CN116910632B - Radio frequency fingerprint identification method, device, equipment and medium based on transfer learning - Google Patents
Radio frequency fingerprint identification method, device, equipment and medium based on transfer learning Download PDFInfo
- Publication number
- CN116910632B CN116910632B CN202311183840.1A CN202311183840A CN116910632B CN 116910632 B CN116910632 B CN 116910632B CN 202311183840 A CN202311183840 A CN 202311183840A CN 116910632 B CN116910632 B CN 116910632B
- Authority
- CN
- China
- Prior art keywords
- usrp
- training
- model
- domain
- source domain
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 59
- 238000013526 transfer learning Methods 0.000 title claims abstract description 42
- 238000012549 training Methods 0.000 claims abstract description 141
- 230000002457 bidirectional effect Effects 0.000 claims abstract description 60
- 238000013508 migration Methods 0.000 claims description 98
- 230000005012 migration Effects 0.000 claims description 98
- 238000012360 testing method Methods 0.000 claims description 48
- 238000004590 computer program Methods 0.000 claims description 22
- 230000008569 process Effects 0.000 claims description 20
- 238000012795 verification Methods 0.000 claims description 16
- 238000012545 processing Methods 0.000 claims description 15
- 238000007781 pre-processing Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 11
- 230000000694 effects Effects 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 9
- 230000001788 irregular Effects 0.000 claims description 7
- 238000002372 labelling Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 6
- 230000008014 freezing Effects 0.000 claims description 3
- 238000007710 freezing Methods 0.000 claims description 3
- 238000012546 transfer Methods 0.000 abstract description 5
- 230000009977 dual effect Effects 0.000 abstract 1
- 238000004891 communication Methods 0.000 description 8
- 238000013135 deep learning Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0475—Generative 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
- G06N3/094—Adversarial learning
-
- 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
- G06N3/096—Transfer learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The application provides a radio frequency fingerprint identification method, a device, equipment and a medium based on Transfer learning, which adopt bidirectional Transfer to generate a countermeasure network (Dual Transfer-GAN, DTGAN), use a model parameter Transfer method to reserve weight parameters of a pre-training model shallow convolution structure, and perform fine adjustment on a final classification layer so as to align the distribution of target domain data and adapt to radio frequency fingerprint identification tasks under a variable environment. By utilizing the idea of transfer learning, under the influence of multiple factors such as individual cross-acquisition equipment and cross-acquisition equipment characteristics, the directions of transfer learning from a model domain and data domain are respectively changed, the problems of insufficient label samples and sample characteristic space deviation are solved, and the recognition performance robustness of the recognition model under different scenes is improved.
Description
Technical Field
The application relates to the technical field of communication, in particular to a radio frequency fingerprint identification method, a device, equipment and a medium based on transfer learning.
Background
Currently, wireless networks bring subversive changes to society, which have become an integral part of everyday life. However, the explosive growth of wireless application demands and the number of wireless devices inevitably raise security and privacy concerns, and authentication of the identity of the wireless devices is urgent.
The ideal case of radio frequency fingerprint identification is mostly based on the fact that the sample feature space is of the same distribution. Through characteristic offset analysis, different acquisition devices acquire the same radio frequency emission source, the acquired radio frequency fingerprint characteristic distribution is mostly the same, and a small part of characteristic distribution has certain deviation, so that the recognition model is not sensitive enough to the characteristic distribution difference, and the recognition performance of the pre-training model is reduced, and therefore, the performance of the pre-training recognition model can be influenced by a changed electromagnetic environment. At present, wireless device identification based on physical layer radio frequency fingerprints often faces insufficient label samples and sample characteristic space deviation, and the wireless device individual identification problem under the condition of research and transfer learning is significant.
At present, the academic circles at home and abroad have theoretically proved that the transfer learning can solve the problem of data domain mismatch in the deep learning training process. Mainly because the feature differences between the data can lead to degradation of the source model in different scenarios. The idea of transfer learning is to rely on models with excellent performance, transfer parameters of the models to specific tasks, and selectively adjust deep parameters of a network aiming at task data sets, because shallow characteristics of the network are kept unchanged in a training process, the deep parameters of the network are transitive, and the optimization of transfer learning is focused on the deep parameters of the network, so that the network can capture common data characteristics of different domains, and the model can be more suitable for tasks of a target domain.
However, in most machine learning and deep learning application scenarios, new data may be obtained under a new learning task, resulting in the fact that the pre-training model obtained by training the source data is not applicable to the new data, and the time cost and the computational cost of constantly repeating the construction of the new task model are expensive. In addition, when using deep learning networks, the training signal data and the test signal data originate from different acquisition conditions and channel transmission conditions, resulting in characteristic deviations of the training and test data.
In view of this, the present application has been proposed.
Disclosure of Invention
In view of the above, the present application aims to provide a method, an apparatus, a device and a medium for recognizing a radio frequency fingerprint based on transfer learning, which can effectively solve the problem that in the prior art, in most machine learning and deep learning application scenarios, new data may be obtained under a new learning task, so that a pre-training model obtained by training source data is not applicable to the new data, and the time cost and the computational cost for continuously and repeatedly constructing the new task model are expensive. In addition, when using a deep learning network, the training signal data and the test signal data originate from different acquisition conditions and channel transmission conditions, resulting in the problem of characteristic deviation of the training and test data.
The application discloses a radio frequency fingerprint identification method based on transfer learning, which comprises the following steps:
acquiring a USRP signal, preprocessing the USRP signal, and generating double-domain training set signal data and a target domain USRP test signal, wherein the USRP signal comprises a source domain USRP acquisition signal and a target domain USRP acquisition signal, and the double-domain training set signal data comprises a source domain data set and a target domain training set;
creating a bidirectional migration generation countermeasure network model, and performing alignment enhancement processing of training radio frequency fingerprint characteristics on the bidirectional migration generation countermeasure network model according to the two-domain training set signal data to complete radio frequency fingerprint migration based on the data characteristics, wherein the method specifically comprises the following steps:
constructing a bidirectional migration generation countermeasure network model on the basis of a CycleGAN bidirectional structure model, and replacing a transpose convolution process of the CycleGAN bidirectional structure model with an up-sampling and convolution process;
adding an Identity loss function to limit generator irregular migration;
creating a source domain pre-training LRFN model, inputting the source domain USRP acquisition signal into the source domain pre-training LRFN model for verification, storing the parameters of the source domain pre-training LRFN model after verification as a pth file, and loading the source domain pre-training LRFN model for fine tuning at the same time, wherein the method specifically comprises the following steps:
inputting the source domain USRP acquisition signals into the bidirectional migration generation countermeasure network model for training, and generating a source domain pre-training LRFN model;
inputting the source domain USRP acquisition signal into the source domain pre-training LRFN model for verification, and storing the verified parameters in the source domain pre-training LRFN model as a pth file;
performing overall bidirectional migration generation countermeasure network model parameter training until the bidirectional migration generation countermeasure network model loss converges, and the prediction effect in the source domain reaches a preset value;
performing DTGAN feature migration test on the counternetwork model according to the target domain USRP test signal and the bidirectional migration generation, and generating a feature migration test signal;
and inputting the source domain USRP acquisition signal into the source domain pretraining LRFN model for preprocessing, and generating a characteristic migration test signal and an identification result.
Preferably, the USRP signal is collected and preprocessed to generate the double-domain training set signal data and the target domain USRP test signal, specifically:
acquiring a plurality of source domain USRP signals acquired by source domain USRP equipment, and performing label labeling processing on each source domain USRP signal to generate a source domain data set;
and acquiring a plurality of target domain USRP signals acquired by target domain USRP equipment, and performing label labeling processing on each target domain USRP signal to generate a target domain training set and target domain USRP test signals.
Preferably, the source domain pre-training LRFN model is loaded for fine tuning, specifically:
extracting the two-domain training set signal data to generate a target domain USRP sample with a label;
a long-short-time memory unit of the convolution unit module and the feature serialization optimization module of the source domain pre-training LRFN model is not frozen, and deep feature boundaries of target domain data are learned again on the basis of the pth file;
updating parameters of a last layer of classifier of the source domain pre-training LRFN model, freezing a convolution kernel unit or a long-short-term memory unit of the source domain pre-training LRFN model, and controlling the classifier to judge potential target domain features after the features are extracted.
The application also discloses a radio frequency fingerprint identification device based on transfer learning, which comprises:
the signal acquisition unit is used for acquiring a USRP signal, preprocessing the USRP signal and generating double-domain training set signal data and a target domain USRP test signal, wherein the USRP signal comprises a source domain USRP acquisition signal and a target domain USRP acquisition signal, and the double-domain training set signal data comprises a source domain data set and a target domain training set;
the bidirectional migration generation countermeasure network model unit is used for creating a bidirectional migration generation countermeasure network model, carrying out alignment enhancement processing of training radio frequency fingerprint characteristics on the bidirectional migration generation countermeasure network model according to the two-domain training set signal data, and completing radio frequency fingerprint migration based on the data characteristics, and specifically comprises the following steps:
constructing a bidirectional migration generation countermeasure network model on the basis of a CycleGAN bidirectional structure model, and replacing a transpose convolution process of the CycleGAN bidirectional structure model with an up-sampling and convolution process;
adding an Identity loss function to limit generator irregular migration;
a source domain pre-training LRFN model unit, configured to create a source domain pre-training LRFN model, input the source domain USRP acquisition signal into the source domain pre-training LRFN model for verification, store the verified parameters of the source domain pre-training LRFN model as a pth file, and load the source domain pre-training LRFN model for fine tuning at the same time, where the method specifically includes:
inputting the source domain USRP acquisition signals into the bidirectional migration generation countermeasure network model for training, and generating a source domain pre-training LRFN model;
inputting the source domain USRP acquisition signal into the source domain pre-training LRFN model for verification, and storing the verified parameters in the source domain pre-training LRFN model as a pth file;
performing overall bidirectional migration generation countermeasure network model parameter training until the bidirectional migration generation countermeasure network model loss converges, and the prediction effect in the source domain reaches a preset value;
the feature migration unit is used for performing DTGAN feature migration test on the countermeasure network model according to the target domain USRP test signal and the bidirectional migration generation, and generating a feature migration test signal;
and the identification unit is used for inputting the source domain USRP acquisition signal into the source domain pretraining LRFN model for preprocessing, and generating a characteristic migration test signal and an identification result.
The application also discloses a radio frequency fingerprint identification device based on the transfer learning, which comprises a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, wherein the radio frequency fingerprint identification method based on the transfer learning is realized when the processor executes the computer program.
The application also discloses a readable storage medium, which stores a computer program, the computer program can be executed by a processor of a device where the storage medium is located, so as to realize the radio frequency fingerprint identification method based on transfer learning.
In summary, according to the radio frequency fingerprint identification method, device, equipment and medium based on transfer learning provided by the embodiment, through the transfer learning thought and the source model training of putting a small amount of target domain samples in the DTGAN model domain, the identification rate of global domain parameter updating is higher than that of directly inputting target domain test data, and the radio frequency fingerprint identification of a small sample can be effectively realized. In the data domain direction, combining with the deep migration learning idea, a DTGAN network algorithm suitable for radio frequency signal data is provided, and the signal enhancement processing can be carried out on signals with characteristic offset, so that the characteristic boundary of the target domain signal fits the source domain signal, and the stability of classifier identification under the characteristic offset is improved.
Drawings
Fig. 1 is a schematic flow chart of a radio frequency fingerprint identification method based on transfer learning according to an embodiment of the present application.
Fig. 2 is a schematic diagram of a DTGAN network flow structure of a radio frequency fingerprint identification method based on transfer learning according to an embodiment of the present application.
Fig. 3 is a schematic diagram of transpose convolution alignment of a radio frequency fingerprint identification method based on transfer learning according to an embodiment of the present application.
Fig. 4 is a schematic diagram of a DTGAN feature migration simulation flow of a radio frequency fingerprint identification method based on migration learning according to an embodiment of the present application.
Fig. 5 is a schematic representation of a characteristic migration experiment result of a cross-acquisition device of a radio frequency fingerprint identification method based on migration learning according to an embodiment of the present application.
Fig. 6 is a schematic block diagram of a radio frequency fingerprint identification device based on transfer learning according to an embodiment of the present application.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present application more apparent, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, based on the embodiments of the application, which are apparent to those of ordinary skill in the art without inventive faculty, are intended to be within the scope of the application. Thus, the following detailed description of the embodiments of the application, as presented in the figures, is not intended to limit the scope of the application, as claimed, but is merely representative of selected embodiments of the application. All other embodiments, based on the embodiments of the application, which are apparent to those of ordinary skill in the art without inventive faculty, are intended to be within the scope of the application.
Specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.
Referring to fig. 1 to 3, a first embodiment of the present application provides a method for recognizing a radio frequency fingerprint based on transfer learning, which may be performed by a radio frequency fingerprint recognition device (hereinafter, recognition device), in particular, by one or more processors within the recognition device, to implement the following steps:
in this embodiment, the identification device may be a user terminal device (such as a smart phone, a smart computer or other smart devices), and the user terminal device may establish a communication connection with a cloud server to implement data interaction.
S101, acquiring a USRP signal, preprocessing the USRP signal, and generating double-domain training set signal data and a target domain USRP test signal, wherein the USRP signal comprises a source domain USRP acquisition signal and a target domain USRP acquisition signal, and the double-domain training set signal data comprises a source domain data set and a target domain training set;
specifically, step S101 includes: acquiring a plurality of source domain USRP signals acquired by source domain USRP equipment, and performing label labeling processing on each source domain USRP signal to generate a source domain data set;
and acquiring a plurality of target domain USRP signals acquired by target domain USRP equipment, and performing label labeling processing on each target domain USRP signal to generate a target domain training set and target domain USRP test signals.
Different acquisition devices on the market acquire the same radio frequency emission source, the acquired radio frequency fingerprint feature distribution is mostly the same, and a small part of the feature distribution has a certain deviation, so that the recognition model is not sensitive enough to the feature distribution difference, and the recognition performance of the pre-training model is reduced, and therefore, the performance of the pre-training recognition model can be influenced by a changed electromagnetic environment.
Specifically, in this embodiment, signals are collected and preprocessed; the problem of insufficient target domain data samples under an individual cross-collection device is simulated, so that two batches of data are respectively collected, different USRP individuals of the same USRP B200 model are used for actually collecting the Internet of things module, the two collection devices are defined as a source domain USRP and a target domain USRP, the source domain USRP device collects enough signal data and has corresponding label labels to form a source domain data set, the target domain USRP takes a small number of samples of each Internet of things device and labels as a target domain training set, and the enough target domain signal data is collected as a test set to simulate a scene with few label samples under a complex communication environment.
In this embodiment, in the performance analysis experiment of DTGAN, two different motherboards with the model USRP B200 are adopted as the signals of different software radio acquisition devices and 18 wireless communication modules with the same model of the internet of things, and after the signals are detected by the starting point, the signals are directly input into the model for training and recognition, and the sampling length reserved for each signal is 14400. Each transmitting board signal acquires 1000 training set samples for training a DTGAN characteristic migration network, 500 target domain test set samples for testing the migration effect of the DTGAN, and the migrated signals are put into a source domain pre-training model for recognition. The DTGAN model works based on the Pytorch framework, the optimization algorithm adopts Adam, the learning rate is set to be 0.001, and the iteration times are set to be 100 in the training process. 3000 signals were randomly sampled as training data for each wireless device from the total signal sample library without overlapping with the test data. Different USRP B200 software radio equipment individuals are used for acquiring radio frequency signals of the same Internet of things equipment in the same acquisition environment, and a radio frequency characteristic offset experiment is carried out. The two USRP devices also collect signals of a plurality of radio frequency communication individuals to simulate multi-classification recognition, so that the data sets are divided into a source domain data set and a target domain data set, the two domains both comprise 18 signals of the same type of wireless communication module of the internet of things, 1000 training set samples are collected for training a DTGAN characteristic migration network by each transmitting plate signal, 500 target domain test set samples are used for testing the migration effect of the DTGAN, and the migrated signals are put into a source domain pre-training model for recognition.
S102, creating a bidirectional migration generation countermeasure network model, and performing alignment enhancement processing of training radio frequency fingerprint characteristics on the bidirectional migration generation countermeasure network model according to the two-domain training set signal data to finish radio frequency fingerprint migration based on the data characteristics;
specifically, step S102 includes: constructing a bidirectional migration generation countermeasure network model on the basis of a CycleGAN bidirectional structure model, and replacing a transpose convolution process of the CycleGAN bidirectional structure model with an up-sampling and convolution process;
an Identity loss function is added to limit generator irregular migration.
Specifically, in this embodiment, a two-way migration generation countermeasure network (DTGAN) is designed, and the network model is mainly used for improving the up-sampling process of the reconstruction features, performing alignment enhancement of the radio frequency fingerprint features on the data domain, and completing the radio frequency fingerprint migration based on the data features. Specifically, on the basis of a CycleGAN bidirectional structure, the transpose convolution process is replaced by an up-sampling and convolution process, and identity loss is added to limit irregular migration of the generator, so that the generator is prevented from modifying potential characteristic distribution of the original domain radio frequency data.
S103, creating a source domain pre-training LRFN model, inputting the source domain USRP acquisition signal into the source domain pre-training LRFN model for verification, storing the parameters of the source domain pre-training LRFN model after verification as a pth file, and loading the source domain pre-training LRFN model for fine adjustment;
specifically, step S103 includes: inputting the source domain USRP acquisition signals into the bidirectional migration generation countermeasure network model for training, and generating a source domain pre-training LRFN model;
inputting the source domain USRP acquisition signal into the source domain pre-training LRFN model for verification, and storing the verified parameters in the source domain pre-training LRFN model as a pth file;
and carrying out overall bidirectional migration generation countermeasure network model parameter training until the bidirectional migration generation countermeasure network model loss converges, and the prediction effect in the source domain reaches a preset value.
Extracting the two-domain training set signal data to generate a target domain USRP sample with a label;
a long-short-time memory unit of the convolution unit module and the feature serialization optimization module of the source domain pre-training LRFN model is not frozen, and deep feature boundaries of target domain data are learned again on the basis of the pth file;
updating parameters of a last layer of classifier of the source domain pre-training LRFN model, freezing a convolution kernel unit or a long-short-term memory unit of the source domain pre-training LRFN model, and controlling the classifier to judge potential target domain features after the features are extracted.
Specifically, in this embodiment, the source domain model is trained and verified, a source training set is input into a network for training to obtain a source domain model, a test set is input for verifying the source model, model parameters are saved as pth files after the source domain pre-training model is verified, and global network model parameter training is performed until model loss converges and a source domain prediction effect is good.
In this embodiment, the pre-training model is loaded for fine tuning, and the training set data is changed into a small number of labeled target domain USRP samples, so that the learning rate can be improved due to the fact that the source model has the capability of extracting shallow features, so that the model can learn classification feature boundaries more quickly. The model migration process is divided into global domain model migration and local domain model migration by the experiment, a long-short-term memory unit of a global domain migration freezing-free convolution unit module and a characteristic serialization optimization module further learns deep characteristic boundaries of target domain data on source model parameters, and local domain updating is to update parameters of a last layer of classifier only, freeze a convolution kernel unit or a long-short-term memory unit, and directly enable the classifier to judge potential target domain characteristics after extracting characteristics.
Referring to fig. 4 to 5, S104 performs a DTGAN feature migration test on the target domain USRP test signal and the bidirectional migration generation countermeasure network model, and generates a feature migration test signal;
s105, inputting the source domain USRP acquisition signal into the source domain pretraining LRFN model for preprocessing, and generating a characteristic migration test signal and an identification result.
Specifically, in this embodiment, retraining and target domain migration are performed, and a predictive test is performed on the migrated model using target domain data, so as to verify the validity of the model migration for radio frequency identification under a small number of target domain samples.
In summary, the radio frequency fingerprint identification method based on Transfer learning adopts a two-way Transfer generation countermeasure network (DTGAN) to realize radio frequency fingerprint identification of wireless signals under the conditions of insufficient label samples and sample characteristic space offset, and uses a model parameter Transfer method to reserve weight parameters of a shallow convolution structure of a pre-training model to finely adjust a final classification layer so as to align the distribution of target domain data and adapt to radio frequency fingerprint identification tasks under a variable environment. According to the method, the problem of insufficient label samples and sample characteristic space deviation is solved by utilizing the idea of transfer learning in the directions of model domain transfer learning and data domain transfer learning under the influence of multiple factors such as cross-acquisition equipment individuals and cross-acquisition equipment characteristics, and the recognition performance robustness of the recognition model under different scenes is improved. The radio frequency fingerprint identification method based on transfer learning can effectively solve the problems of insufficient label samples and sample characteristic space deviation, and improves the identification performance robustness of the identification model in different scenes.
Referring to fig. 6, a second embodiment of the present application provides a radio frequency fingerprint identification device based on transfer learning, including:
the signal acquisition unit 201 is configured to acquire a USRP signal, and pre-process the USRP signal to generate dual-domain training set signal data and a target domain USRP test signal, where the USRP signal includes a source domain USRP acquisition signal and a target domain USRP acquisition signal, and the dual-domain training set signal data includes a source domain data set and a target domain training set;
a bidirectional migration generating countermeasure network model unit 202, configured to create a bidirectional migration generating countermeasure network model, and perform alignment enhancement processing of training radio frequency fingerprint features on the bidirectional migration generating countermeasure network model according to the two-domain training set signal data, so as to complete radio frequency fingerprint migration based on data features;
a source domain pre-training LRFN model unit 203, configured to create a source domain pre-training LRFN model, input the source domain USRP acquisition signal into the source domain pre-training LRFN model for verification, store the parameters of the source domain pre-training LRFN model after verification as a pth file, and load the source domain pre-training LRFN model for fine tuning;
the feature migration unit 204 is configured to perform a DTGAN feature migration test on the counternetwork model according to the target domain USRP test signal and the bidirectional migration generation, and generate a feature migration test signal;
the identifying unit 205 is configured to input the source domain USRP acquisition signal into the source domain pre-training LRFN model for preprocessing, and generate a feature migration test signal and an identification result.
Preferably, the bidirectional migration generation countermeasure network model unit 202 is specifically configured to:
constructing a bidirectional migration generation countermeasure network model on the basis of a CycleGAN bidirectional structure model, and replacing a transpose convolution process of the CycleGAN bidirectional structure model with an up-sampling and convolution process;
an Identity loss function is added to limit generator irregular migration.
Preferably, the source domain pre-training LRFN model unit 203 is specifically configured to:
inputting the source domain USRP acquisition signals into the bidirectional migration generation countermeasure network model for training, and generating a source domain pre-training LRFN model;
inputting the source domain USRP acquisition signal into the source domain pre-training LRFN model for verification, and storing the verified parameters in the source domain pre-training LRFN model as a pth file;
and carrying out overall bidirectional migration generation countermeasure network model parameter training until the bidirectional migration generation countermeasure network model loss converges, and the prediction effect in the source domain reaches a preset value.
A third embodiment of the present application provides a radio frequency fingerprint identification device based on transfer learning, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, the processor implementing the radio frequency fingerprint identification method based on transfer learning as described in any one of the above when executing the computer program.
A fourth embodiment of the present application provides a readable storage medium storing a computer program executable by a processor of a device in which the storage medium is located to implement a radio frequency fingerprint identification method based on transfer learning as described in any one of the above.
Illustratively, the computer programs described in the third and fourth embodiments of the present application may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present application. The one or more modules may be a series of computer program instruction segments capable of performing specific functions for describing the execution of the computer program in the transfer learning based radio frequency fingerprinting device. For example, the device described in the second embodiment of the present application.
The processor may be a central processing unit (Central Processing Unit, CPU), other general purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. The general processor may be a microprocessor or the processor may be any conventional processor, etc., where the processor is a control center of the rf fingerprint identification method based on the transfer learning, and various interfaces and lines are used to connect the various parts of the overall rf fingerprint identification method based on the transfer learning.
The memory may be used to store the computer program and/or the module, and the processor may implement various functions of the radio frequency fingerprint identification method based on transfer learning by running or executing the computer program and/or the module stored in the memory and invoking data stored in the memory. The memory may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program (such as a sound playing function, a text conversion function, etc.) required for at least one function, and the like; the storage data area may store data (such as audio data, text message data, etc.) created according to the use of the cellular phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as a hard disk, memory, plug-in hard disk, smart Media Card (SMC), secure Digital (SD) Card, flash Card (Flash Card), at least one disk storage device, flash memory device, or other volatile solid-state storage device.
Wherein the modules may be stored in a computer readable storage medium if implemented in the form of software functional units and sold or used as a stand alone product. Based on this understanding, the present application may implement all or part of the flow of the method of the above embodiment, or may be implemented by a computer program to instruct related hardware, where the computer program may be stored in a computer readable storage medium, and the computer program may implement the steps of each method embodiment described above when executed by a processor. Wherein the computer program comprises computer program code which may be in source code form, object code form, executable file or some intermediate form etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer Memory, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), an electrical carrier signal, a telecommunications signal, a software distribution medium, and so forth. It should be noted that the computer readable medium contains content that can be appropriately scaled according to the requirements of jurisdictions in which such content is subject to legislation and patent practice, such as in certain jurisdictions in which such content is subject to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.
It should be noted that the above-described apparatus embodiments are merely illustrative, and the units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the drawings of the embodiment of the device provided by the application, the connection relation between the modules represents that the modules have communication connection, and can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art will understand and implement the present application without undue burden.
The above is only a preferred embodiment of the present application, and the protection scope of the present application is not limited to the above examples, and all technical solutions belonging to the concept of the present application belong to the protection scope of the present application.
Claims (6)
1. The radio frequency fingerprint identification method based on transfer learning is characterized by comprising the following steps:
acquiring a USRP signal, preprocessing the USRP signal, and generating double-domain training set signal data and a target domain USRP test signal, wherein the USRP signal comprises a source domain USRP acquisition signal and a target domain USRP acquisition signal, and the double-domain training set signal data comprises a source domain data set and a target domain training set;
creating a bidirectional migration generation countermeasure network model, and performing alignment enhancement processing of training radio frequency fingerprint characteristics on the bidirectional migration generation countermeasure network model according to the two-domain training set signal data to complete radio frequency fingerprint migration based on the data characteristics, wherein the method specifically comprises the following steps:
constructing a bidirectional migration generation countermeasure network model on the basis of a CycleGAN bidirectional structure model, and replacing a transpose convolution process of the CycleGAN bidirectional structure model with an up-sampling and convolution process;
adding an Identity loss function to limit generator irregular migration;
creating a source domain pre-training LRFN model, inputting the source domain USRP acquisition signal into the source domain pre-training LRFN model for verification, storing the parameters of the source domain pre-training LRFN model after verification as a pth file, and loading the source domain pre-training LRFN model for fine tuning at the same time, wherein the method specifically comprises the following steps:
inputting the source domain USRP acquisition signals into the bidirectional migration generation countermeasure network model for training, and generating a source domain pre-training LRFN model;
inputting the source domain USRP acquisition signal into the source domain pre-training LRFN model for verification, and storing the verified parameters in the source domain pre-training LRFN model as a pth file;
performing overall bidirectional migration generation countermeasure network model parameter training until the bidirectional migration generation countermeasure network model loss converges, and the prediction effect in the source domain reaches a preset value;
performing DTGAN feature migration test on the counternetwork model according to the target domain USRP test signal and the bidirectional migration generation, and generating a feature migration test signal;
and inputting the source domain USRP acquisition signal into the source domain pretraining LRFN model for preprocessing, and generating a characteristic migration test signal and an identification result.
2. The method for identifying the radio frequency fingerprint based on the transfer learning according to claim 1, wherein the method is characterized by collecting USRP signals and preprocessing the USRP signals to generate double-domain training set signal data and target domain USRP test signals, and specifically comprises the following steps:
acquiring a plurality of source domain USRP signals acquired by source domain USRP equipment, and performing label labeling processing on each source domain USRP signal to generate a source domain data set;
and acquiring a plurality of target domain USRP signals acquired by target domain USRP equipment, and performing label labeling processing on each target domain USRP signal to generate a target domain training set and target domain USRP test signals.
3. The radio frequency fingerprint identification method based on transfer learning of claim 1, wherein loading the source domain pre-training LRFN model for fine tuning is specifically as follows:
extracting the two-domain training set signal data to generate a target domain USRP sample with a label;
a long-short-time memory unit of the convolution unit module and the feature serialization optimization module of the source domain pre-training LRFN model is not frozen, and deep feature boundaries of target domain data are learned again on the basis of the pth file;
updating parameters of a last layer of classifier of the source domain pre-training LRFN model, freezing a convolution kernel unit or a long-short-term memory unit of the source domain pre-training LRFN model, and controlling the classifier to judge potential target domain features after the features are extracted.
4. Radio frequency fingerprint identification device based on migration study, its characterized in that includes:
the signal acquisition unit is used for acquiring a USRP signal, preprocessing the USRP signal and generating double-domain training set signal data and a target domain USRP test signal, wherein the USRP signal comprises a source domain USRP acquisition signal and a target domain USRP acquisition signal, and the double-domain training set signal data comprises a source domain data set and a target domain training set;
the bidirectional migration generation countermeasure network model unit is used for creating a bidirectional migration generation countermeasure network model, carrying out alignment enhancement processing of training radio frequency fingerprint characteristics on the bidirectional migration generation countermeasure network model according to the two-domain training set signal data, and completing radio frequency fingerprint migration based on the data characteristics, and specifically comprises the following steps:
constructing a bidirectional migration generation countermeasure network model on the basis of a CycleGAN bidirectional structure model, and replacing a transpose convolution process of the CycleGAN bidirectional structure model with an up-sampling and convolution process;
adding an Identity loss function to limit generator irregular migration;
a source domain pre-training LRFN model unit, configured to create a source domain pre-training LRFN model, input the source domain USRP acquisition signal into the source domain pre-training LRFN model for verification, store the verified parameters of the source domain pre-training LRFN model as a pth file, and load the source domain pre-training LRFN model for fine tuning at the same time, where the method specifically includes:
inputting the source domain USRP acquisition signals into the bidirectional migration generation countermeasure network model for training, and generating a source domain pre-training LRFN model;
inputting the source domain USRP acquisition signal into the source domain pre-training LRFN model for verification, and storing the verified parameters in the source domain pre-training LRFN model as a pth file;
performing overall bidirectional migration generation countermeasure network model parameter training until the bidirectional migration generation countermeasure network model loss converges, and the prediction effect in the source domain reaches a preset value;
the feature migration unit is used for performing DTGAN feature migration test on the countermeasure network model according to the target domain USRP test signal and the bidirectional migration generation, and generating a feature migration test signal;
and the identification unit is used for inputting the source domain USRP acquisition signal into the source domain pretraining LRFN model for preprocessing, and generating a characteristic migration test signal and an identification result.
5. A radio frequency fingerprint identification device based on transfer learning, characterized by comprising a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, the processor implementing the radio frequency fingerprint identification method based on transfer learning as claimed in any one of claims 1 to 3 when executing the computer program.
6. A readable storage medium, characterized in that a computer program is stored, said computer program being executable by a processor of a device in which the storage medium is located, for implementing a radio frequency fingerprint identification method based on transfer learning as claimed in any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311183840.1A CN116910632B (en) | 2023-09-14 | 2023-09-14 | Radio frequency fingerprint identification method, device, equipment and medium based on transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311183840.1A CN116910632B (en) | 2023-09-14 | 2023-09-14 | Radio frequency fingerprint identification method, device, equipment and medium based on transfer learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116910632A CN116910632A (en) | 2023-10-20 |
| CN116910632B true CN116910632B (en) | 2023-12-12 |
Family
ID=88363370
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311183840.1A Active CN116910632B (en) | 2023-09-14 | 2023-09-14 | Radio frequency fingerprint identification method, device, equipment and medium based on transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116910632B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114332649A (en) * | 2022-03-07 | 2022-04-12 | 湖北大学 | Cross-scene remote sensing image depth countermeasure transfer learning method based on dual-channel attention mechanism |
| CN114936575A (en) * | 2022-05-07 | 2022-08-23 | 重庆大学 | Motor bearing fault diagnosis method based on sample generation and deep migration learning |
| CN116257750A (en) * | 2022-11-09 | 2023-06-13 | 南京大学 | Radio frequency fingerprint identification method based on sample enhancement and deep learning |
| CN116304820A (en) * | 2023-03-14 | 2023-06-23 | 广东石油化工学院 | Bearing fault type prediction method and system based on multi-source domain transfer learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US12061991B2 (en) * | 2020-09-23 | 2024-08-13 | International Business Machines Corporation | Transfer learning with machine learning systems |
-
2023
- 2023-09-14 CN CN202311183840.1A patent/CN116910632B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114332649A (en) * | 2022-03-07 | 2022-04-12 | 湖北大学 | Cross-scene remote sensing image depth countermeasure transfer learning method based on dual-channel attention mechanism |
| CN114936575A (en) * | 2022-05-07 | 2022-08-23 | 重庆大学 | Motor bearing fault diagnosis method based on sample generation and deep migration learning |
| CN116257750A (en) * | 2022-11-09 | 2023-06-13 | 南京大学 | Radio frequency fingerprint identification method based on sample enhancement and deep learning |
| CN116304820A (en) * | 2023-03-14 | 2023-06-23 | 广东石油化工学院 | Bearing fault type prediction method and system based on multi-source domain transfer learning |
Non-Patent Citations (1)
| Title |
|---|
| 无线发射机信号指纹的形成机理及特征识别研究;赵彩丹;中国硕士学位论文;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116910632A (en) | 2023-10-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110798417A (en) | Signal modulation identification method and device based on cyclic residual error network | |
| CN110427893A (en) | A kind of specific emitter identification method, apparatus and computer storage medium based on convolutional neural networks | |
| CN110046622B (en) | Targeted attack sample generation method, device, equipment and storage medium | |
| CN110929806B (en) | Picture processing method and device based on artificial intelligence and electronic equipment | |
| CN112785086A (en) | Credit overdue risk prediction method and device | |
| CN112231416B (en) | Knowledge graph body updating method and device, computer equipment and storage medium | |
| CN112926654A (en) | Pre-labeling model training and certificate pre-labeling method, device, equipment and medium | |
| CN112685787B (en) | Big data information security protection method applied to artificial intelligence and cloud server | |
| CN110781856B (en) | Heterogeneous face recognition model training method, face recognition method and related device | |
| CN113673260A (en) | Model processing method, device, storage medium and processor | |
| CN116992299A (en) | Training method, detecting method and device of blockchain transaction anomaly detection model | |
| US11881052B2 (en) | Face search method and apparatus | |
| US20220146707A1 (en) | Temperature prediction system | |
| CN116910632B (en) | Radio frequency fingerprint identification method, device, equipment and medium based on transfer learning | |
| CN113886821A (en) | Malicious process identification method and device based on twin network, electronic equipment and storage medium | |
| CN113283388A (en) | Training method, device and equipment of living human face detection model and storage medium | |
| CN112820404A (en) | Information processing method applied to big data intelligent medical treatment and intelligent medical treatment server | |
| CN117009539A (en) | Entity alignment method, device, equipment and storage medium of knowledge graph | |
| CN114118411A (en) | Training method of image recognition network, image recognition method and device | |
| CN113627514A (en) | Data processing method and device of knowledge graph, electronic equipment and storage medium | |
| CN116778534B (en) | Image processing method, device, equipment and medium | |
| CN111126503A (en) | Training sample generation method and device | |
| CN117058498B (en) | Training method of segmentation map evaluation model, and segmentation map evaluation method and device | |
| CN114648978B (en) | Voice verification processing method and related device | |
| US20240054346A1 (en) | Systems and methods for simultaneous network pruning and parameter optimization |
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 |