CN109871805B - Electromagnetic signal open set identification method - Google Patents
Electromagnetic signal open set identification method Download PDFInfo
- Publication number
- CN109871805B CN109871805B CN201910126528.6A CN201910126528A CN109871805B CN 109871805 B CN109871805 B CN 109871805B CN 201910126528 A CN201910126528 A CN 201910126528A CN 109871805 B CN109871805 B CN 109871805B
- Authority
- CN
- China
- Prior art keywords
- electromagnetic signal
- neural network
- sample
- convolutional neural
- layer
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to an electromagnetic signal open set identification method, belongs to the technical field of signal processing, and solves the problems of low discrimination accuracy and poor performance of the existing open set identification method. The method comprises the following steps: acquiring an electromagnetic signal sample set and the category of each electromagnetic signal sample; dividing the electromagnetic signal sample set into a training set and a test set; training the convolutional neural network by using a training set, and evaluating the trained convolutional neural network by using a test set to obtain an optimal convolutional neural network; constructing a characteristic parameter weibull distribution model according to the optimal convolutional neural network; and performing open set identification on the unknown electromagnetic signals by using the optimal convolutional neural network, and judging the accuracy of the identification result according to the constructed characteristic parameter weibull distribution model. The method effectively improves the accuracy of electromagnetic signal open set identification.
Description
Technical Field
The invention relates to the technical field of signal processing, in particular to an electromagnetic signal open set identification method.
Background
With the rapid development of science and technology, electromagnetic signal identification has wide application and important research value in the aspects of national defense safety, intelligent traffic, security industry and the like. Electromagnetic signal recognition in the general sense means that the recognition result of an input electromagnetic signal is given by calculating the similarity of the input electromagnetic signal sample and samples in a known signal library. Thus, there are two types of identification problems: 1) closed Set (Close Set) electromagnetic signal identification without rejection, i.e. assuming that the input electromagnetic signal sample must belong to a certain individual in the signal library; 2) and (3) Open Set (Open Set) identification with rejection, namely, firstly, judging whether an input electromagnetic signal sample belongs to a known signal library, and then giving an identification result on the basis of determination. In a real-world scenario, the electromagnetic identification system faces more open electromagnetic environments, not only known signals, but also unknown signals. Compared with closed set signal identification without rejection, the open set signal identification can effectively distinguish unknown signals from known signals, so that the method is more suitable for practical application of an identification system and has more research significance.
For closed set electromagnetic signal identification, a plurality of identification algorithms with good performance, such as Fisher linear discrimination and Garbor characteristic discrimination classification methods, can achieve high correct identification rate, but the performance of the algorithms used for open set identification is not satisfactory, so that the open set electromagnetic signal identification problem is concerned more. The nearest neighbor algorithm calculates the minimum distance between the test sample and each sample of the known class, and determines the class to which the test sample belongs to be received or rejected by comparing the minimum distance with a preset threshold, but the identification effect is seriously influenced by factors such as noise, disturbance and the like. In order to eliminate the changes, researchers propose a normalization method, and open set discrimination is carried out by using the minimum normalization distance so as to improve discrimination accuracy. Although the open set identification methods reduce the false acceptance rate under the condition of the same false rejection rate, the open set identification methods only use the information of one dimension of the minimum distance or the maximum confidence coefficient to carry out open set judgment, and abandon the distribution of the distance or the confidence coefficient between the test sample and each known sample in the space containing a large amount of judgment information, so the open set judgment has low accuracy and unsatisfactory performance.
Disclosure of Invention
In view of the above analysis, the present invention aims to provide an electromagnetic signal open set identification method, so as to solve the problems of low discrimination accuracy and poor performance of the existing open set identification method.
The purpose of the invention is mainly realized by the following technical scheme:
an electromagnetic signal open set identification method comprises the following steps:
acquiring an electromagnetic signal sample set and the category of each electromagnetic signal sample;
dividing the electromagnetic signal sample set into a training set and a test set; training the convolutional neural network by using a training set, and evaluating the trained convolutional neural network by using a test set to obtain an optimal convolutional neural network;
constructing a characteristic parameter weibull distribution model according to the optimal convolutional neural network;
and performing open set identification on the unknown electromagnetic signals by using the optimal convolutional neural network, and judging the accuracy of the identification result according to the constructed characteristic parameter weibull distribution model.
The invention has the following beneficial effects: the strong characteristic characterization capability of the convolutional neural network is utilized, an extreme value theory is introduced, correctly detected sample parameters are fully utilized, a characteristic parameter weibull distribution model is constructed, the accuracy of an unknown electromagnetic signal identification result is judged according to the constructed characteristic parameter weibull distribution model, and a better unknown sample rejection result can be obtained.
On the basis of the scheme, the invention is further improved as follows:
further, training the convolutional neural network using the training set includes:
and taking each electromagnetic signal sample in the training set as the input of a convolutional neural network, taking the class of the corresponding electromagnetic signal sample as the output of the convolutional neural network, and training the convolutional neural network.
Further, the convolutional neural network sequentially comprises K basic layers and a full connection layer; k is not less than 1 and is an integer; wherein the content of the first and second substances,
each basic layer comprises a convolution layer, a ReLu layer and a pooling layer, wherein the convolution layer performs convolution processing on signals and extracts features; the ReLu layer provides coefficients for nonlinear transformation; the pooling layer compresses the input feature map, extracts main features and reduces the network computation complexity;
the full connection layer comprises an activation layer and a classification layer, wherein the activation layer is used for connecting all the characteristics output by the Kth basic layer and classifying all the characteristics output by the classification layer.
Further, the classification layer utilizes a softmax classifier to implement classification functions.
The beneficial effect of adopting the further scheme is that: the softmax classifier can realize classification of various categories and has the advantages of simple method, easiness in realization, high accuracy and the like.
Further, the method for constructing the characteristic parameter weibull distribution model according to the optimal convolutional neural network further comprises the following steps:
extracting activation layer coefficient v in optimal convolutional neural network1(x)、...、vc(x) Wherein c is the number of categories of test results;
recording the coefficient of the active layer of the electromagnetic signal sample correctly classified in the test process as Si,j＝vj(xi,j) Wherein i is a serial number corresponding to the ith electromagnetic signal sample, j is a sample type corresponding to the ith electromagnetic signal sample, and j is 1, 2.. and c;
calculating the mean value center of each type of sample according to the correctly classified electromagnetic signal sample activation layer coefficients in the test process, and recording the mean value as uj＝meani(Si,j)；
Calculating the distance between the correctly classified electromagnetic signal sample activation layer coefficients in each class of samples and the corresponding class mean value center, and sequencing the distance from large to small to obtain a corresponding activation layer coefficient sequencing sequence Sj(l) And (3) carrying out weibull distribution fitting on the first items in the sequencing sequence to obtain a parameter distribution model, and recording as:
ρj＝(τj,κj,λj)＝weibullfit(||Sj(l)-μj||,η) (1)
wherein, tauj,κj,λjRespectively is a displacement parameter, a shape parameter and a scale parameter of the parameter weibull distribution, and eta is a selection sequencing sequence Sj(l) And (4) performing weibull distribution fitting on the corresponding parameters by utilizing weibull fit.
The beneficial effect of adopting the further scheme is that: the constructed convolutional neural network can only identify the electromagnetic signals to be identified as one of the known c-type signals, and by introducing a weibull distribution model, the statistical characteristics of the weibull distribution are fully utilized, the internal relation between the coefficients of the active layer and the fitting of the weibull distribution is analyzed, the identification of the electromagnetic signals of unknown types can be realized, and the accuracy of the diversity identification of the electromagnetic signals is improved.
Further, the method for performing open-set identification on the unknown electromagnetic signals by using the optimal convolutional neural network and judging the accuracy of the identification result according to the constructed characteristic parameter weibull distribution model further comprises the following steps:
processing unknown electromagnetic signals to obtain corresponding electromagnetic signal samples, inputting the electromagnetic signal samples into an optimal convolutional neural network, and obtaining the current active layer coefficient v1(x)、...、vc(x) Taking out the maximum coefficient vm(x) Wherein m is more than or equal to 1 and less than or equal to c, and m is a positive integer; and obtaining coefficient v by the output result of the classification layerm(x) A corresponding category label P; wherein the class label P is the mth label in the classification result of the classification layer;
v is calculated according to a characteristic parameter weibull distribution modelm(x) Cumulative probability distribution value w corresponding to Euclidean distance between P-type sample mean value centerss：
Wherein u ispIs the mean center of the class P samples, τp,κp,λpRespectively representing the weibull distribution displacement parameter, the shape parameter and the scale parameter of the P-type sample;
setting a cumulative probability distribution threshold when wsWhen the electromagnetic signal is larger than or equal to the threshold value, the unknown electromagnetic signal is judged to be unknownA signal; when w issAnd if the electromagnetic signal is smaller than the threshold value, the unknown electromagnetic signal is judged to be a known signal, and the class of the known signal is P.
The beneficial effect of adopting the further scheme is that: the method comprises the following steps of (1) realizing primary identification of an unknown electromagnetic signal through a convolutional neural network, wherein the identification result is not necessarily accurate, and the convolutional neural network can only identify the unknown electromagnetic signal as one of known classes; in order to eliminate the uncertainty, the accuracy of the identification result is judged according to the constructed characteristic parameter weibull distribution model, and if the current cumulative probability distribution value is larger than the cumulative probability distribution threshold, the unknown electromagnetic signal is judged as an unknown signal; the asphyxia is a known signal, and the signal type is the recognition result of the convolutional neural network. The accuracy of electromagnetic signal open set identification can be effectively improved in the aspect.
Further, the cumulative probability distribution threshold is set by:
extracting an active layer coefficient V in an optimal convolutional neural network1(x)、...、Vc(x) Wherein c is the number of categories of test results;
calculating the cumulative probability distribution value w corresponding to all electromagnetic signal samples in the test set in the optimal convolutional neural networks' (k), k is 1,2, …, N-M, and the maximum value of the cumulative probability distribution that 95% of the samples of the electromagnetic signal satisfy is set as the cumulative probability distribution threshold.
The beneficial effect of adopting the further scheme is that: the cumulative probability distribution corresponding to the classified test data is fully utilized to correspondingly set a cumulative probability distribution threshold, so that the electromagnetic signals can be better identified; meanwhile, in order to avoid the influence caused by individual result deviation, the maximum value of the cumulative probability distribution satisfied by 95% of the electromagnetic signal samples is set as the cumulative probability distribution threshold.
Further, by selecting characteristic parameters capable of reflecting the attributes of the electromagnetic signals, the electromagnetic signals are correspondingly processed, and the electromagnetic signal sample set is obtained.
Furthermore, the signal amplitude characteristic is selected as a characteristic parameter, each electromagnetic signal is subjected to A/D sampling and digital orthogonal transformation,obtain corresponding I path signal xI(n) and Q-path signal xQ(n), and calculating to obtain an electromagnetic signal amplitude characteristic sequence sample A (n) according to the following formula:
and obtaining electromagnetic signal amplitude characteristic sequence samples corresponding to the electromagnetic signals to form an electromagnetic signal sample set.
The beneficial effect of adopting the further scheme is that: the signal amplitude characteristics have the advantages of being convenient to obtain, capable of reflecting the signal characteristics more and the like, the signal amplitude characteristics can be selected as characteristic parameters, electromagnetic signal amplitude characteristic sequence samples corresponding to the electromagnetic signals are obtained, and an electromagnetic signal sample set is formed.
Furthermore, the category to which each electromagnetic signal sample in the electromagnetic signal sample set belongs needs to be set in advance.
In the invention, the technical schemes can be combined with each other to realize more preferable combination schemes. Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout.
FIG. 1 is a flowchart of an electromagnetic signal open set identification method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a convolutional neural network structure according to an embodiment of the present invention;
fig. 3 is a schematic diagram of an electromagnetic signal open set identification method according to an embodiment of the present invention.
Detailed Description
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate preferred embodiments of the invention and together with the description, serve to explain the principles of the invention and not to limit the scope of the invention.
The embodiment of the invention discloses an electromagnetic signal open set identification method, a flow chart is shown in fig. 1, and the method specifically comprises the following steps:
step S1: acquiring an electromagnetic signal sample set and the category of each electromagnetic signal sample;
and correspondingly processing the electromagnetic signals by selecting characteristic parameters capable of reflecting the attributes of the electromagnetic signals to obtain the electromagnetic signal sample set.
Preferably, because the signal amplitude characteristics have the advantages of being convenient to obtain, being capable of reflecting the signal characteristics more and the like, the signal amplitude characteristics are preferably used as characteristic parameters, and the corresponding I-path signal x is obtained by performing A/D sampling and digital orthogonal transformation on each electromagnetic signalI(n) and Q-path signal xQ(n), and calculating to obtain an electromagnetic signal amplitude characteristic sequence sample A (n) according to the following formula:
obtaining electromagnetic signal amplitude characteristic sequence samples corresponding to the electromagnetic signals to form an electromagnetic signal sample set;
preferably, the category to which each electromagnetic signal sample in the electromagnetic signal sample set belongs needs to be set in advance.
Preferably, the electromagnetic signal sample set includes N electromagnetic signal samples in total, and all of the N electromagnetic signal samples belong to one of the c-type signal samples;
marking the categories of N electromagnetic signal samples in the electromagnetic signal sample set, and sequentially marking the electromagnetic signal samples as
Step S2: dividing the electromagnetic signal sample set into a training set and a test set; training the convolutional neural network by using a training set, and evaluating the trained convolutional neural network by using a test set to obtain an optimal convolutional neural network;
step S21: dividing the set of electromagnetic signal samples into a training set and a test set, and further performing the following operations:
dividing an electromagnetic signal sample set into a training set and a testing set according to a certain proportion; generally, the number of training set samples is slightly greater than the number of test set samples: such as 6 to 4.
Wherein the training set contains M electromagnetic signal samples (M)<N), each electromagnetic signal sample is labeled in sequence as:
in order to improve the accuracy of the later network training, the M samples need to cover all the classes.
The test set contains N-M electromagnetic signal samples, and each electromagnetic signal sample is sequentially marked as:
in order to improve the accuracy of the later network test optimization, the N-M samples need to cover all the categories.
Step S22: training the convolutional neural network with a training set, further performing the following operations:
taking each electromagnetic signal sample in the training set as the input of a convolutional neural network, taking the class of the corresponding electromagnetic signal sample as the output of the convolutional neural network, and training the convolutional neural network;
preferably, the schematic structural diagram of the convolutional neural network in the present application is shown in fig. 2: the convolutional neural network sequentially comprises K basic layers and a full connection layer; k is not less than 1 and is an integer; each base layer comprises a convolution layer, a ReLu (Rectifledinearer unit) layer and a pooling layer, wherein the convolution layer performs convolution processing and feature extraction on signals; the ReLu layer provides coefficients for nonlinear transformation; the pooling layer compresses the input feature map, extracts main features and reduces the network computation complexity. The full connection layer is further divided into an activation layer and a classification layer,
the activation layer is used for connecting all the characteristics output by the Kth basic layer and classifying all the output characteristics through the classification layer; preferably, the classification layer in the present application implements a classification function using a softmax classifier.
Step S23: evaluating the trained convolutional neural network by using a test set to obtain an optimal convolutional neural network;
taking each electromagnetic signal sample in the test set as the input of the trained convolutional neural network, taking the class of the corresponding electromagnetic signal sample as the output of the trained convolutional neural network, evaluating the performance of the trained convolutional neural network, and adjusting the learning rate according to the test accuracy (generally, the accuracy is up to more than 95%); and when the accuracy reaches a set threshold, stopping iterative updating in algorithm training to obtain the optimal convolutional neural network.
In the actual process of identifying the open set of electromagnetic signals, the electromagnetic signals of unknown types can be included in addition to the electromagnetic signals of known types, and the electromagnetic signals of unknown types cannot be identified by adopting the convolutional neural network, so that the accuracy of identifying the open set of electromagnetic signals is reduced. The problem that electromagnetic signals of unknown classes cannot be identified is solved by introducing the weibull distribution model.
Step S3: constructing a characteristic parameter weibull distribution model according to the optimal convolutional neural network:
step S31: extracting activation layer coefficient v in optimal convolutional neural network1(x)、...、vc(x) Wherein c is the number of categories of test results;
step S32: recording the coefficient of the active layer of the electromagnetic signal sample correctly classified in the test process as Si,j＝vj(xi,j) Wherein i is a serial number corresponding to the ith electromagnetic signal sample, j is a sample type corresponding to the ith electromagnetic signal sample, and j is 1, 2.. and c;
step S33: calculating the mean value center of each type of sample according to the correctly classified electromagnetic signal sample activation layer coefficients in the test process, and recording the mean value as uj＝meani(Si,j)；
Step S34: calculating the distance between the correctly classified electromagnetic signal sample activation layer coefficients in each class of samples and the corresponding class mean value center, and sequencing the distance from large to small to obtain a corresponding activation layer coefficient sequencing sequence Sj(l) And (3) carrying out weibull distribution fitting on the first items in the sequencing sequence to obtain a parameter distribution model, and recording as:
ρj＝(τj,κj,λj)＝weibullfit(||Sj(l)-μj||,η) (1)
wherein, tauj,κj,λjRespectively is a displacement parameter, a shape parameter and a scale parameter of the parameter weibull distribution, and eta is a selection sequencing sequence Sj(l) And (4) performing weibull distribution fitting on the corresponding parameters by utilizing weibull fit.
Step S4: and performing open set identification on the unknown electromagnetic signals by using the optimal convolutional neural network, and judging the accuracy of the identification result according to the constructed characteristic parameter weibull distribution model.
Step S41: processing unknown electromagnetic signals to obtain corresponding electromagnetic signal samples, inputting the electromagnetic signal samples into an optimal convolutional neural network, and obtaining the current active layer coefficient v1(x)、...、vc(x) Taking out the maximum coefficient vm(x) Wherein m is more than or equal to 1 and less than or equal to c, and m is a positive integer; and obtaining coefficient v by the output result of the classification layerm(x) A corresponding category label P; wherein the class label P is the mth label in the classification result of the softmax classifier;
step S42: v is calculated according to a characteristic parameter weibull distribution modelm(x) Cumulative probability distribution value w corresponding to Euclidean distance between P-type sample mean value centerss：
Wherein u ispIs the mean center of the class P samples, τp,κp,λpRespectively representing the weibull distribution displacement parameter, the shape parameter and the scale parameter of the P-type sample;
step S43: setting a cumulative probability distribution threshold when wsWhen the electromagnetic signal is larger than or equal to the threshold value, the unknown electromagnetic signal is judged to be an unknown signal; when w issAnd if the electromagnetic signal is smaller than the threshold value, the unknown electromagnetic signal is judged to be a known signal, and the class of the known signal is P.
Preferably, the cumulative probability distribution threshold is set by:
extracting an active layer coefficient V in an optimal convolutional neural network1(x)、...、Vc(x) Wherein c is the number of categories of test results;
calculating the cumulative probability distribution value w corresponding to all electromagnetic signal samples in the test set in the optimal convolutional neural networks' (k), k is 1,2, …, N-M, and the maximum value of the cumulative probability distribution that 95% of the samples of the electromagnetic signal satisfy is set as the cumulative probability distribution threshold.
Fig. 3 is a schematic diagram of an electromagnetic signal open set identification method in an embodiment of the present invention, which corresponds to the open set identification process.
According to the electromagnetic signal open-set identification method provided by the invention, the strong characteristic characterization capability of a convolutional neural network is utilized, an extreme value theory is introduced, the correctly detected sample parameters are fully utilized, extreme value weibull distribution of the characteristic parameters and the class center distance is constructed, the cumulative probability distribution value is calculated for the characteristics of the activation layer of the test sample according to the corresponding weibull distribution condition, whether the signal is known or not is judged through a threshold value, and a better unknown sample rejection result can be obtained.
Those skilled in the art will appreciate that all or part of the flow of the method implementing the above embodiments may be implemented by hardware associated with computer program instructions, and the program may be stored in a computer readable storage medium. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.
Claims (9)
1. An electromagnetic signal open set identification method is characterized by comprising the following steps:
acquiring an electromagnetic signal sample set and the category of each electromagnetic signal sample;
dividing the electromagnetic signal sample set into a training set and a test set; training the convolutional neural network by using a training set, and evaluating the trained convolutional neural network by using a test set to obtain an optimal convolutional neural network;
according to the optimal convolutional neural network, constructing a characteristic parameter weibull distribution model, comprising the following steps:
extracting activation layer coefficient v in optimal convolutional neural network1(x)、...、vc(x) Wherein c is the number of categories of the test result, and x is the input electromagnetic signal sample;
recording the coefficient of the active layer of the electromagnetic signal sample correctly classified in the test process as Si,j＝vj(xi,j) Wherein i is a serial number corresponding to the ith electromagnetic signal sample, j is a sample type corresponding to the ith electromagnetic signal sample, and j is 1, 2.. and c;
calculating the mean value center of each type of sample according to the correctly classified electromagnetic signal sample activation layer coefficients in the test process, and recording the mean value as uj＝meani(Si,j)；
Calculating the distance between the correctly classified electromagnetic signal sample activation layer coefficients in each type of sample and the corresponding type mean value center, sequencing the samples from large to small to obtain a corresponding sequencing sequence, and then carrying out weibull distribution fitting on a plurality of items in the sequencing sequence to obtain a parameter distribution model which is recorded as:
ρj＝(τj,κj,λj)＝weibullfit(||Sj-uj||,η) (1)
wherein, tauj,κj,λjRespectively is a displacement parameter, a shape parameter and a scale parameter of the parameter weibull distribution, and eta is a selection sequencing sequence Sj-ujNumber of items of (S)jActivating for j-type electromagnetic signal sampleLayer coefficient, ujTaking the mean value center of a jth sample, wherein j is a sample type corresponding to the electromagnetic signal sample, and carrying out weibull distribution fitting on corresponding parameters by utilizing weibull fit;
and performing open set identification on the unknown electromagnetic signals by using the optimal convolutional neural network, and judging the accuracy of the identification result according to the constructed characteristic parameter weibull distribution model.
2. The recognition method of claim 1, wherein the training the convolutional neural network with the training set comprises:
and taking each electromagnetic signal sample in the training set as the input of a convolutional neural network, taking the class of the corresponding electromagnetic signal sample as the output of the convolutional neural network, and training the convolutional neural network.
3. The identification method according to claim 2, wherein the convolutional neural network comprises K base layers and a full connection layer in sequence, K is greater than or equal to 1 and K is an integer; wherein the content of the first and second substances,
each basic layer comprises a convolution layer, a ReLu layer and a pooling layer, and the convolution layer performs convolution processing and feature extraction on the sample signal; the ReLu layer provides coefficients for nonlinear transformation; the pooling layer compresses the input feature map and extracts main features;
the full connection layer comprises an activation layer and a classification layer, wherein the activation layer is used for connecting all the characteristics output by the Kth basic layer and classifying all the characteristics output by the classification layer.
4. The identification method of claim 3, wherein the classification layer implements a classification function using a softmax classifier.
5. The identification method according to any one of claims 1 to 4, wherein the open-set identification of the unknown electromagnetic signals by using the optimal convolutional neural network and the accuracy of the identification result is judged according to the constructed characteristic parameter weibull distribution model, further comprising the following steps:
processing unknown electromagnetic signals to obtain corresponding electromagnetic signal samples, inputting the electromagnetic signal samples into an optimal convolutional neural network, and obtaining the current active layer coefficient v1(x)、...、vc(x) X is the input electromagnetic signal sample, and the maximum coefficient v is takenm(x) Wherein m is more than or equal to 1 and less than or equal to c, m is a positive integer, and c is the number of test result categories; and obtaining coefficient v by the output result of the classification layerm(x) A corresponding category label P; wherein the class label P is the mth label in the classification result of the classification layer;
v is calculated according to a characteristic parameter weibull distribution modelm(x) Cumulative probability distribution value w corresponding to Euclidean distance between P-type sample mean value centerss：
Wherein u ispIs the mean center of the class P samples, τp,κp,λpRespectively representing the weibull distribution displacement parameter, the shape parameter and the scale parameter of the P-type sample;
setting a cumulative probability distribution threshold when wsWhen the electromagnetic signal is larger than or equal to the threshold value, the unknown electromagnetic signal is judged to be an unknown signal; when w issAnd if the electromagnetic signal is smaller than the threshold value, the unknown electromagnetic signal is judged to be a known signal, and the class of the known signal is P.
6. The identification method of claim 5, wherein the cumulative probability distribution threshold is set by:
extracting activation layer coefficient v in optimal convolutional neural network1(x)、...、vc(x) Wherein c is the number of categories of the test result, and x is the input electromagnetic signal sample;
calculating the cumulative probability distribution value w corresponding to all electromagnetic signal samples in the test set in the optimal convolutional neural networks' (k), k is 1,2, …, N-M, N is the number of electromagnetic signal samples contained in the electromagnetic signal sample set, M is the number of electromagnetic signal samples in the training setAnd setting the maximum value of the cumulative probability distribution satisfied by 95% of the electromagnetic signal samples as the cumulative probability distribution threshold value, wherein M is less than N.
7. The identification method according to claim 1, wherein the electromagnetic signal sample set is obtained by selecting characteristic parameters capable of reflecting the attributes of the electromagnetic signal and performing corresponding processing on the electromagnetic signal.
8. The identification method according to claim 7, wherein the signal amplitude characteristic is selected as the characteristic parameter, and the I-path signal x of the corresponding input electromagnetic signal sample x is obtained by A/D sampling and digital orthogonal transformation of each electromagnetic signalI(n) and Q-path signal xQ(n), and calculating to obtain an electromagnetic signal amplitude characteristic sequence sample A (n) according to the following formula:
and obtaining electromagnetic signal amplitude characteristic sequence samples corresponding to the electromagnetic signals to form an electromagnetic signal sample set.
9. The method according to any one of claims 6 to 8, wherein the class to which each electromagnetic signal sample in the set of electromagnetic signal samples belongs is set in advance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910126528.6A CN109871805B (en) | 2019-02-20 | 2019-02-20 | Electromagnetic signal open set identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910126528.6A CN109871805B (en) | 2019-02-20 | 2019-02-20 | Electromagnetic signal open set identification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109871805A CN109871805A (en) | 2019-06-11 |
| CN109871805B true CN109871805B (en) | 2020-10-27 |
Family
ID=66918956
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910126528.6A Active CN109871805B (en) | 2019-02-20 | 2019-02-20 | Electromagnetic signal open set identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109871805B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112163636B (en) * | 2020-10-15 | 2023-09-26 | 电子科技大学 | Unknown mode identification method of electromagnetic signal radiation source based on twin neural network |
| CN113705446B (en) * | 2021-08-27 | 2023-04-07 | 电子科技大学 | Open set identification method for individual radiation source |
| CN116578913B (en) * | 2023-03-31 | 2024-05-24 | 中国人民解放军陆军工程大学 | Reliable unmanned aerial vehicle detection and recognition method oriented to complex electromagnetic environment |
| CN116628619B (en) * | 2023-07-26 | 2023-10-20 | 西南交通大学 | Unknown abnormal electrical phenomenon identification method based on vehicle network coupling |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102520697A (en) * | 2011-12-16 | 2012-06-27 | 西安建筑科技大学 | Onsite information preprocessing method of remote cooperative diagnosis |
| CN105654037A (en) * | 2015-12-21 | 2016-06-08 | 浙江大学 | Myoelectric signal gesture recognition method based on depth learning and feature images |
| CN106980367A (en) * | 2017-02-27 | 2017-07-25 | 浙江工业大学 | A kind of gesture identification method based on myoelectricity topographic map |
| CN108520199A (en) * | 2018-03-04 | 2018-09-11 | 天津大学 | Based on radar image and the human action opener recognition methods for generating confrontation model |
| CN108846412A (en) * | 2018-05-08 | 2018-11-20 | 复旦大学 | A kind of method of extensive zero sample learning |
| CN109086700A (en) * | 2018-07-20 | 2018-12-25 | 杭州电子科技大学 | Radar range profile's target identification method based on depth convolutional neural networks |
| CN109342909A (en) * | 2018-12-14 | 2019-02-15 | 中国测试技术研究院电子研究所 | A kind of cable accessory Partial Discharge Pattern Recognition Method based on SLLE |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140114609A1 (en) * | 2012-10-23 | 2014-04-24 | Hewlett-Packard Development Company, L.P. | Adaptive analysis of signals |
| CN104316277B (en) * | 2014-09-26 | 2016-08-17 | 西安交通大学 | Air-tightness monitoring method based on sound detection Yu Blind Signal Separation |
| CN107908299B (en) * | 2017-11-16 | 2020-12-11 | 青岛罗博数码科技有限公司 | Electromagnetic signal processing method, device and system |
-
2019
- 2019-02-20 CN CN201910126528.6A patent/CN109871805B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102520697A (en) * | 2011-12-16 | 2012-06-27 | 西安建筑科技大学 | Onsite information preprocessing method of remote cooperative diagnosis |
| CN105654037A (en) * | 2015-12-21 | 2016-06-08 | 浙江大学 | Myoelectric signal gesture recognition method based on depth learning and feature images |
| CN106980367A (en) * | 2017-02-27 | 2017-07-25 | 浙江工业大学 | A kind of gesture identification method based on myoelectricity topographic map |
| CN108520199A (en) * | 2018-03-04 | 2018-09-11 | 天津大学 | Based on radar image and the human action opener recognition methods for generating confrontation model |
| CN108846412A (en) * | 2018-05-08 | 2018-11-20 | 复旦大学 | A kind of method of extensive zero sample learning |
| CN109086700A (en) * | 2018-07-20 | 2018-12-25 | 杭州电子科技大学 | Radar range profile's target identification method based on depth convolutional neural networks |
| CN109342909A (en) * | 2018-12-14 | 2019-02-15 | 中国测试技术研究院电子研究所 | A kind of cable accessory Partial Discharge Pattern Recognition Method based on SLLE |
Non-Patent Citations (4)
| Title |
|---|
| A Robust Voice Activity Detector Based on Weibull and Gaussian Mixture Distribution;Yuan Liang 等;《2010 2nd International Conference on Signal Processing Systems (ICSPS)》;20101231;第V2-26-V2-30页 * |
| Probability Models for Open Set Recognition;Walter J.Scheirer 等;《IEEE Transactions on Pattern Analysis and Machine Intelligence》;20141130;第36卷(第11期);摘要，第1-6节 * |
| 基于机器学习的无线电信号识别;尹良 等;《2017年全国无线及移动通信学术大会论文集》;20171231;摘要，第1-5节 * |
| 通信辐射源个体开集识别中的二分类SVDD算法;骆振兴 等;《通信对抗》;20170630;第36卷(第2期);第1-6页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109871805A (en) | 2019-06-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109871805B (en) | Electromagnetic signal open set identification method | |
| CN113378632B (en) | Pseudo-label optimization-based unsupervised domain adaptive pedestrian re-identification method | |
| CN109934269B (en) | Open set identification method and device for electromagnetic signals | |
| CN111833172A (en) | Consumption credit fraud detection method and system based on isolated forest | |
| CN111444951B (en) | Sample recognition model generation method, device, computer equipment and storage medium | |
| CN110942091B (en) | Semi-supervised few-sample image classification method for searching reliable abnormal data center | |
| CN109993236A (en) | Few sample language of the Manchus matching process based on one-shot Siamese convolutional neural networks | |
| CN110120218A (en) | Expressway oversize vehicle recognition methods based on GMM-HMM | |
| CN113052126B (en) | Dual-threshold open-set signal modulation identification method based on deep learning | |
| CN108171976B (en) | Vehicle use identification method based on in-transit data | |
| CN112560596B (en) | Radar interference category identification method and system | |
| CN114844840B (en) | Method for detecting distributed external network flow data based on calculated likelihood ratio | |
| CN116347318A (en) | Intelligent production test method and system for sound equipment | |
| Zhu et al. | Adversarial anomaly detection for marked spatio-temporal streaming data | |
| CN114980122A (en) | Small sample radio frequency fingerprint intelligent identification system and method | |
| CN108694375B (en) | Imaging white spirit identification method applicable to multi-electronic nose platform | |
| CN117131436A (en) | Radiation source individual identification method oriented to open environment | |
| CN116304941A (en) | Ocean data quality control method and device based on multi-model combination | |
| CN116170187A (en) | Industrial Internet intrusion monitoring method based on CNN and LSTM fusion network | |
| CN112350745B (en) | Sorting method of frequency hopping communication radio station | |
| CN111667000B (en) | Earthquake early warning method of adaptive field deep neural network | |
| CN113822048B (en) | Social media text denoising method based on space-time burst characteristics | |
| CN112633399B (en) | Sparse collaborative joint representation pattern recognition method | |
| CN114220016B (en) | Unmanned aerial vehicle aerial image domain adaptive identification method oriented to open scene | |
| CN114998747B (en) | Long flight path real-time identification method based on deep learning template matching |
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 |