CN113705446B - Open set identification method for individual radiation source - Google Patents
Open set identification method for individual radiation source Download PDFInfo
- Publication number
- CN113705446B CN113705446B CN202110997031.9A CN202110997031A CN113705446B CN 113705446 B CN113705446 B CN 113705446B CN 202110997031 A CN202110997031 A CN 202110997031A CN 113705446 B CN113705446 B CN 113705446B
- Authority
- CN
- China
- Prior art keywords
- layer
- radiation source
- training
- sample
- convolution
- 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
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
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computing Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Signal Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of signal processing, and particularly relates to an open set identification method for radiation source individuals. In practical application scenarios, the radiation source individual identification system faces an open electromagnetic environment, and receiving radiation source signals of unknown types which are not recorded by a database cannot be avoided, so that open-set identification of radiation sources has important research significance. On the basis of using a deep neural network as a radiation source signal feature extractor, firstly, a combined loss function which gives consideration to classification and clustering effects is designed, the signal depth features extracted by the neural network are ensured to have good classification characteristics and clustering characteristics, then, an extreme value distribution model is constructed by using feature vectors obtained by training data, a discrimination threshold value is determined, a discrimination algorithm is realized, and high-accuracy open-set identification of a radiation source signal is completed.
Description
Technical Field
The invention relates to the technical field of signal processing, in particular to an open set identification method facing to a radiation source individual.
Background
The individual identification of the radiation source is to extract the inherent characteristics of an external emitter by utilizing a received radiation source signal so as to realize the identification of the communication radiation source equipment. With the rapid development of communication technology and increasingly complex communication environment, the individual identification of the radiation source has wide application prospect and important research value in the fields of military communication, communication safety and the like. The conventional radiation source individual identification technology needs to design expert features in advance for classification identification, but the expert features seriously depend on prior information such as received signal types, and the application scene is greatly limited. In recent years, deep neural networks have been used for radiation source individual recognition and have been used for obtaining excellent performance as deep learning has made many breakthroughs in the fields of computer vision and the like. But most approaches focus on closed-set identification, i.e. assuming that the radiation source signal samples that need to be classified must belong to some known class in the radiation source database. However, in practical application scenarios, the individual radiation source identification system is faced with an open electromagnetic environment, and the reception of radiation source signals of unknown types which are not recorded by the database is inevitable. The open set identification aiming at the individual radiation source can not only correctly classify the known radiation source, but also reject the unknown individual radiation source, thereby being more in line with the actual application requirement of the individual radiation source identification and having more research significance.
For the open set identification problem, a common method is to train a classification neural network, such as a convolutional neural network, by using a training data set to obtain a closed set classifier, and then set a threshold value for the class probability of a test sample obtained by the classifier, if the maximum probability is smaller than the threshold value, the test sample is determined as an unknown class, otherwise, the test sample is classified as a known class corresponding to the maximum probability. However, since the convolutional neural network is normalized to obtain the class probability, an incorrect high prediction probability may be output to cause the identification system to crash. Aiming at the defects of the traditional neural network, bendale et al propose an Openmax structure, and reconstruct the probability that a test sample belongs to an unknown class from a closed set probability vector output by the convolutional neural network by using Weibull distribution. However, the Openmax is used for realizing the open set identification, and the known class samples need to have a good clustering effect in the feature space, that is, the distance between classes is large, and the distance within the classes is small. For radiation source signals with slight fingerprint feature differences, the classification neural network is used as a feature extractor, so that the features with good clustering effect cannot be obtained, and the accuracy and the performance are poor when Openmax is used for individual collection identification of radiation sources.
Disclosure of Invention
In view of the analysis, in order to solve the problems of low accuracy and poor performance of the open set identification of the radiation source individual, the invention provides a radiation source individual open set identification scheme based on a deep neural network and an extremum theory. The core technology of the invention comprises two parts: designing a joint loss function considering both classification performance and clustering performance aiming at a radiation source signal, applying the joint loss function to a ResNet11 convolutional neural network, training the network by using a radiation source training set, and taking the obtained neural network as a feature extractor; and secondly, a discrimination algorithm is provided based on the training sample and the extreme value theory, and discrimination and classification are carried out on the test sample.
In order to realize the method, the technical scheme of the invention is realized as follows:
an open set identification method facing to individual radiation source comprises the following steps:
step 1: acquiring a radiation source signal sample set and a category corresponding to each radiation source signal sample to obtain a training set;
and 2, step: training a ResNet11 convolutional neural network model by using a training set based on a designed joint loss function, and taking the trained model as a feature extractor; the loss function is marked as join _ loss, and the specific expression is as follows:
joint_loss＝CE+α*Tripletloss
CE represents a cross entropy loss function for ensuring the classification performance of features extracted by a neural network, and the specific expression is as follows:
where K represents the number of classes to be classified in the training set, t K Correct label, y, representing training sample K Representing the output result of the training sample after passing through the neural network;
tripletloss represents a clustering loss function for ensuring the clustering performance of features extracted by a neural network, and the specific expression is as follows:
Tripletloss＝max(d(a,p)-d(a,n)+margin,0)
wherein a represents a standard sample in the network training process, p represents a sample of the same type as a standard sample point, n represents a sample of different type from the standard sample point, d represents the calculation of the distance between the two points, and margin is a hyper-parameter for controlling the clustering effect achieved by the network training;
alpha is a scale factor used for controlling the proportion of the clustering target and the classification target in the optimization process, when the alpha is less than 1, the network pays more attention to the classification loss, and when the alpha is more than 1, the network pays more attention to the clustering loss;
and step 3: obtaining a training sample feature vector through a feature extractor, and recording as v (x) = (v) 1 (x),...v k (x) K represents the number of classes of the training set, and the set of the feature vectors of the jth class of training samples is recorded as T j (j＝1,...k)；
And 4, step 4: calculating the mean center point of the feature vectors of the known classes according to the training sample labels, and recording the mean center point as beta j ＝mean(T j ) Calculating the distance from the feature vector of each training sample to the center point of the corresponding class to obtain a set of distances from the feature vectors of the known classes to the center point, and recording the set as D j ；
And 5: to D j Arranging the middle distance values from large to small and taking out the front gamma values, and marking as S j ；
Step 6: for sequence S j Carrying out Weibull distribution fitting to obtain a probability distribution function of the maximum distance from the j-th type known sample characteristic vector to the mean value center, and recording the probability distribution function as PDF j ；
And 7: setting a probability threshold eta, and calculating by using a probability distribution inverse function to obtain a jth distance threshold tau j ＝PDF j -1 (η)；
And step 8: for identifying the target, obtaining a feature vector through a feature extractor, calculating the distance from the feature vector to various mean value centers, taking the minimum value as chi, and recording the corresponding category as j if the chi is less than or equal to tau j And classifying the test samples into j categories, otherwise classifying the test samples into unknown categories, and completing the open set identification of the radiation source signals.
Further, the ResNet11 neural network in step 2 includes an input layer, a residual layer, and an output layer; the input layer inputs radiation source signal samples, each sample is boot IQ data with the sampling point number of 10000, and the size of the input layer is 2x10000; the residual error layer is formed by connecting three identical residual error blocks in series, each residual error block is composed of a convolution module and a direct connection module, the direct connection module is a convolution layer with a convolution kernel size of 1x1, the input size is not changed, the purpose is to enable original information to be transmitted in a depth network and prevent overfitting, the convolution module is formed by connecting three convolution units in series and is used for extracting features from original data, each convolution unit comprises a convolution layer, a regular layer and an activation layer, the convolution kernel sizes of the convolution layers of the first convolution unit and the third convolution unit are 1x1, the size of the second convolution layer is 2x3, and Relu is used as an activation function of the activation layer; the output layer comprises an average pooling layer, a flat layer and a full-connection layer, the average pooling layer pools a plurality of feature maps from the residual error layer, the dimensionality is reduced, the flat layer fuses the plurality of feature maps, the full-connection layer completes the mapping from the high-dimensional flat layer to the feature space, and the output dimensionality is consistent with the classified category number.
The invention has the following beneficial effects: 1. the classification loss function and the clustering loss function are combined, and a scale factor is introduced to construct a combined loss function to be applied to the deep neural network, so that the obtained radiation source signal feature vector has good classification and clustering performances; 2. and constructing a known Weibull distribution model by using an extreme value theory, fully considering the distribution characteristic of the characteristic vector, obtaining a threshold value by using the probability distribution characteristic, realizing a discrimination algorithm, and realizing open set identification aiming at high accuracy of individual signals of the radiation source.
Drawings
FIG. 1 is a flow chart of the present invention
FIG. 2 is a general structure diagram of a deep neural network ResNet11 used in the present invention
FIG. 3 is a diagram of residual block in ResNet11
FIG. 4 is a confusion matrix obtained from four-classification radiation source open set identification experiments
Detailed Description
The technical scheme of the invention is described in detail below with reference to the accompanying drawings and embodiments:
taking four communication radio stations as radiation sources, recording as radiation sources 1-4, wherein a training set comprises signal samples of the radiation sources 1-3, and the number of each type of sample is 500; the test set comprises signal samples of the radiation sources 1-4, wherein the radiation source 4 is a sample of unknown class, the number of samples of each class being 100. The open set identification is carried out according to the scheme provided by the invention, the flow chart of the scheme is shown in figure 1, and the specific steps are as follows:
step 1: dividing a radiation source signal sample set into a training set and a test set, wherein the test set contains a radiation source 4 signal which is not contained in the training set so as to accord with an open set recognition scene hypothesis;
and 2, step: and training a ResNet11 convolutional neural network model based on a designed joint loss function by using a training set, and taking the trained model as a feature extractor.
The joint loss function is marked as join _ loss, and the specific expression is as follows:
joint_loss＝CE+α*Tripletloss
CE represents a cross entropy loss function for ensuring the classification performance of features extracted by a neural network, and the specific expression is as follows:
where K denotes the number of classes to be classified in the training set, K =3,t in this example K Correct label, y, representing training sample K And representing the output result of the training sample after passing through the neural network.
Tripletloss represents a clustering loss function for ensuring the clustering performance of features extracted by a neural network, and the specific expression is as follows:
Tripletloss＝max(d(a,p)-d(a,n)+margin,0)
wherein a represents a standard sample in the network training process, p represents a sample of the same type as a standard sample point, n represents a sample of different type from the standard sample point, d represents the calculation of the distance between two points, margin is a super parameter for controlling the clustering effect achieved by the network training, and margin =1 in this example.
Alpha is a proportion factor used for controlling the proportion of clustering targets and classification targets in the optimization process, when alpha is smaller than 1, the network pays more attention to the classification loss, when alpha is larger than 1, the network pays more attention to the clustering loss, and as the fingerprint characteristics of radiation source signals are finer, the network pays more attention to the clustering loss in order to realize open set identification, alpha =2 in the example.
The ResNet11 neural network used for training comprises an input layer, a residual error layer and an output layer, and the overall structure is shown in FIG. 2; the input layer inputs radiation source signal samples, each sample is boot IQ data with the sampling point number of 10000, and the size of the input layer is 2x10000; the residual error layer is formed by connecting three identical residual error blocks in series, each residual error block is composed of a convolution module and a direct connection module, the specific structure of the residual error layer is shown in fig. 3, the direct connection module is a convolution layer with a convolution kernel size of 1x1, the input size is not changed, the purpose is to enable original information to be transmitted in a depth network, overfitting is prevented, the convolution module is formed by connecting three convolution units in series and is used for extracting characteristics from original data, each convolution unit comprises a convolution layer, a regularization layer and an activation layer, the convolution kernel sizes of the convolution layers of the first convolution unit and the third convolution unit are 1x1, the size of the second convolution layer is 2x3, and the activation layer uses Relu as an activation function; the output layer comprises an average pooling layer, a flat layer and a full-connection layer, the average pooling layer pools a plurality of characteristic graphs from the residual error layer, the dimensionality is reduced, the flat layer fuses the characteristic graphs, the full-connection layer completes the mapping of the high-dimensional flat layer output to the characteristic space, the output dimensionality is consistent with the classified class number, and the value in the example is 3.
And step 3: obtaining a training sample feature vector through a feature extractor, and recording as v (x) = (v) 1 (x),v 2 (x),v 3 (x) Let the set of feature vectors of the jth class of training samples be denoted as T) j (j＝1,2,3)；
And 4, step 4: calculating the mean center point of the feature vectors of the known classes according to the training sample labels, and recording as beta j ＝mean(T j ) Calculating the distance from the feature vector of each training sample to the center point of the corresponding class to obtain a set of distances from the feature vectors of the known classes to the center point, and recording the set as D j ；
And 5: to D j The middle distance values are arranged from large to small, and the front gamma =20 values are taken out and recorded as S j
Step 6: for the sequence S j Carrying out Weibull distribution fitting to obtain a probability distribution function of the maximum distance from the j-th type known sample characteristic vector to the mean value center, and recording the probability distribution function as PDF j
And 7: setting a probability threshold eta =0.8, and calculating by using a probability distribution inverse function to obtain a distance threshold of the jth class
And 8: for a test sample, obtaining a feature vector through a feature extractor, calculating the distance from the feature vector to various mean value centers, taking the minimum value as χ, and recording the minimum value as χ, wherein the corresponding category is j, and if χ is less than or equal to tau j Then the test sample is divided into j types, otherwise, the test sample is divided into unknown types, the open set identification of the radiation source signal is completed, finally, the open set identification confusion matrix is shown in figure 4, and the accuracy is shown in table 1:
TABLE 1 radiation Source four-class open set identification accuracy
| Radiation source target | Rate of identification accuracy |
| Radiation source 1 | 0.94 |
| Radiation source 2 | 0.99 |
| Radiation source 3 | 0.95 |
| Radiation source 4 (unknown class) | 0.98 |
| Integrated recognition accuracy | 0.965 |
Claims (2)
1. An open set identification method facing an individual with a radiation source is characterized by comprising the following steps:
step 1: acquiring a radiation source signal sample set and categories corresponding to the radiation source signal samples to obtain a training set;
step 2: training a ResNet11 convolutional neural network model by using a training set based on a designed joint loss function, and taking the trained model as a feature extractor; the loss function is marked as join _ loss, and the specific expression is as follows:
joint_loss＝CE+α*Tripletloss
CE represents a cross entropy loss function for ensuring the classification performance of features extracted by a neural network, and the specific expression is as follows:
where K represents the number of classes to be classified in the training set, t K Correct label representing training sample, y K Representing the output result of the training sample after passing through the neural network;
tripletloss represents a clustering loss function for ensuring the clustering performance of features extracted by a neural network, and the specific expression is as follows:
Tripletloss＝max(d(a,p)-d(a,n)+margin,0)
wherein a represents a standard sample in the network training process, p represents a sample of the same type as a standard sample point, n represents a sample of different type from the standard sample point, d represents the calculation of the distance between the two points, and margin is a hyper-parameter for controlling the clustering effect achieved by the network training;
alpha is a scale factor used for controlling the proportion of the clustering target and the classification target in the optimization process, when the alpha is less than 1, the network pays more attention to the classification loss, and when the alpha is more than 1, the network pays more attention to the clustering loss;
and step 3: obtaining a training sample feature vector through a feature extractor, and marking as v (x) = (v) 1 (x),...v k (x) K represents the number of classes of the training set, and the set of the feature vectors of the jth class of training samples is recorded as T j (j＝1,...k)；
And 4, step 4: calculating the mean center point of the feature vectors of the known classes according to the training sample labels, and recording as beta j ＝mean(T j ) Calculating the distance from the feature vector of each training sample to the center point of the corresponding class to obtain a set of distances from the feature vectors of the known classes to the center point, and recording the set as D j ；
And 5: to D j Arranging the middle distance values from large to small and taking out the front gamma values, and marking as S j ；
Step 6: for sequence S j Weibull distribution fittingObtaining the probability distribution function of the maximum distance from the j-th known sample feature vector to the mean center, and recording the probability distribution function as PDF j ；
And 7: setting a probability threshold eta, and calculating by utilizing a probability distribution inverse function to obtain a jth distance threshold
And step 8: for identifying the target, obtaining a feature vector through a feature extractor, calculating the distance from the feature vector to various mean value centers, taking the minimum value as chi, and recording the corresponding category as j if the chi is less than or equal to tau j And classifying the test samples into j categories, otherwise classifying the test samples into unknown categories, and completing the open set identification of the radiation source signals.
2. The individual radiation source-oriented open set identification method according to claim 1, wherein the ResNet11 convolutional neural network comprises an input layer, a residual layer and an output layer; the input layer inputs radiation source signal samples, each sample is boot IQ data with the sampling point number of 10000, and the size of the input layer is 2x10000; the residual error layer is formed by connecting three identical residual error blocks in series, each residual error block is composed of a convolution module and a direct connection module, the direct connection module is a convolution layer with a convolution kernel size of 1x1, the input size is not changed, the purpose is to enable original information to be transmitted in a depth network, overfitting is prevented, the convolution module is formed by connecting three convolution units in series and used for extracting features from original data, each convolution unit comprises a convolution layer, a regularization layer and an activation layer, the convolution kernel sizes of the convolution layers of the first convolution unit and the third convolution unit are 1x1, the size of the second convolution layer is 2x3, and Relu is used as an activation function of the activation layer; the output layer comprises an average pooling layer, a flat layer and a full-connection layer, the average pooling layer pools a plurality of feature maps from the residual error layer, the dimensionality is reduced, the flat layer fuses the plurality of feature maps, the full-connection layer completes the mapping from the high-dimensional flat layer to the feature space, and the output dimensionality is consistent with the classified category number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110997031.9A CN113705446B (en) | 2021-08-27 | 2021-08-27 | Open set identification method for individual radiation source |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110997031.9A CN113705446B (en) | 2021-08-27 | 2021-08-27 | Open set identification method for individual radiation source |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113705446A CN113705446A (en) | 2021-11-26 |
| CN113705446B true CN113705446B (en) | 2023-04-07 |
Family
ID=78656143
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110997031.9A Active CN113705446B (en) | 2021-08-27 | 2021-08-27 | Open set identification method for individual radiation source |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113705446B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114567528B (en) * | 2022-01-26 | 2023-05-23 | 中国人民解放军战略支援部队信息工程大学 | Communication signal modulation mode open set recognition method and system based on deep learning |
| CN114626418A (en) * | 2022-03-18 | 2022-06-14 | 中国人民解放军32802部队 | Radiation source identification method and device based on multi-center complex residual error network |
| CN114940142B (en) * | 2022-05-31 | 2023-10-13 | 中国人民解放军国防科技大学 | Automobile anti-theft method and system based on radiation source individual verification and automobile |
| CN114781554B (en) * | 2022-06-21 | 2022-09-20 | 华东交通大学 | Open set identification method and system based on small sample condition |
| CN114866162B (en) * | 2022-07-11 | 2023-09-26 | 中国人民解放军国防科技大学 | Signal data enhancement method and system and communication radiation source identification method and system |
| CN116522242B (en) * | 2023-04-28 | 2024-01-26 | 哈尔滨工程大学 | Radiation source signal open set identification method based on diffusion model |
| CN117150265B (en) * | 2023-08-17 | 2024-05-17 | 中国人民解放军陆军工程大学 | Robust radio frequency signal open set identification method under low signal-to-noise ratio condition |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108846412A (en) * | 2018-05-08 | 2018-11-20 | 复旦大学 | A kind of method of extensive zero sample learning |
| CN109376786A (en) * | 2018-10-31 | 2019-02-22 | 中国科学院深圳先进技术研究院 | A kind of image classification method, device, terminal device and readable storage medium storing program for executing |
| CN109871805A (en) * | 2019-02-20 | 2019-06-11 | 中国电子科技集团公司第三十六研究所 | A kind of electromagnetic signal opener recognition methods |
| CN109934269A (en) * | 2019-02-25 | 2019-06-25 | 中国电子科技集团公司第三十六研究所 | A kind of opener recognition methods of electromagnetic signal and device |
| CN110807467A (en) * | 2019-09-18 | 2020-02-18 | 北京大学 | Support point learning-based mining method and device for category of development |
| CN111914919A (en) * | 2020-07-24 | 2020-11-10 | 天津大学 | Open set radiation source individual identification method based on deep learning |
| CN112308008A (en) * | 2020-11-12 | 2021-02-02 | 电子科技大学 | Radar radiation source individual identification method based on working mode open set of transfer learning |
| CN112422213A (en) * | 2020-11-11 | 2021-02-26 | 杭州电子科技大学 | Efficient spectrum sensing method based on support vector machine |
| CN112749633A (en) * | 2020-12-25 | 2021-05-04 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Separate and reconstructed individual radiation source identification method |
| CN112904299A (en) * | 2021-03-03 | 2021-06-04 | 西安电子科技大学 | Radar high-resolution range profile open set target identification method based on deep intra-class division |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11106969B2 (en) * | 2017-01-19 | 2021-08-31 | International Business Machines Corporation | Method and apparatus for driver identification leveraging telematics data |
-
2021
- 2021-08-27 CN CN202110997031.9A patent/CN113705446B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108846412A (en) * | 2018-05-08 | 2018-11-20 | 复旦大学 | A kind of method of extensive zero sample learning |
| CN109376786A (en) * | 2018-10-31 | 2019-02-22 | 中国科学院深圳先进技术研究院 | A kind of image classification method, device, terminal device and readable storage medium storing program for executing |
| CN109871805A (en) * | 2019-02-20 | 2019-06-11 | 中国电子科技集团公司第三十六研究所 | A kind of electromagnetic signal opener recognition methods |
| CN109934269A (en) * | 2019-02-25 | 2019-06-25 | 中国电子科技集团公司第三十六研究所 | A kind of opener recognition methods of electromagnetic signal and device |
| CN110807467A (en) * | 2019-09-18 | 2020-02-18 | 北京大学 | Support point learning-based mining method and device for category of development |
| CN111914919A (en) * | 2020-07-24 | 2020-11-10 | 天津大学 | Open set radiation source individual identification method based on deep learning |
| CN112422213A (en) * | 2020-11-11 | 2021-02-26 | 杭州电子科技大学 | Efficient spectrum sensing method based on support vector machine |
| CN112308008A (en) * | 2020-11-12 | 2021-02-02 | 电子科技大学 | Radar radiation source individual identification method based on working mode open set of transfer learning |
| CN112749633A (en) * | 2020-12-25 | 2021-05-04 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Separate and reconstructed individual radiation source identification method |
| CN112904299A (en) * | 2021-03-03 | 2021-06-04 | 西安电子科技大学 | Radar high-resolution range profile open set target identification method based on deep intra-class division |
Non-Patent Citations (5)
| Title |
|---|
| Guillem Reus-Muns等.Trust in 5G open RANs through machine learning: RF fingerprinting on the POWDER PAWR platform.《GLOBECOM 2020 - 2020 IEEE Global Communications Conference》.2021,1-6. * |
| Souvik Hazra等.Short-range radar-based gesture recognition system using 3D CNN with triplet loss.《IEEE Access》.2019,第7卷125623 - 125633. * |
| 丁纲松.通信辐射源识别关键技术研究.《中国优秀硕士学位论文全文数据库 信息科技辑》.2021,I136-1202. * |
| 汤鹏等.非合作辐射源识别技术发展动态分析.《数据采集与处理》.2020,第35卷1033-1048. * |
| 贾永强等.基于特征降维的卫星通信辐射源个体识别方法.《测控技术》.2017,第36卷14-17+22. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113705446A (en) | 2021-11-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113705446B (en) | Open set identification method for individual radiation source | |
| CN113378632B (en) | Pseudo-label optimization-based unsupervised domain adaptive pedestrian re-identification method | |
| CN108564129B (en) | Trajectory data classification method based on generation countermeasure network | |
| CN112464713B (en) | Communication radiation source radio frequency fingerprint identification method based on deep learning | |
| CN105138998B (en) | Pedestrian based on the adaptive sub-space learning algorithm in visual angle recognition methods and system again | |
| CN110175615B (en) | Model training method, domain-adaptive visual position identification method and device | |
| CN102324038A (en) | A kind of floristics recognition methods based on digital picture | |
| CN104834941A (en) | Offline handwriting recognition method of sparse autoencoder based on computer input | |
| CN111008575B (en) | Robust face recognition method based on multi-scale context information fusion | |
| Dev et al. | Multi-level semantic labeling of sky/cloud images | |
| CN104978569B (en) | A kind of increment face identification method based on rarefaction representation | |
| CN113139512B (en) | Depth network hyperspectral image classification method based on residual error and attention | |
| CN112800249A (en) | Fine-grained cross-media retrieval method based on generation of countermeasure network | |
| CN112766378A (en) | Cross-domain small sample image classification model method focusing on fine-grained identification | |
| CN110781828A (en) | Fatigue state detection method based on micro-expression | |
| CN114463812A (en) | Low-resolution face recognition method based on dual-channel multi-branch fusion feature distillation | |
| ul Haq et al. | Neural network based adaboosting approach for hyperspectral data classification | |
| Ali et al. | Sindhi handwritten-digits recognition using machine learning techniques | |
| CN110991247A (en) | Electronic component identification method based on deep learning and NCA fusion | |
| CN103942572A (en) | Method and device for extracting facial expression features based on bidirectional compressed data space dimension reduction | |
| CN112489689B (en) | Cross-database voice emotion recognition method and device based on multi-scale difference countermeasure | |
| CN113947101A (en) | Unsupervised pedestrian re-identification method and system based on softening similarity learning | |
| CN111832391A (en) | Image dimension reduction method and image identification method based on truncated nuclear norm low-rank discriminant embedding method | |
| CN116520154A (en) | Performance evaluation method and system for lithium battery pole piece detection | |
| CN115329821A (en) | Ship noise identification method based on pairing coding network and comparison learning |
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 |