CN114943253A - Radio frequency fingerprint small sample identification method based on meta-learning model - Google Patents
Radio frequency fingerprint small sample identification method based on meta-learning model Download PDFInfo
- Publication number
- CN114943253A CN114943253A CN202210549015.8A CN202210549015A CN114943253A CN 114943253 A CN114943253 A CN 114943253A CN 202210549015 A CN202210549015 A CN 202210549015A CN 114943253 A CN114943253 A CN 114943253A
- Authority
- CN
- China
- Prior art keywords
- sample
- training
- model
- task
- function
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 56
- 230000006870 function Effects 0.000 claims description 46
- 239000013598 vector Substances 0.000 claims description 15
- 238000005070 sampling Methods 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 abstract description 2
- 238000004519 manufacturing process Methods 0.000 abstract 1
- 238000012360 testing method Methods 0.000 description 11
- 238000004891 communication Methods 0.000 description 10
- 238000010801 machine learning Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
Images
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Probability & Statistics with Applications (AREA)
- Signal Processing (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of neural network and target identification, and particularly relates to a radio frequency fingerprint small sample identification method based on a meta-learning model. The invention comprises the following steps: taking the original I/Q signal as a data set, and manufacturing a training task sample set; constructing a meta-learning model-matching network model; inputting the training task set into a matching network model, extracting radio frequency fingerprint characteristics from I/Q signals, measuring the similarity between samples, and training the model; and inputting the recognition task into the trained meta-learning model, and outputting the recognition result of the model. The method has better identification precision when facing a radio frequency fingerprint small sample data set, and avoids the problem of model overfitting caused by small data volume.
Description
Technical Field
The invention belongs to the technical field of neural network and target identification, and particularly relates to a radio frequency fingerprint small sample identification method based on a meta-learning model.
Background
In recent years, with the advent of the 5G communication era, wireless communication has been applied to various scenes such as smart homes, internet of things, internet of vehicles, smart cities, smart battles and the like, and plays an irreplaceable role in the civil and military fields. In contrast to wired networks, in wireless networks, security of communication between devices is typically implemented based on various security protocols at the network layer and the transport layer. This makes devices in the wireless network more vulnerable to attacks, such as the ip (internet protocol) address and key password used by the communication device in access authentication, which may be falsified. Therefore, new security mechanisms and device authentication and identification technologies are urgently needed for wireless communication to defend against potential threats of wireless devices in networks.
Like a person's fingerprint, the rf fingerprint of each device is unique, and the signals sent by different wireless communication devices vary due to hardware differences. By analyzing the small difference of the radio frequency signals, the extracted hardware characteristic is the radio frequency fingerprint of the equipment, and the hardware characteristic can be used for equipment identification.
At present, the existing radio frequency fingerprint identification technology mainly trains a large amount of original I/Q signal data sets of wireless communication equipment through traditional machine learning or deep learning, extracts signal characteristics, namely radio frequency fingerprints, trains to obtain a classifier, and the classifier is used for confirming the identity of the wireless communication equipment. However, the method is weak in the case of a small sample, and especially in some special environments, such as a satellite and other communication devices in a relatively hidden environment, a large number of signals of the communication devices cannot be acquired for radio frequency fingerprint feature extraction. Traditional machine learning and deep learning methods are also no longer suitable because they are aimed at data-intensive applications, and when facing small data sets, too few training samples are prone to problems such as low algorithm accuracy, overfitting, and the like.
A machine learning method, small sample learning, is proposed in recent years. Through the existing priori knowledge, the small sample learning can quickly generalize the model to a new task with only a few labeled samples, and helps solve the problem of machine learning when facing a small sample data set. And meta learning is one of the methods for implementing small sample learning. In addition, most of the existing small sample learning is realized based on image recognition, but the research in the field of radio frequency fingerprint recognition is less. Therefore, an effective radio frequency fingerprint small sample identification method is found to have great research value.
Disclosure of Invention
Aiming at the defects in the prior art, the radio frequency fingerprint small sample identification method based on the meta-learning model provided by the invention solves the problems in the prior art.
The technical scheme of the invention is as follows:
a radio frequency fingerprint small sample identification method based on a meta-learning model comprises the following steps:
s1, making a training sample, specifically: randomly selecting k categories from the acquired radio equipment I/Q signals, randomly selecting s + Q samples for each category, taking k & lts & gt samples as a support set, taking k & ltq & gt samples as a query set, taking a group of support sets and the query set as a training task, and obtaining all training tasks required in a training stage as training samples through random sampling;
s2, constructing a meta-learning model, specifically comprising an embedding function g for processing a support set, an embedding function f for processing a query set and an attention kernel function a;
the embedding function g and the embedding function f have the same structure and respectively comprise four convolution modules with the same structure, each convolution module consists of a convolution layer, batch normalization and a nonlinear activation function, and input data are subjected to four times of convolution and then unfolded by using a Flatten function to obtain a feature vector extracted by the embedding function and used as the input of an attention kernel function; the method specifically comprises the following steps:
for each given support set S ═ x i ,y i } i＝1,2…n Where n-k-S represents the number of samples of the support set S, x i Represents the ith sample, y i Labels representing the ith sample, respectively learning a classifier
This classifier gives a sample of the specified query set
Time, output prediction tag
The classifier will support the weighted sum of the sample label similarity in the set as the output result:
the weight in the formula is the attention kernel function
Calculating a query set sample feature vector by using Euclidean distance
And support set sample feature vector g (x) i ) Similarity between them:
performing softmax normalization to obtain an attention kernel function calculation formula:
s3, inputting the training sample set obtained in the step S1 into the meta learning model constructed in the step S2 for training, specifically: for a training task process, firstly, respectively calculating the feature vectors of a training task support set and a query set sample by an embedded function, then calculating the weight between the task query sample and each support set sample by an attention kernel function, finally, carrying out weighted summation on the labels of all the support sets by combining the weights to obtain the probability distribution of the query set sample in the support set, continuously sampling a training task after the model is trained for one round, and carrying out the next round of training on the model; at the moment, updating parameters of the whole model, and performing gradient reduction on the parameters by adopting a loss function in a parameter updating mode to obtain a loss value of one round of training, and then reversely transmitting the loss value to bring the loss value into the next round of training to continuously optimize the parameters of the model until the model converges;
s4, obtaining an I/Q signal of the target device, sampling an identification task in the mode of the step S1, inputting the identification task into the trained model obtained in the step S3, then respectively calculating the feature vectors of the support set and the sample of the query set of the identification task by the embedding function, then calculating the weight between the query sample of the identification task and each sample of the support set by the attention kernel function, and finally carrying out weighted summation on the labels of all the support sets by combining the weight to obtain the probability distribution of the sample of the query set in the support set, thereby obtaining the classification result of the identification task.
The invention has the beneficial effects that:
(1) the invention applies the meta-learning method to the problem of small samples of the radio frequency fingerprint, avoids the problem of model overfitting and provides a new idea for the radio frequency fingerprint identification in some special scenes.
(2) In order to adapt to the characteristics of I/Q signals, the matching network model established by the invention adopts an Euclidean distance calculation formula to construct an attention kernel function, and the similarity between samples is more accurately compared, so that the identification accuracy of the model is improved.
(3) The method can be applied to the intelligent identification scene of the radio frequency fingerprint small sample, the model training speed is high, and the calculation cost of the required server is low.
Drawings
Fig. 1 is a flowchart of a method for identifying a small sample of a radio frequency fingerprint based on a meta-learning model according to the present invention.
FIG. 2 is a diagram illustrating a meta-learning training and testing task stage according to the present invention.
FIG. 3 is a diagram of the matching network model body architecture of the present invention.
Fig. 4 is a diagram of a matching network embedding function structure according to the present invention.
Detailed Description
The technical scheme of the invention is described in detail below with reference to the accompanying drawings:
the invention relates to an identification method based on a meta-learning model, and the basic concept of meta-learning is as follows:
meta Learning (Meta Learning) is just like the human brain, and can use the existing prior knowledge to learn a new task, so that the model has Learning ability and is one of the methods for solving the problem of small samples. Meta learning mainly has two stages. The first stage is a meta-training (meta-training) stage, a large number of tasks are trained, knowledge of each task is learned, and a parameter theta of a model F is obtained; in the second stage, namely the meta-testing stage, for a new task class, after gradient descent calculation updating iteration is performed by taking the parameter theta of the model F in the first stage as an initial parameter, a new task model can be learned according to the existing model, and then samples in the new task are classified and identified. Training tasks and testing tasks are used as training sets and testing sets, each task contains training and testing data, and in order to avoid confusion, the data sets in each task are called Support sets (Support sets) and Query sets (Query sets).
As shown in fig. 1, the identification method of the present invention includes the steps of:
s1, taking the original I/Q signal as a data set, and making a training task sample set;
s2, constructing a meta learning model-matching network model, as shown in FIG. 3;
s3, inputting the training task set into a meta-learning model, and training to obtain a model;
and S4, inputting the task to be recognized into the trained model to obtain the recognition result of the model.
The above method is described in detail with reference to specific examples below:
step S1 specifically includes:
s1.1, a data set D, namely original I/Q signal data transmitted by 12 devices is included, wherein the number of I/Q signals of each device is 600. Division into training task data sets D tain And a test task data set D test ，D tain I/Q signal data containing 8 devices, D test Contains the I/Q signal data of 4 devices. And the sample classes in the two data sets do not overlap.
S1.2, as shown in fig. 2, the meta-learning is generally divided into a meta-training (meta-training) phase and a meta-testing (meta-testing) phase, and the present embodiment will perform a 4-way 5-shot task experiment, that is, a support set of each task of a sample includes 4 categories, each category includes 5I/Q samples, and each category in a query set includes 15 samples.
From the data set D tain In the method, 4 categories are randomly selected, 1+15 samples are randomly selected for each category, 4 × 1 samples are used as a support set, 4 × 15 samples are used as a query set, a group of support sets and the query set are used as a training task, and all training tasks required in a training stage are obtained through random sampling and are used as training samples. And similarly, sampling the task to be identified according to the method.
Step S2 specifically includes:
s2.1, constructing an embedding function. As shown in fig. 4, the embedding function contains four convolution modules with the same structure, each module consisting of Conv + BatchNorm + ReLU (convolution layer + batch normalization + nonlinear activation function). The convolution layer of each module adopts convolution kernels with the size of 3 multiplied by 3, the filling of 1 and the step length of 1 as a filter, and a feature matrix with the output of 64 channels is obtained. Each convolution operation is followed by a normalization and nonlinear activation operation. The purpose of normalization is to prevent gradient disappearance and accelerate convergence speed, and ReLU activation is to better divide the feature space between different signals. After four times of convolution, using a Flatten function to expand to obtain a feature vector extracted by the embedding function, and using the feature vector as the input of the attention kernel function. Furthermore, since the support set is the same form as the I/Q signal samples input to the query set, the embedding functions g and f use the same structure. In addition, because the I/Q signals only have two paths and the parameter quantity of the whole network is not large, the pooling operation is not carried out on each layer of module so as to avoid losing useful information.
And S2.2, constructing an attention kernel function. For each given support set S ═ x i ,y i } i＝1,2…n Where n-k-S represents the number of samples in the support set S, x i Represents the ith sample, y i Labels representing the ith sample, respectively learning a classifier
This classifier gives a sample of the specified query set
Time, output prediction tag
The classifier will support the weighted sum of the sample label similarity in the set as the output result:
the weight in the formula is the attention kernel function
Calculating the sample characteristic direction of the query set by using Euclidean distanceMeasurement of
And support set sample feature vector g (x) i ) Similarity between them:
performing softmax normalization on the formula to obtain an attention kernel function calculation formula:
step S3 specifically includes:
and S3.1, loading the I/Q signals in the training task data set to a matching network model. The size of each I/Q signal sample is 2 x 128. To fit the model input, it is warped to 16 × 16 dimensions before input. The model will task sample the input data in the manner of step S2.
S3.2, setting the learning rate to be 0.001, the iteration times to be 50 and the initial channel number of the sample to be 1.
S3.3, conveying the preprocessed I/Q signals to a matching network model, for a training task process, firstly, respectively calculating the feature vectors of a training task support set and a query set sample by an embedded function, then, calculating the weight between the task query sample and each support set sample by an attention kernel function, finally, carrying out weighted summation on the labels of all the support sets by combining the weights to obtain the probability distribution of the query set sample in the support sets, continuously sampling the training task after the model is trained for one round, and carrying out the next round of training on the model; at the moment, the parameters of the whole model are updated, and the mode of updating the parameters adopts a loss function to carry out gradient reduction to obtain the loss value of one round of training, and then the loss value is propagated reversely to bring the loss value into the next round of training to continuously optimize the model parameters until the model converges.
Step S4 specifically includes:
and S4.1, taking the model parameters after the training in the meta-training stage as initialization parameters in the meta-testing stage.
And S4.2, loading the I/Q signals in the task data set to be identified to the matching network model. The size of each I/Q signal sample is 2 x 128. Also to fit the model input, it is warped to 16 x 16 size before input. The model will task sample the input data in the manner of step S2.
And S4.3, conveying the preprocessed I/Q signals to a trained matching network model, sampling an identification task, wherein the identification task also comprises a support set and a query set sample, inputting the identification task to the trained model obtained in the step S3, respectively calculating respective feature vectors of the support set and the query set sample of the identification task by an embedded function, then calculating weights between the query sample of the identification task and each support set sample by an attention kernel function, and finally performing weighted summation on labels of all support sets by combining the weights to obtain probability distribution of the query set sample in the support set, thereby obtaining a classification result of the identification task.
Claims (1)
1. A radio frequency fingerprint small sample identification method based on a meta-learning model is characterized by comprising the following steps:
s1, making a training sample, specifically: randomly selecting k categories from the acquired radio equipment I/Q signals, randomly selecting s + Q samples for each category, taking k & lts & gt samples as a support set, taking k & ltq & gt samples as a query set, taking a group of support sets and the query set as a training task, and obtaining all training tasks required in a training stage as training samples through random sampling;
s2, constructing a meta-learning model, specifically comprising an embedding function g for processing a support set, an embedding function f for processing a query set and an attention kernel function a;
the embedding function g and the embedding function f have the same structure and respectively comprise four convolution modules with the same structure, each convolution module consists of a convolution layer, batch normalization and a nonlinear activation function, and input data are subjected to four times of convolution and then unfolded by using a Flatten function to obtain a feature vector extracted by the embedding function and used as the input of an attention kernel function; the method specifically comprises the following steps:
for each given support set S ═ x i ,y i } i＝1,2…n Where n-k-S represents the number of samples of the support set S, x i Represents the ith sample, y i Labels representing the ith sample, respectively learning a classifier
This classifier gives a sample of the specified query set
Time, output prediction tag
The classifier will support the weighted sum of the sample label similarity in the set as the output result:
the weight in the formula is the attention kernel function
Calculating a query set sample feature vector by using Euclidean distance
And support set sample feature vector g (x) i ) Similarity between them:
performing softmax normalization to obtain an attention kernel function calculation formula:
s3, inputting the training sample set obtained in the step S1 into the meta learning model constructed in the step S2 for training, specifically: for a training task process, firstly, respectively calculating respective feature vectors of a training task support set and a query set sample by an embedded function, then calculating weights between the task query sample and each support set sample by an attention kernel function, finally, carrying out weighted summation on labels of all support sets by combining the weights to obtain probability distribution of the query set samples in the support sets, continuously sampling a training task after the model is trained for one round, and carrying out the next round of training on the model; at the moment, updating parameters of the whole model, namely performing gradient reduction on the parameters by adopting a loss function in a parameter updating mode to obtain a loss value of one round of training, and then reversely transmitting the loss value to bring the loss value into the next round of training to continuously optimize the parameters of the model until the model converges;
s4, obtaining an I/Q signal of the target device, sampling an identification task in the mode of the step S1, inputting the identification task into the trained model obtained in the step S3, then respectively calculating the feature vectors of the support set and the sample of the query set of the identification task by the embedding function, then calculating the weight between the query sample of the identification task and each sample of the support set by the attention kernel function, and finally carrying out weighted summation on the labels of all the support sets by combining the weight to obtain the probability distribution of the sample of the query set in the support set, thereby obtaining the classification result of the identification task.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210549015.8A CN114943253A (en) | 2022-05-20 | 2022-05-20 | Radio frequency fingerprint small sample identification method based on meta-learning model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210549015.8A CN114943253A (en) | 2022-05-20 | 2022-05-20 | Radio frequency fingerprint small sample identification method based on meta-learning model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114943253A true CN114943253A (en) | 2022-08-26 |
Family
ID=82909602
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210549015.8A Pending CN114943253A (en) | 2022-05-20 | 2022-05-20 | Radio frequency fingerprint small sample identification method based on meta-learning model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114943253A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115527269A (en) * | 2022-10-10 | 2022-12-27 | 动自由(北京)科技有限公司 | Intelligent human body posture image identification method and system |
| CN115879049A (en) * | 2023-03-02 | 2023-03-31 | 国网江西省电力有限公司电力科学研究院 | Induction identification output method and system of automatic vertical rod |
| CN116127298A (en) * | 2023-02-22 | 2023-05-16 | 北京邮电大学 | Small sample radio frequency fingerprint identification method based on triplet loss |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109961089A (en) * | 2019-02-26 | 2019-07-02 | 中山大学 | Small sample and zero sample image classification method based on metric learning and meta learning |
| CN114092747A (en) * | 2021-11-30 | 2022-02-25 | 南通大学 | Small sample image classification method based on depth element metric model mutual learning |
| CN114387627A (en) * | 2022-01-11 | 2022-04-22 | 厦门大学 | Small sample wireless device radio frequency fingerprint identification method and device based on depth measurement learning |
-
2022
- 2022-05-20 CN CN202210549015.8A patent/CN114943253A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109961089A (en) * | 2019-02-26 | 2019-07-02 | 中山大学 | Small sample and zero sample image classification method based on metric learning and meta learning |
| CN114092747A (en) * | 2021-11-30 | 2022-02-25 | 南通大学 | Small sample image classification method based on depth element metric model mutual learning |
| CN114387627A (en) * | 2022-01-11 | 2022-04-22 | 厦门大学 | Small sample wireless device radio frequency fingerprint identification method and device based on depth measurement learning |
Non-Patent Citations (1)
| Title |
|---|
| 孙晶明等: ""基于元学习的雷达小样本目标识别方法及改进"" * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115527269A (en) * | 2022-10-10 | 2022-12-27 | 动自由(北京)科技有限公司 | Intelligent human body posture image identification method and system |
| CN115527269B (en) * | 2022-10-10 | 2023-05-16 | 动自由(北京)科技有限公司 | Intelligent human body posture image recognition method and system |
| CN116127298A (en) * | 2023-02-22 | 2023-05-16 | 北京邮电大学 | Small sample radio frequency fingerprint identification method based on triplet loss |
| CN116127298B (en) * | 2023-02-22 | 2024-03-19 | 北京邮电大学 | Small sample radio frequency fingerprint identification method based on triplet loss |
| CN115879049A (en) * | 2023-03-02 | 2023-03-31 | 国网江西省电力有限公司电力科学研究院 | Induction identification output method and system of automatic vertical rod |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114943253A (en) | Radio frequency fingerprint small sample identification method based on meta-learning model | |
| Kang et al. | SAR image classification based on the multi-layer network and transfer learning of mid-level representations | |
| CN111160217B (en) | Method and system for generating countermeasure sample of pedestrian re-recognition system | |
| CN110929839B (en) | Method and device for training neural network, electronic equipment and computer storage medium | |
| CN111340850A (en) | Ground target tracking method of unmanned aerial vehicle based on twin network and central logic loss | |
| US11558810B2 (en) | Artificial intelligence radio classifier and identifier | |
| CN109446804B (en) | Intrusion detection method based on multi-scale feature connection convolutional neural network | |
| CN112215255A (en) | Training method of target detection model, target detection method and terminal equipment | |
| CN108564040B (en) | Fingerprint activity detection method based on deep convolution characteristics | |
| CN111079847A (en) | Remote sensing image automatic labeling method based on deep learning | |
| CN111914919A (en) | Open set radiation source individual identification method based on deep learning | |
| Selvaraj et al. | Digital image steganalysis: A survey on paradigm shift from machine learning to deep learning based techniques | |
| CN107491729B (en) | Handwritten digit recognition method based on cosine similarity activated convolutional neural network | |
| CN114896887B (en) | Frequency-using equipment radio frequency fingerprint identification method based on deep learning | |
| CN108062538A (en) | Face identification method and device | |
| CN111126570A (en) | SAR target classification method for pre-training complex number full convolution neural network | |
| CN112395941B (en) | Radio signal radiation source individual identification method, computer device and storage medium | |
| CN113095370A (en) | Image recognition method and device, electronic equipment and storage medium | |
| CN112200123B (en) | Hyperspectral open set classification method combining dense connection network and sample distribution | |
| CN111310743A (en) | Face recognition method and device, electronic equipment and readable storage medium | |
| CN114980122A (en) | Small sample radio frequency fingerprint intelligent identification system and method | |
| CN114758113A (en) | Confrontation sample defense training method, classification prediction method and device, and electronic equipment | |
| CN110995631B (en) | Communication signal modulation mode identification method and system based on LSTM and SVM | |
| CN116482618B (en) | Radar active interference identification method based on multi-loss characteristic self-calibration network | |
| CN113109782A (en) | Novel classification method directly applied to radar radiation source amplitude sequence |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20220826 |