal distribution is defined, the randomness of the samples is transferred to epsilon, and mu and sigma which really need to be trained are fixed at the specific training moment, so that the reverse propagation can be normally performed, and the encoding network E can be normally trained. And sampling the sample vector space Z to obtain a low-dimensional characterization vector Z ', wherein the characterization vector Z' comprises the probability distribution characteristic of the original data, and decoding the probability distribution characteristic to obtain a real sample. In the actual process, the generating network G decodes z 'to obtain a real sample r'. In addition, the generation network also decodes the implicit vector z and the implicit code c to obtain a generated sample r ═ G (z, c). In the encoding and decoding processes, the encoding network E and the generating network G are combined to be a variational auto-encoder (VAE), and the combination can effectively avoid the problem of mode collapse of a network model caused by the complexity of a radiation source signal. From the mean vector μ and the standard deviation vector σ, the loss function of the coding network can be expressed as:
for the discrimination network D, the input comprises a real sample r' and a false sample r, the function is to discriminate whether the input sample is the real sample or the false sample, the specific working process is like a two-classifier, and the label "1" is assigned to the real sample, and the label "0" is assigned to the false sample. The optimization target of the discrimination network is as follows:
the penalty function for a discriminant network can be expressed as:
the role of the auxiliary classification network Q is to make the mutual information value I [ G (z, c) between the generated sample r ═ G (z, c) and the implicit code c; c ] reaches a maximum. When the mutual information value of the two is larger, the relation between the implicit code c and the generated sample r is larger. In the actual calculation process, mutual information has edge probability which is difficult to calculate, so a variation inference method is adopted to solve the problem. First, mutual information of the implicit code c and the generated sample r can be expressed as follows:
I[G(z,c)；c]＝H(c)-H(c|G(z,c)) (7)
then, further derivation of-H (c | G (z, c)) was performed to obtain its lower bound:
thus, the lower bound of mutual information can be obtained:
thus far, the present invention defines LI(G, Q) to approximate mutual information I [ G (z, c); c. C]When the difference between the two is sufficiently small, L can be usedIThe (G, Q) function directly replaces the mutual information between the implicit code c and the generated network generated data. L isIThe edge probabilities still exist in (G, Q), which can be approximated using Monte Carlo (Monte Carlo) simulation methods.
In the actual training process, when the implicit code c is obtained by sampling a uniform distribution (K, p is 1/K) containing radiation source individual class information, the auxiliary classification network can be regarded as a classifier of a labeled generation sample G (z, c), the classification result is compared with the corresponding implicit code c, and a cross entropy function of the two is calculated as a loss function of the network:
furthermore, when the original signal data is tagged, the true sample generated by the generating network r ═ G (z', y)l) With corresponding tag information ylAnd the method also participates in the training of the auxiliary classification network, and the generated loss function is added into the original loss function in a regularization mode, wherein the loss function of the network is as follows:
wherein λQRepresenting the regularization weighting coefficients.
The generation network G has the functions of decoding the implicit vector z and the implicit code c to obtain a generated sample r ═ G (z, c), and simultaneously decoding the low-dimensional characterization vector z 'to obtain a real sample r ═ G (z', y ═ Gl) The training goal being to generate a probability distribution p of the sampleG(z, c) probability distribution p of sample as true as possibleG(z') approximation, discrimination of the loss function of the network generated during training by the discriminant network and assistanceThe loss function of the classification network is jointly determined, and specifically, the loss function is as follows:
LG＝λG1LD+λG2LQ (12)
wherein λG1、λG2Representing the regularization weighting coefficients.
In the training process, firstly, the network is subjected to instructive supervised training by adopting the signal data with the label, and then the network is subjected to optimization unsupervised training by adopting the signal data without the label. When the whole network is converged after training is finished, the auxiliary classification network can be used for classifying the received radio frequency communication signals so as to complete the individual identification task of the radiation source.
The training and recognition process of the network model is as follows:
(1) receiving initial signal data, dividing the initial signal data into two groups of data sets according to the existence of label information, performing instructive training on a network model by using the labeled signal data in a supervised learning mode, and then optimizing the network model by using the unlabeled signal data in an unsupervised learning mode;
(2) training is first performed using signal data with labels. Firstly, the vector is sent into a coding network to be coded so as to obtain a low-dimensional characterization vector zl′＝E(xl,yl) Then the token vector zl' sending into a generation network for decoding to obtain a real sample rl′＝G(zl′,yl). On the other hand, sampling is respectively carried out from uniform distribution (K, p is 1/K) containing radiation source individual category information and multi-modal Nakagami-m distribution containing wireless channel prior information, implicit codes c and implicit vectors z are respectively obtained, and then the implicit codes c and the implicit vectors z are sent into a generation network to be decoded, so that generated samples r are G (z, c);
(3) network parameters of the fixed coding network and the generating network are used for generating a sample r and a real sample rlThe data is sent to a discrimination network to discriminate the authenticity of the data, and a loss function of the discrimination network is calculated according to the formula (6), so that the network parameters are optimized. At the same time, the generated sample r and the real sample rl' feeding into an auxiliary classification network for predictive classification,and (3) classifying the classification result and corresponding label information: implicit code c and signal tag ylComparing, and calculating a cross entropy function of the cross entropy function as a loss function of the auxiliary classification network according to the formula (11), so as to optimize network parameters of the cross entropy function;
(4) network parameters of a fixed discrimination network and an auxiliary classification network, a training coding network and a generating network, and calculating loss functions of the coding network and the generating network according to the formulas (4) and (12) respectively so as to optimize the network parameters;
(5) and then training is carried out by using the signal data without the label. Similarly, a true sample r is obtained according to step 2u′＝G(zu') and generate the sample r-G (z, c), then train each sub-network according to step 3, calculate the respective loss function and optimize the parameters accordingly. When training the auxiliary classification network, only sending the generated samples r into the auxiliary classification network for prediction classification, comparing the classification result with the corresponding implicit code c, and calculating the cross entropy function of the classification result as the loss function of the auxiliary classification network according to the formula (10) so as to optimize the network parameters of the auxiliary classification network;
(6) and (5) repeating the step (2) to the step (5) for a plurality of iterations, and after the whole network is trained, classifying the received radio frequency communication signals by using the auxiliary classification network, thereby finally completing the individual identification task of the radiation source.
The invention embeds a Variational self-encoder (VAE) based on a variant network model (InfoGAN) (information maximum adaptive network) of a Generative Antagonistic Network (GAN), and then carries out semi-supervised learning by comprehensively utilizing labeled and unlabeled training signal samples. When the whole network model converges, an Auxiliary classification network (Auxiliary Classifier Networks) is used for classifying the received radio frequency communication signals, and therefore the specific radiation source individual identification task is completed.
In a specific embodiment: MATLAB software is utilized to generate simulation signal data based on a Taylor polynomial of a power amplifier, a Taylor series is taken as 5, 5 types of radiation source radio-frequency signals are generated, and the coefficient of the corresponding Taylor polynomial is as follows: α 1 ═ 10.50.30.050.2,. α 2 ═ 10.60.040.050.4,. α 3 ═ 10.080.60.40.8,. α 4 ═ 10.10.80.040.06,. α 5 ═ 10.10.010.030.15. The carrier frequency of the signal is set to be 2GHz, the modulation mode is 16-QAM, the sampling frequency of the signal is 10GHz, and the number of sampling points of each signal sample is 1000. The radio-frequency signals of various radiation sources are respectively generated under different signal-to-noise ratios, and the values of the signal-to-noise ratios are set to be 0dB, 2dB, … … dB, 22dB and 24 dB. 20000 samples are respectively generated by each type of radiation source radio-frequency signals under different signal-to-noise ratios, 2000 samples are randomly selected as labeled training samples, 8000 samples are unlabeled training samples, and 1000 samples are selected as test samples.
Fig. 3 shows the identification rates of the semi-supervised individual identification method of specific radiation source based on the generative countermeasure network, which is provided by the present invention, at different signal-to-noise ratios when the types of radiation sources are 3, 4, and 5, respectively. The embodiment results show that although the identification rates are different when the radiation source types are different, the overall identification rate can be kept high, and the identification rate can reach more than 90% under the signal-to-noise ratio of 10 dB. In addition, the individual identification task of a specific radiation source can be realized under the condition of low signal-to-noise ratio, and the identification rate can be maintained at a higher level.
In fig. 4, the type of the radiation source is set to 3, and each type of radiation source generates 20000 radio frequency signal samples under the signal-to-noise ratios of 4dB, 12dB and 20dB, and 8000 non-label training samples and 1000 test samples are randomly selected. Then, 200, 400 and … … 1800 samples are randomly selected as labeled training samples respectively, namely the ratio of the number of the labeled training samples to the number of the unlabeled training samples is respectively 2.5%, 5%, … …% and 22.5%. The recognition rates at different ratios are shown in fig. 4. The embodiment result shows that when the ratio of the number of the labeled training samples to the number of the unlabeled training samples reaches 7.5%, the recognition rate can be stable, which indicates that the method can be well adapted to the individual recognition task of the specific radiation source under the condition of small samples.
FIG. 5 is a diagram illustrating the convergence performance of the network model VAE-InfoGAN proposed by the present invention compared with the conventional InfoGAN network model. And setting the maximum training times as 100, and testing the loss function values of the two network models under different training times. The embodiment result shows that the VAE-InfoGAN has higher convergence speed than the InfoGAN, which shows that the method can reduce the training difficulty of the network, reduce the training time and enable the network to reach the convergence state more quickly.
The invention also provides a semi-supervised specific radiation source individual identification system based on the generative countermeasure network, which comprises the following steps:
the sample acquisition module is used for acquiring a real sample and generating a sample; the real samples comprise labeled real samples and unlabeled real samples; the generated sample is obtained by coding an implicit code and an implicit vector; the implicit code is obtained by sampling uniform distribution containing radiation source individual category information, and the implicit vector is obtained by sampling multimode Nakagami-m distribution containing wireless channel prior information;
the auxiliary classification network acquisition module is used for acquiring an auxiliary classification network;
the first judging module is used for judging the generated sample and the labeled real sample;
the auxiliary classification network training module is used for training an auxiliary classification network through the labeled real sample and the generated sample obtained through discrimination;
the second judging module is used for judging the generated sample and the label-free real sample;
the retraining module is used for retraining the trained auxiliary classification network through the generated sample obtained by discrimination;
and the recognition module is used for recognizing the specific radiation source individuals through the retrained auxiliary classification network.
Wherein, the first judging module specifically comprises:
a discrimination network acquisition unit for acquiring a discrimination network;
the judgment network training unit is used for training the judgment network through generating samples and labeled real samples;
and the judging unit is used for judging the generated sample and the labeled real sample through the trained judging network.
Wherein the sample acquisition module comprises:
the characterization vector acquisition unit is used for acquiring a characterization vector; the characterization vector is obtained by encoding sample data; the sample data comprises labeled sample data and unlabeled sample data;
the first sampling unit is used for sampling the uniform distribution containing the individual category information of the radiation source to obtain implicit codes;
the second sampling unit is used for sampling the multi-modal Nakagami-m distribution containing the prior information of the wireless channel to obtain an implicit vector;
a generation network acquisition unit for acquiring a generation network;
and the generating network training unit is used for training the generating network through the characterization vector, the implicit codes and the implicit vectors, and the generating network outputs a generating sample and a real sample.
Wherein, the characterization vector obtaining unit specifically includes:
the data acquisition subunit is used for acquiring sample data; the sample data comprises labeled sample data and unlabeled sample data;
the coding network acquiring subunit is used for acquiring a coding network;
the coding network training unit is used for training the coding network through the sample data; the coded network output is a characterization vector.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (8)
1. A semi-supervised individual identification method of a specific radiation source based on a generative countermeasure network is characterized by comprising the following steps:
acquiring a real sample and generating a sample; the real samples comprise labeled real samples and unlabeled real samples; the generated sample is obtained by coding an implicit code and an implicit vector; the implicit code is obtained by sampling uniform distribution containing radiation source individual category information, and the implicit vector is obtained by sampling multimode Nakagami-m distribution containing wireless channel prior information;
acquiring an auxiliary classification network;
judging the generated sample and the labeled real sample;
training the auxiliary classification network through the obtained labeled real sample and the generated sample;
judging the generated sample and the label-free real sample;
training the trained auxiliary classification network again through the generated sample obtained by discrimination;
and identifying the specific radiation source individuals through the retrained auxiliary classification network.
2. The semi-supervised specific radiation source individual identification method based on the generative countermeasure network as recited in claim 1, wherein the distinguishing the generated sample and the labeled real sample specifically comprises:
acquiring a discrimination network;
training the discrimination network by generating a sample and a labeled real sample;
and judging the generated sample and the labeled real sample through the trained judging network.
3. The semi-supervised specific radiation source individual identification method based on the generative countermeasure network as recited in claim 1, wherein the acquiring process of the real sample and the generated sample comprises:
obtaining a characterization vector; the characterization vector is obtained by encoding sample data; the sample data comprises labeled sample data and unlabeled sample data;
sampling the uniform distribution containing the individual category information of the radiation source to obtain implicit codes;
sampling multi-modal Nakagami-m distribution containing wireless channel prior information to obtain an implicit vector;
acquiring a generation network;
training the generation network through the characterization vector, the implicit codes and the implicit vectors, wherein the generation network outputs generation samples and real samples.
4. The semi-supervised individual identification method for a specific radiation source based on a generative countermeasure network as claimed in claim 3, wherein the obtaining process of the characterization vector comprises:
acquiring sample data; the sample data comprises labeled sample data and unlabeled sample data;
acquiring a coding network;
training the coding network through the sample data; the coded network output is a characterization vector.
5. A semi-supervised radiation source-specific individual identification system based on a generative countermeasure network, comprising:
the sample acquisition module is used for acquiring a real sample and generating a sample; the real samples comprise labeled real samples and unlabeled real samples; the generated sample is obtained by coding an implicit code and an implicit vector; the implicit code is obtained by sampling uniform distribution containing radiation source individual category information, and the implicit vector is obtained by sampling multimode Nakagami-m distribution containing wireless channel prior information;
the auxiliary classification network acquisition module is used for acquiring an auxiliary classification network;
the first judging module is used for judging the generated sample and the labeled real sample;
the auxiliary classification network training module is used for training an auxiliary classification network through the labeled real sample and the generated sample obtained through discrimination;
the second judging module is used for judging the generated sample and the label-free real sample;
the retraining module is used for retraining the trained auxiliary classification network through the generated sample obtained by discrimination;
and the recognition module is used for recognizing the specific radiation source individuals through the retrained auxiliary classification network.
6. The semi-supervised specific radiation source individual identification system based on the generative countermeasure network as recited in claim 5, wherein the first discrimination module specifically comprises:
a discrimination network acquisition unit for acquiring a discrimination network;
the judgment network training unit is used for training the judgment network through generating samples and labeled real samples;
and the judging unit is used for judging the generated sample and the labeled real sample through the trained judging network.
7. The semi-supervised specific radiation source individual identification system based on the generative countermeasure network of claim 5, wherein the sample acquisition module comprises:
the characterization vector acquisition unit is used for acquiring a characterization vector; the characterization vector is obtained by encoding sample data; the sample data comprises labeled sample data and unlabeled sample data;
the first sampling unit is used for sampling the uniform distribution containing the individual category information of the radiation source to obtain implicit codes;
the second sampling unit is used for sampling the multi-modal Nakagami-m distribution containing the prior information of the wireless channel to obtain an implicit vector;
a generation network acquisition unit for acquiring a generation network;
and the generating network training unit is used for training the generating network through the characterization vector, the implicit codes and the implicit vectors, and the generating network outputs a generating sample and a real sample.
8. The semi-supervised specific radiation source individual identification system based on the generative countermeasure network as recited in claim 7, wherein the characterization vector obtaining unit specifically comprises:
the data acquisition subunit is used for acquiring sample data; the sample data comprises labeled sample data and unlabeled sample data;
the coding network acquiring subunit is used for acquiring a coding network;
the coding network training unit is used for training the coding network through the sample data; the coded network output is a characterization vector.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110490858.0A CN113205140B (en) | 2021-05-06 | 2021-05-06 | Semi-supervised specific radiation source individual identification method based on generative countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110490858.0A CN113205140B (en) | 2021-05-06 | 2021-05-06 | Semi-supervised specific radiation source individual identification method based on generative countermeasure network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113205140A true CN113205140A (en) | 2021-08-03 |
| CN113205140B CN113205140B (en) | 2022-11-15 |
Family
ID=77029861
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110490858.0A Active CN113205140B (en) | 2021-05-06 | 2021-05-06 | Semi-supervised specific radiation source individual identification method based on generative countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113205140B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111582320A (en) * | 2020-04-17 | 2020-08-25 | 电子科技大学 | Dynamic individual identification method based on semi-supervised learning |
| CN115130743A (en) * | 2022-06-22 | 2022-09-30 | 国网山东省电力公司电力科学研究院 | Wind turbine generator regulation rate prediction method and system based on variation inference |
| CN115277189A (en) * | 2022-07-27 | 2022-11-01 | 中国人民解放军海军航空大学 | Unsupervised intrusion flow detection and identification method based on generative countermeasure network |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080102881A1 (en) * | 2006-10-25 | 2008-05-01 | Samsung Electronics Co., Ltd. | Method and apparatus for adaptively allocating transmission power for beam-forming combined with OSTBCs in a distributed wireless communication system |
| CN104052557A (en) * | 2014-06-24 | 2014-09-17 | 西安电子科技大学 | Method for modeling Nakagami repeated fading channel |
| CN106993851A (en) * | 2017-05-18 | 2017-08-01 | 李文谦 | A kind of shoe tree parameter automatic prediction method and prediction meanss based on shoes image and foot type image |
| CN107330954A (en) * | 2017-07-14 | 2017-11-07 | 深圳市唯特视科技有限公司 | A kind of method based on attenuation network by sliding attribute manipulation image |
| CN107862293A (en) * | 2017-09-14 | 2018-03-30 | 北京航空航天大学 | Radar based on confrontation generation network generates colored semantic image system and method |
| CN108985464A (en) * | 2018-07-17 | 2018-12-11 | 重庆科技学院 | The continuous feature generation method of face for generating confrontation network is maximized based on information |
| CN109410179A (en) * | 2018-09-28 | 2019-03-01 | 合肥工业大学 | A kind of image abnormity detection method based on generation confrontation network |
| CN110297218A (en) * | 2019-07-09 | 2019-10-01 | 哈尔滨工程大学 | Based on the unknown modulation system detection method of radar signal for generating confrontation network |
| CN111027439A (en) * | 2019-12-03 | 2020-04-17 | 西北工业大学 | SAR target recognition method for generating countermeasure network based on auxiliary classification |
| CN111106856A (en) * | 2019-12-31 | 2020-05-05 | 华侨大学 | Cooperative relay network and transmission method thereof |
| CN112270351A (en) * | 2020-10-24 | 2021-01-26 | 国网江苏省电力有限公司信息通信分公司 | Semi-supervised encryption traffic identification method for generating countermeasure network based on auxiliary classification |
-
2021
- 2021-05-06 CN CN202110490858.0A patent/CN113205140B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080102881A1 (en) * | 2006-10-25 | 2008-05-01 | Samsung Electronics Co., Ltd. | Method and apparatus for adaptively allocating transmission power for beam-forming combined with OSTBCs in a distributed wireless communication system |
| CN104052557A (en) * | 2014-06-24 | 2014-09-17 | 西安电子科技大学 | Method for modeling Nakagami repeated fading channel |
| CN106993851A (en) * | 2017-05-18 | 2017-08-01 | 李文谦 | A kind of shoe tree parameter automatic prediction method and prediction meanss based on shoes image and foot type image |
| CN107330954A (en) * | 2017-07-14 | 2017-11-07 | 深圳市唯特视科技有限公司 | A kind of method based on attenuation network by sliding attribute manipulation image |
| CN107862293A (en) * | 2017-09-14 | 2018-03-30 | 北京航空航天大学 | Radar based on confrontation generation network generates colored semantic image system and method |
| CN108985464A (en) * | 2018-07-17 | 2018-12-11 | 重庆科技学院 | The continuous feature generation method of face for generating confrontation network is maximized based on information |
| CN109410179A (en) * | 2018-09-28 | 2019-03-01 | 合肥工业大学 | A kind of image abnormity detection method based on generation confrontation network |
| CN110297218A (en) * | 2019-07-09 | 2019-10-01 | 哈尔滨工程大学 | Based on the unknown modulation system detection method of radar signal for generating confrontation network |
| CN111027439A (en) * | 2019-12-03 | 2020-04-17 | 西北工业大学 | SAR target recognition method for generating countermeasure network based on auxiliary classification |
| CN111106856A (en) * | 2019-12-31 | 2020-05-05 | 华侨大学 | Cooperative relay network and transmission method thereof |
| CN112270351A (en) * | 2020-10-24 | 2021-01-26 | 国网江苏省电力有限公司信息通信分公司 | Semi-supervised encryption traffic identification method for generating countermeasure network based on auxiliary classification |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111582320A (en) * | 2020-04-17 | 2020-08-25 | 电子科技大学 | Dynamic individual identification method based on semi-supervised learning |
| CN111582320B (en) * | 2020-04-17 | 2022-10-14 | 电子科技大学 | Dynamic individual identification method based on semi-supervised learning |
| CN115130743A (en) * | 2022-06-22 | 2022-09-30 | 国网山东省电力公司电力科学研究院 | Wind turbine generator regulation rate prediction method and system based on variation inference |
| CN115277189A (en) * | 2022-07-27 | 2022-11-01 | 中国人民解放军海军航空大学 | Unsupervised intrusion flow detection and identification method based on generative countermeasure network |
| CN115277189B (en) * | 2022-07-27 | 2023-08-15 | 中国人民解放军海军航空大学 | Unsupervised intrusion flow detection and identification method based on generation type countermeasure network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113205140B (en) | 2022-11-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113205140B (en) | Semi-supervised specific radiation source individual identification method based on generative countermeasure network | |
| Li et al. | Robust automated VHF modulation recognition based on deep convolutional neural networks | |
| Zhang et al. | An efficient deep learning model for automatic modulation recognition based on parameter estimation and transformation | |
| CN112308133A (en) | Modulation identification method based on convolutional neural network | |
| CN109890043B (en) | Wireless signal noise reduction method based on generative countermeasure network | |
| CN112115821B (en) | Multi-signal intelligent modulation mode identification method based on wavelet approximate coefficient entropy | |
| CN111490853A (en) | Channel coding parameter identification method based on deep convolutional neural network | |
| CN112039820B (en) | Communication signal modulation and identification method for quantum image group mechanism evolution BP neural network | |
| CN112364729A (en) | Modulation identification method based on characteristic parameters and BP neural network | |
| CN112464837A (en) | Shallow sea underwater acoustic communication signal modulation identification method and system based on small data samples | |
| CN109525369B (en) | Channel coding type blind identification method based on recurrent neural network | |
| CN112910811B (en) | Blind modulation identification method and device under unknown noise level condition based on joint learning | |
| CN114268388B (en) | Channel estimation method based on improved GAN network in large-scale MIMO | |
| CN113055107B (en) | Interference strategy generation method for radio station with unknown communication mode | |
| CN110929842B (en) | Accurate intelligent detection method for non-cooperative radio signal burst time region | |
| CN114337880B (en) | Spectrum sensing method considering random arrival and departure of main user signals | |
| Zhang et al. | Open set recognition of communication signal modulation based on deep learning | |
| CN114943246A (en) | Semi-supervised specific radiation source identification method based on bispectrum feature extraction | |
| CN115982613A (en) | Signal modulation identification system and method based on improved convolutional neural network | |
| CN115392285A (en) | Deep learning signal individual recognition model defense method based on multiple modes | |
| CN105959004A (en) | Single-precision ADC adaptive threshold quantification method based on large-scale MIMO | |
| CN115270891A (en) | Method, device, equipment and storage medium for generating signal countermeasure sample | |
| CN116680608A (en) | Signal modulation identification method based on complex graph convolutional neural network | |
| CN116243248A (en) | Multi-component interference signal identification method based on multi-label classification network | |
| Zhou et al. | Msom based automatic modulation recognition and demodulation |
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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20221031 Address after: 264001 No. two, 188 Road, Shandong, Yantai Applicant after: NAVAL AVIATION UNIVERSITY OF PLA Address before: 264001 No. two, 188 Road, Shandong, Yantai Applicant before: School of Aeronautical foundation, Naval Aeronautical University PLA |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |