CN111800811B - Unsupervised detection method, unsupervised detection device, unsupervised detection equipment and storage medium for frequency spectrum abnormality - Google Patents
Unsupervised detection method, unsupervised detection device, unsupervised detection equipment and storage medium for frequency spectrum abnormality Download PDFInfo
- Publication number
- CN111800811B CN111800811B CN202010934719.8A CN202010934719A CN111800811B CN 111800811 B CN111800811 B CN 111800811B CN 202010934719 A CN202010934719 A CN 202010934719A CN 111800811 B CN111800811 B CN 111800811B
- Authority
- CN
- China
- Prior art keywords
- frequency image
- time
- network
- original
- abnormal
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 108
- 238000001228 spectrum Methods 0.000 title claims abstract description 37
- 230000005856 abnormality Effects 0.000 title claims abstract description 18
- 230000002159 abnormal effect Effects 0.000 claims abstract description 119
- 238000012360 testing method Methods 0.000 claims description 115
- 230000006870 function Effects 0.000 claims description 56
- 238000012549 training Methods 0.000 claims description 33
- 238000000034 method Methods 0.000 claims description 30
- 238000004590 computer program Methods 0.000 claims description 11
- 238000011478 gradient descent method Methods 0.000 claims description 10
- 230000003595 spectral effect Effects 0.000 claims description 5
- 238000010586 diagram Methods 0.000 abstract description 24
- 230000000694 effects Effects 0.000 abstract description 5
- 238000012986 modification Methods 0.000 abstract description 3
- 230000004048 modification Effects 0.000 abstract description 3
- 238000010606 normalization Methods 0.000 description 14
- 238000004422 calculation algorithm Methods 0.000 description 13
- 238000013528 artificial neural network Methods 0.000 description 9
- 238000012545 processing Methods 0.000 description 8
- 238000004891 communication Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000006872 improvement Effects 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000002547 anomalous effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000001747 exhibiting effect Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 230000017105 transposition Effects 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/04—Arrangements for maintaining operational condition
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/08—Testing, supervising or monitoring using real traffic
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an unsupervised detection method for frequency spectrum abnormality, which comprises the following steps: acquiring a target signal to be detected; generating a corresponding original time-frequency image according to the time-frequency diagram of the target signal; inputting an original time-frequency image into a generation countermeasure network to obtain a reconstructed time-frequency image; and determining an abnormal value by using the original time-frequency image and the reconstructed time-frequency image, and generating an abnormal detection result of the target signal according to the abnormal value and the threshold value. Therefore, after the signals are converted into the corresponding time-frequency images, the anomaly detection can be automatically carried out on the time-frequency images through the generation of the countermeasure network; and by generating the countermeasure network for anomaly detection, prior information of any signal is not needed, so that the countermeasure network can detect some burst interference, and meanwhile, because of the universality, the countermeasure network can be applied to various scenes without large-scale modification. The invention also discloses an unsupervised detection device, equipment and a storage medium for the frequency spectrum abnormity, and the technical effects can be realized.
Description
Technical Field
The present invention relates to the field of signal processing technologies, and in particular, to a method, an apparatus, a device, and a storage medium for unsupervised detection of spectrum anomaly.
Background
With the development of electronic countermeasure technology, interference and anti-interference technology in military communication systems plays an increasingly important role. An interference detection technology is used as an important link in the field of interference resistance, and aims to detect whether an interference signal exists in a communication channel, even the information such as the energy size, the working frequency point/frequency band, the interference type and the like of the interference signal needs to be known in some special scenes, and necessary prior information is provided for subsequent interference resistance.
The traditional detection scheme comprises a series of steps such as energy detection, interference positioning, interference identification and interference classification, the completion of the steps depends on a great deal of professional knowledge, and for different communication frequency bands and signals, corresponding algorithms need to be called to adapt to specific scenes, so that a great deal of manpower, material resources and financial resources are consumed. In addition, in the conventional detection scheme, a certain amount of priori knowledge is often provided for the signals, and an optimal detection scheme can be obtained by utilizing the signal to perform modeling. However, nowadays, the spectrum environment is increasingly complex, the interference signal is complex and variable, the modeling is difficult, and the detection can not be carried out by a model-driven method. Therefore, it is a problem to be solved by those skilled in the art to provide an intelligent spectrum detection system, which can be used in different spectrum environments in a universal manner, and automatically detect an abnormal situation by monitoring a spectrum.
Disclosure of Invention
The invention aims to provide an unsupervised detection method, a device, equipment and a storage medium of frequency spectrum abnormity, and provides a high-performance generalized frequency spectrum abnormity detection method, so that a wireless communication system can automatically detect the occurrence of an abnormal condition, position the position of an abnormal signal and provide support for subsequent frequency spectrum use and management.
In order to achieve the above object, the present invention provides an unsupervised detection method for spectrum abnormality, comprising:
acquiring a target signal to be detected;
generating a corresponding original time-frequency image according to the time-frequency graph of the target signal;
inputting the original time-frequency image into a pre-established generation countermeasure network to obtain a reconstructed time-frequency image corresponding to the original time-frequency image;
and determining an abnormal value by using the original time-frequency image and the reconstructed time-frequency image, and generating an abnormal detection result of the target signal according to the abnormal value and a threshold value.
The creating method for generating the countermeasure network comprises the following steps:
creating an initial generative countermeasure network, the initial generative countermeasure network comprising: an encoder network, a generator network, and a discriminator network;
constructing a combined loss function according to the countermeasure loss function, the reconstruction loss function, the corresponding countermeasure loss function weight and the reconstruction loss function weight;
and updating the weights of the encoder network, the generator network and the discriminator network through the pre-acquired training data and the joint loss function to obtain the generation countermeasure network.
Wherein the obtaining the generation countermeasure network by updating the weights of the encoder network, the generator network, and the discriminator network through the pre-obtained training data and the joint loss function includes:
the method comprises the following steps: setting fixed default weights for the encoder network and the generator network, and inputting a first original test time-frequency image into the encoder network and the generator network to generate a first reconstructed test time-frequency image; inputting a first original test time-frequency image and a first reconstructed test time-frequency image into a discriminator network, and updating the weight of the discriminator network by using the combined loss function gradient descent method;
step two: fixing the weight of the discriminator network, inputting a second original test time-frequency image into the encoder network and the generator network to generate a second reconstruction test time-frequency image, inputting the second original test time-frequency image and the second reconstruction test time-frequency image into the discriminator network, and updating the weights of the encoder network and the generator network by using the combined loss function gradient descent method;
step three: and repeatedly executing the step one and the step two for preset iteration times to obtain the trained generation countermeasure network.
Wherein, after obtaining the generation countermeasure network, the method further comprises:
inputting each original test time-frequency image in the training data into the generation countermeasure network to obtain a reconstructed test time-frequency image corresponding to each original test time-frequency image;
determining a test abnormal value corresponding to each original test time-frequency image by using the corresponding original test time-frequency image and the reconstructed test time-frequency image;
arranging the test abnormal values in a descending order, wherein each test abnormal value has a corresponding arrangement serial number;
and (3) generating a rule by using a threshold value:
determining a threshold value eta; wherein, PFAThe preset false alarm probability, N is the number of the original test time-frequency images,
represents rounding down, A is an abnormal test value, and the threshold value eta is a sequence number
Test outliers of (2).
Inputting the original time-frequency image into a pre-established generation countermeasure network to obtain a reconstructed time-frequency image corresponding to the original time-frequency image, wherein the method comprises the following steps:
inputting the original time-frequency image into an encoder network in the generation countermeasure network to obtain encoded data;
and inputting the coded data into a generator network in the generation countermeasure network to obtain a reconstructed time-frequency image corresponding to the original time-frequency image.
Wherein, the determining an abnormal value by using the original time-frequency image and the reconstructed time-frequency image, and generating an abnormal detection result of the target signal according to the abnormal value and a threshold value comprises:
by using
Calculating the original time-frequency image x and the reconstructed time-frequency image
Wherein | represents the norm of L1;
judging whether the abnormal value is larger than the threshold value;
if so, determining that the target signal is abnormal, and determining that no abnormality exists in the abnormal detection result of the target signal; and if not, determining that the target signal is abnormal according to the abnormal detection result.
Wherein if the anomaly detection result indicates that there is an anomaly, the unsupervised detection method further includes:
and subtracting the reconstructed time-frequency image from the original time-frequency image to obtain an abnormal time-frequency image of the abnormal signal, and positioning the time-frequency domain of the abnormal signal through the abnormal time-frequency image.
In order to achieve the above object, the present invention further provides an unsupervised spectrum abnormality detection device, including:
the signal acquisition module is used for acquiring a target signal to be detected;
the original time-frequency image generating module is used for generating a corresponding original time-frequency image according to the time-frequency image of the target signal;
the reconstruction time-frequency image generation module is used for inputting the original time-frequency image into a pre-established generation countermeasure network to obtain a reconstruction time-frequency image corresponding to the original time-frequency image;
an abnormal value determining module, configured to determine an abnormal value by using the original time-frequency image and the reconstructed time-frequency image;
and the abnormity detection result generation module is used for generating an abnormity detection result of the target signal according to the abnormity value and the threshold value.
To achieve the above object, the present invention further provides an electronic device comprising:
a memory for storing a computer program;
and the processor is used for realizing the steps of the unsupervised detection method of the spectrum abnormality when executing the computer program.
To achieve the above object, the present invention further provides a computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, implements the steps of the above-mentioned unsupervised detection method of spectrum abnormality.
According to the scheme, the unsupervised detection method for the spectrum abnormality provided by the embodiment of the invention comprises the following steps: acquiring a target signal to be detected; generating a corresponding original time-frequency image according to the time-frequency graph of the target signal; inputting the original time-frequency image into a pre-established generation countermeasure network to obtain a reconstructed time-frequency image corresponding to the original time-frequency image; and determining an abnormal value by using the original time-frequency image and the reconstructed time-frequency image, and generating an abnormal detection result of the target signal according to the abnormal value and a threshold value. Therefore, after the signals are converted into the corresponding time-frequency images, the anomaly detection can be automatically carried out on the time-frequency images through the generation of the countermeasure network; and by generating the countermeasure network for anomaly detection, prior information of any signal is not needed, so that the countermeasure network can detect some burst interference, and meanwhile, because of the universality, the countermeasure network can be applied to various scenes without large-scale modification. In addition, the original time-frequency image and the reconstructed time-frequency image can be subtracted to obtain an abnormal time-frequency image of the abnormal signal, and the time-frequency domain of the abnormal signal is located through the abnormal time-frequency image. The invention also discloses an unsupervised detection device, equipment and a storage medium for the frequency spectrum abnormity, and the technical effects can be realized.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flowchart of an overall implementation of the disclosed embodiments of the present invention;
FIG. 2 is a schematic diagram of a network structure of an encoder according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a generator network structure disclosed in the embodiment of the present invention;
FIG. 4 is a schematic diagram of a network structure of a discriminator according to an embodiment of the present invention;
fig. 5 is a schematic flow chart of an unsupervised spectrum anomaly detection method according to an embodiment of the present invention;
FIG. 6a is a schematic diagram of a detection probability curve of an abnormal signal according to an embodiment of the present invention;
FIG. 6b is a schematic diagram of another abnormal signal detection probability curve according to the embodiment of the present invention;
FIG. 6c is a schematic diagram of another abnormal signal detection probability curve according to the embodiment of the present invention;
FIG. 6d is a schematic diagram of another abnormal signal detection probability curve according to the embodiment of the present invention;
fig. 7 is a schematic structural diagram of an unsupervised spectrum anomaly detection device according to an embodiment of the present invention;
fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the disclosure.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the conventional anomaly detection scheme relies on a large amount of professional knowledge, and for different communication bands and signals, parameters need to be adjusted by an algorithm to adapt to a specific scene, which consumes a large amount of manpower, material resources, and financial resources. In addition, in a traditional detection scheme, a certain amount of prior knowledge is often required for signal detection, but due to increasingly complex spectrum environment, interference signals are complex and variable and difficult to model, and cannot be detected by a model-driven method.
Therefore, the method is based on the difficulties of the two existing interference detection algorithms, the deep learning anomaly detection algorithm based on data driving is designed by using a neural network which is a generalized tool, the neural network can extract recognizable low-dimensional data features from high-dimensional data features of signals, prior information of any signal is not needed, and the neural network can detect some sudden interference. The premise of deep learning application is that enough data are used for training a network, and complex abnormal signals cannot be modeled, that is, enough abnormal signal samples and corresponding sample labels cannot be generated manually, so that common supervised classification learning is not feasible for a spectrum abnormality detection scene. Therefore, an unsupervised learning mode is adopted, the normal frequency spectrum is used as training data, and the network learns the distribution of the normal frequency spectrum. When a network test is performed, the network detects an abnormality when a spectrum including an abnormal signal is input.
The current unsupervised anomaly detection is mainly divided into prediction type anomaly detection and reconstruction type anomaly detection, and considering that in a real frequency spectrum environment, an anomaly signal can change rapidly, and the reconstruction type anomaly detection is simpler and more effective than the prediction type anomaly detection. The most common reconstruction Model is an Auto Encoder (AE), but its effect is not good, and in 2014 Ian an GoodFellow proposed generation of a confrontation network (GAN), which has a very good effect on generating pictures due to its confrontation training concept. Therefore, the time-frequency diagram of the received signal is used as a 2-dimensional image, the image is reconstructed by utilizing the generated countermeasure network GAN, the reconstruction error can be used as the basis of abnormal detection, and once the reconstruction error is larger than a set threshold, the abnormal situation can be judged. In addition, by comparing the original signal and the reconstructed signal, the position of the abnormal point in time and frequency can be effectively located on the pixel level. The result shows that the method can effectively detect and locate the abnormal signals in the frequency spectrum, and obtains greater performance improvement compared with the existing deep learning algorithm.
Referring to fig. 1, a flowchart is integrally implemented for a scheme provided by an embodiment of the present invention, and as can be seen from fig. 1, in order to detect an anomaly in a frequency spectrum through a countermeasure network, a data set needs to be prepared, a neural network needs to be built, and the neural network needs to be trained to create a network. Specifically, the steps are as follows:
first, prepare the data set
The method firstly needs to select an experimental data set, determine a training data set and a testing data set related to the experimental data set so as to train and test the generation countermeasure network, and specifically comprises the following steps:
(1.1) acquiring signals of a certain specific frequency band by using a universal spectrum acquisition device (HackRF SDR), converting N obtained signal samples into a time-frequency domain through short-time Fourier transform (STFT), and finally obtaining 64 x 64 3-channel 2D images as training data by using an image processing tool to form a training data set;
(1.2) M signal samples are acquired by using a universal spectrum acquisition device HackRF SDR and are divided into 5 parts, wherein one part is directly used as a normal test sample without subsequent processing, and 4 abnormal signals are respectively added to the original signal in the other 4 parts of samples, so that an abnormal test sample is formed. The 4 kinds of abnormal signals are respectively: 1) tonal interference (random center frequency); 2) noise interference (bandwidth and starting frequency random); 3) sweep frequency interference (random frequency range and time range); 4) broadband impulse interference (random in duration). The obtained 5120 signal samples are converted into a time-frequency domain through short-time Fourier transform (STFT), and a 64 x 64 3 channel 2D image is finally obtained by utilizing an image processing tool and serves as final training data to form a test data set.
Second, build up the neural network
Specifically, the created generative countermeasure network of the present application first needs to create an initial generative countermeasure network, which includes: the encoder network, the generator network and the discriminator network are characterized in that the specific structure of each network is as follows:
(2.1) referring to fig. 2, which is a schematic diagram of an encoder network structure provided in the embodiment of the present invention, it can be seen from the diagram that an encoder network E constructed in the present application includes a 9-layer structure of four convolutional layers, four normalization layers, and a full connection layer, and the left and right distribution of the structure is sequentially: the first buildup layer 4 × 4Conv → the second buildup layer 4 × 4Conv → the first normalization layer layerrnorm → the third buildup layer 4 × 4Conv → the second normalization layer layerrnorm → the fourth buildup layer 4 × 4Conv → the third normalization layer layerrnorm → the first fully-connected layer Linear → the fourth normalization layer batotchmld. Wherein the parameters of each layer are set as follows:
the convolution kernel size of the first convolution layer is 4 multiplied by 4, the convolution step is 2, the zero padding number is 1, and the number of convolution kernels is 64;
the convolution kernel size of the second convolution layer is 4 multiplied by 4, the convolution step is 2, the zero padding number is 1, and the number of convolution kernels is 128;
the convolution kernel size of the third convolution layer is 4 multiplied by 4, the convolution step is 2, the zero padding number is 1, and the number of convolution kernels is 256;
the convolution kernel size of the fourth convolution layer is 4 multiplied by 4, the convolution step is 2, the zero padding number is 1, and the number of convolution kernels is 512;
the number of output features of the first fully connected layer is 8192;
(2.2) referring to fig. 3, it can be seen from fig. 3 that the generator network G includes a 7-layer structure of five transposed convolution layers and four normalization layers, and the left and right distribution of the generator network G is sequentially the first transposed convolution layer 4 × 4Up-Conv → the first normalization layer BatchNorm2d → the second transposed convolution layer 4 × 4Up-Conv → the second normalization layer BatchNorm2d → the third transposed convolution layer 4 × 4Up-Conv → the third normalization layer BatchNorm2d → the fourth transposed convolution layer 4 × 4Up-Conv → the fourth normalization layer BatchNorm2d → the fifth transposed convolution layer 4 × 4 Up-Conv. Wherein the parameters of each layer are set as follows:
the convolution kernel size of the first transposition convolution layer is 4 multiplied by 4, the convolution step length is 2, the zero padding number is 1, and the number of convolution kernels is 512;
the convolution kernel size of the second transpose convolution layer is 4 multiplied by 4, the convolution step size is 2, the zero padding number is 1, and the number of convolution kernels is 256;
the convolution kernel size of the third transposed convolution layer is 4 multiplied by 4, the convolution step size is 2, the zero padding number is 1, and the number of convolution kernels is 128;
the convolution kernel size of the fourth transpose convolution layer is 4 multiplied by 4, the convolution step size is 2, the zero padding number is 1, and the number of convolution kernels is 64;
the convolution kernel size of the fifth transposed convolution layer is 4 × 4, the convolution step size is 2, the zero padding number is 1, and the number of convolution kernels is 3.
(2.3) referring to fig. 4, which is a schematic diagram of a discriminator network structure provided in the embodiment of the present invention, as can be seen from fig. 4, the discriminator network D includes an 8-layer structure of four convolutional layers, three normalization layers, and a full connection layer, and the left and right distribution of the discriminator network D sequentially includes: the first buildup layer 4 × 4Conv → the second buildup layer 4 × 4Conv → the first normalization layer layerrnorm → the third buildup layer 4 × 4Conv → the second normalization layer layerrnorm → the fourth buildup layer 4 × 4Conv → the third normalization layer layerrnorm → the first fully-connected layer Linear. Wherein the parameters of each layer are set as follows:
the convolution kernel size of the first convolution layer is 4 multiplied by 4, the convolution step is 2, the zero padding number is 1, and the number of convolution kernels is 64;
the convolution kernel size of the second convolution layer is 4 multiplied by 4, the convolution step is 2, the zero padding number is 1, and the number of convolution kernels is 128;
the convolution kernel size of the third convolution layer is 4 multiplied by 4, the convolution step is 2, the zero padding number is 1, and the number of convolution kernels is 256;
the convolution kernel size of the fourth convolution layer is 4 multiplied by 4, the convolution step is 2, the zero padding number is 1, and the number of convolution kernels is 512;
the number of output features of the first fully-connected layer is 1.
Training neural network
In this application, before training an application network, a joint loss function needs to be constructed according to a countermeasure loss function, a reconstruction loss function, and corresponding countermeasure loss function weight and reconstruction loss function weight, so as to train a neural network through the joint loss function, for example:
(3.1) separately determining the penalty function LadvReconstruction loss function LrWeight parameter lambda ofaAnd λr(ii) a Wherein, the application can set lambdaa＝1、λr＝99；
(3.2) according to the penalty-penalties-of-confrontation function LadvReconstruction loss function LrThese two loss functions and their weight parameter lambdaaAnd λrAnd obtaining a joint loss function:
L＝λaLadv+λrLr
wherein the penalty function is:
wherein x is the original time-frequency image, G (E (x)) is the time-frequency image after reconstruction,
representing expectation, D (x) representing output of discriminator, PxRepresents the distribution obeyed by x;
a reconstruction loss function of
Wherein xiAnd
the ith pixel point in the original time-frequency graph and the reconstructed time-frequency graph respectively represents the L1 norm.
Further, after the joint loss function is determined, the weights of the encoder network, the generator network and the discriminator network need to be updated through the pre-acquired training data and the joint loss function, so as to obtain a generated countermeasure network. Specifically, the method for updating the weights of the encoder network, the generator network and the discriminator network to obtain the generation countermeasure network comprises the following steps:
the method comprises the following steps: setting fixed default weights for the encoder network and the generator network, and inputting a first original test time-frequency image into the encoder network and the generator network to generate a first reconstructed test time-frequency image; inputting the first original test time-frequency image and the first reconstruction test time-frequency image into a discriminator network, and updating the weight of the discriminator network by using a joint loss function gradient reduction method;
specifically, in this step, the weights of the discriminator are mainly updated by fixing the weights of the encoder and the generator, that is to say: after the weights of the encoder and the generator are fixed, an original time-frequency graph in the training data set and a time-frequency graph reconstructed by the encoder and the generator are respectively input into the discriminator, and the weights of the discriminator are updated by using a predefined combined loss function gradient descent method, such as: setting a learning rate of the discriminator network to 0.0002; taking the difference value output by the two discriminators as the gradient value of the discriminator; the weights of the discriminator network are updated using an Adam optimizer.
Step two: fixing the weight of the discriminator network, inputting a second original test time-frequency image into the encoder network and the generator network to generate a second reconstruction test time-frequency image, inputting the second original test time-frequency image and the second reconstruction test time-frequency image into the discriminator network, and updating the weight of the encoder network and the generator network by using a combined loss function gradient descent method;
step three: and repeatedly executing the step one and the step two for preset iteration times to obtain the trained generated countermeasure network.
Specifically, in step two, the weight of the discriminator needs to be fixed, the original time-frequency diagram in the training data set and the reconstructed time-frequency diagram obtained after passing through the encoder and the generator are input into the discriminator, and the weights of the generator and the encoder are updated by using a combined loss function gradient descent method, such as: setting the learning rates of the generator and the encoder to 0.0002 respectively; taking a pixel difference value between the original time-frequency graph and the reconstructed time-frequency graph as a gradient value ^ J of the generator and the encoder2(ii) a By usingThe Adam optimizer updates the weights of the generator and the encoder network; and then, repeatedly executing the first step and the second step, wherein the repeatedly executed times are preset iterative updating times, such as: and if the iterative updating times are set to be 200 times, the trained network model is obtained after the network weight values in the first step and the second step are repeatedly updated for 200 times.
Fourthly, calculating a threshold value
It can be understood that after the trained generated countermeasure network is generated, a threshold value needs to be set to detect whether an abnormal value exists in the signal, and the process of setting the threshold value includes the following steps:
inputting each original test time-frequency image in the training data into the generation countermeasure network to obtain a reconstructed test time-frequency image corresponding to each original test time-frequency image;
determining a test abnormal value corresponding to each original test time-frequency image by using the corresponding original test time-frequency image and the reconstructed test time-frequency image;
arranging the test abnormal values in a descending order, wherein each test abnormal value has a corresponding arrangement serial number;
and (3) generating a rule by using a threshold value:
determining a threshold value eta; wherein, PFAThe preset false alarm probability, N is the number of the original test time-frequency images,
represents rounding down, A is an abnormal test value, and the threshold value eta is a sequence number
Test outliers of (2).
Specifically, the present application first needs to define an outlier as a reconstruction error
Wherein x and
respectively, an original time-frequency graph and a reconstructed time-frequency graph, and then the threshold calculation method is as follows: sample x in training data1,x2,...,xNSequentially inputting the data into the network to obtain abnormal scores, and sequencing all the abnormal scores from large to small to obtain A1,A2,...,ANThe threshold is the first of them
The individual anomaly score is a threshold value eta, i.e.
Wherein P isFAFor a set false alarm probability, N is the number of training samples,
representing a rounding down.
Fifth, anomaly detection
It should be noted that, after the above steps, the trained generated countermeasure network is obtained, and the threshold value is set, so that the generated countermeasure network and the threshold value can be tested through the test data to test whether the corresponding abnormal signal can be detected. After the test is finished, the signal can be detected abnormally through the test. It should be noted that, since the testing process is the same as the process of normally detecting the signal, the present application is specifically described herein by taking the example of detecting the target signal to be detected.
Referring to fig. 5, a schematic flow chart of an unsupervised spectrum anomaly detection method according to an embodiment of the present invention is provided; as can be seen from the figure, the unsupervised detection method specifically includes the following steps:
s101, acquiring a target signal to be detected;
specifically, the target signal in the present application is a target signal to be detected, and if in the test process, the test sample signal used by the test network can be directly obtained from the test data, which may be a normal test sample signal, or an abnormal test sample signal added with different abnormal signals.
S102, generating a corresponding original time-frequency image according to the time-frequency graph of the target signal;
in the application, after a target signal is obtained, the target signal can be converted into a time-frequency domain through short-time Fourier transform (STFT), a time-frequency domain graph is generated, a 64 x 64 3-channel 2D image is finally obtained by using an image processing tool and is called as an original time-frequency image in the application and is marked as x;
s103, inputting the original time-frequency image into a pre-established generation countermeasure network to obtain a reconstructed time-frequency image corresponding to the original time-frequency image;
specifically, after an original video image is obtained, the original time-frequency image needs to be input into an encoder network in a generation countermeasure network to obtain encoded data; then inputting the coded data into a generator network in the generation countermeasure network to obtain a reconstructed time-frequency image corresponding to the original time-frequency image, such as: sending x into the encoder network E, thereby obtaining encoded data z ═ E (x) of a lower dimension; sending z into generator network G to obtain reconstructed time-frequency image
And S104, determining an abnormal value by using the original time-frequency image and the reconstructed time-frequency image, and generating an abnormal detection result of the target signal according to the abnormal value and a threshold value.
Specifically, for each original time-frequency image x, the application needs to calculate an outlier, such as: by using
Calculating an original time-frequency image x and a corresponding reconstructed time-frequency image
Wherein | represents the norm of L1; and according to the formula:
obtaining an abnormality detection result, wherein H0H represents that the time-frequency diagram is normal1This time-frequency diagram is shown to be abnormal, that is: the application needs to judge whether the abnormal value A (x) is larger than the threshold value eta; if so, indicating that the corresponding time-frequency diagram is normal, and judging that the abnormal detection result of the target signal is not abnormal at the moment; if not, the corresponding time-frequency diagram is abnormal, and the abnormal detection result of the target signal is abnormal.
Sixthly, abnormal positioning
It should be noted that, in the present application, if the anomaly detection result indicates that there is an anomaly, the original time-frequency image and the reconstructed time-frequency image may be subtracted from each other to obtain an anomaly time-frequency map of the anomaly signal, so as to locate the time-frequency domain of the anomaly signal through the anomaly time-frequency map. Namely: abnormal time-frequency diagram
The X axis and the Y axis of the abnormal time-frequency diagram respectively represent the time and the frequency of the abnormal signal, and the specific time-frequency domain position of the abnormal signal can be obtained by determining the relation between the pixel and the time frequency.
In summary, the present application can be seen that a neural network is adopted as a generalized tool, a complex model does not need to be established, and features of normal samples are directly extracted by using data, so as to detect an abnormality. In addition, the method and the device have the advantages that the generated countermeasure network is used as a generation model, characteristics of normal time-frequency samples can be well learned, the Wasserstein GAN variant is applied, stability of GAN training is improved by changing a loss function method, and the problem of mode collapse is solved.
In this embodiment, the effect of the present invention is further explained by combining with a simulation experiment:
1. simulation experiment conditions
The simulation condition of the invention is based on a pytorech-GPU-1.5.1 deep learning framework, a countermeasure network and an encoder are built and generated by using a python3.7 scripting language, and the training of the model is completed by using a block of NVIDIA GeForce GTX 1650. All training data and test data were generated using MATLAB, simulating FM frequency bands, training iteration number 200, batch sample number 128, learning rate 0.0002.
2. Emulated content
Firstly, under the above experimental conditions, 8192 time-frequency images without abnormal signals are trained by the method, during the training period, the time-frequency images in the training set are sequentially input into the encoder and the generator, the reconstructed time-frequency images are obtained through forward propagation calculation output, the original time-frequency images and the reconstructed time-frequency images are respectively input into the discriminator, the results obtained by the discriminator and the results obtained by the generator are used for calculating the joint loss function, and the weights of the encoder, the generator and the discriminator are updated by utilizing a gradient descent method to obtain a trained network model.
And secondly, firstly, inputting all training data into the network, and calculating a threshold by using the reconstruction error of the training data. And then inputting the time-frequency graph in the test set to the trained network, calculating a reconstruction error, comparing with a threshold, and judging whether the abnormity is included.
Referring to fig. 6a, 6b, 6c, and 6d, schematic diagrams of detection probability curves of four abnormal signals provided by the embodiment of the present invention; specifically, fig. 6a, 6b, 6c, and 6d are detection probability curves of 4 kinds of abnormal signals when the interference Signal Ratio (JSR) is-20 dB to 20dB, the given false alarm probabilities are 0.01%, 0.1%, 1%, and 10%, and 6a to 6d represent tone interference, noise interference, sweep interference, and broadband pulse interference, respectively. In addition, in order to show the performance advantages of the invention, the application also compares the method with a SAIFE algorithm which is also belonging to the detection of the reconfiguration abnormity and has better performance. It can be seen from the figure that the performance of the present invention is obviously superior to the SAIFE algorithm for any abnormal signal type and any interference-to-signal ratio condition. Taking noise interference as an example: when the detection probability is 1 and the false alarm probability is 0.01%, the invention has 4dB performance improvement compared with the SAIFE algorithm, and the advantage of the detection probability is basically kept stable along with the change of the interference-signal ratio and the false alarm probability.
On the other hand, the two algorithms exhibit differences in performance for different anomaly classes. For example, the two algorithms are excellent for detecting noise interference and impulse interference, the invention can reach near-perfect detection performance when the interference-signal ratio reaches-4 dB under the false alarm probability of 10%, and under the same condition, the required interference-signal ratio of SAIFE is 0dB, which is 4dB lower than that of the invention. This is because noise interference and impulse interference occupy a large amount of energy in the frequency or time dimension, exhibiting a large difference from the original normal spectral signal, resulting in a large reconstruction error. In contrast, both algorithms have poor performance for detecting both tone and swept frequency interference, which is the worst. As can be seen from fig. 6c, the interference-to-signal ratios required by the present invention and the SAIFE algorithm to achieve 100% detection performance with a false alarm probability of 10% are respectively 10dB and 18dB, which is respectively 14dB and 18dB lower than the noise interference detection under the same detection. This is because the tone and sweep disturbances are so close in shape to the FM modulated signal in the normal spectral data that it is difficult for the reconstructed spectral anomaly detector to distinguish the anomalous signal from the normal signal.
Nevertheless, the present invention still exhibits great advantages over the existing SAIFE algorithm, under the above conditions, the present invention has 8dB performance improvement over SAIFE, and when the detection probability is reduced to 70%, the performance difference can be increased to 14 dB. In summary, the detection performance of the reconfigured spectrum abnormal detection is good when the abnormal signal is greatly different from the normal signal, and if the abnormal signal is very close to the working signal, the performance of the method is reduced to a certain extent.
And thirdly, subtracting the original time-frequency graph from the reconstructed time-frequency graph to obtain an image, namely the distribution of the abnormal signals in time and frequency, so that the positioning of the abnormal interference signals can be realized.
In the following, the unsupervised detection device provided by the embodiment of the present invention is introduced, and the unsupervised detection device described below and the unsupervised detection method described above may be referred to each other.
Referring to fig. 7, an unsupervised spectrum anomaly detection apparatus provided in an embodiment of the present invention includes:
a signal acquiring module 100, configured to acquire a target signal to be detected;
an original time-frequency image generating module 200, configured to generate a corresponding original time-frequency image according to the time-frequency graph of the target signal;
a reconstructed time-frequency image generating module 300, configured to input the original time-frequency image into a pre-created generation countermeasure network, and obtain a reconstructed time-frequency image corresponding to the original time-frequency image;
an outlier determination module 400 for determining outliers using the original time-frequency image and the reconstructed time-frequency image;
an anomaly detection result generating module 500 is configured to generate an anomaly detection result of the target signal according to the anomaly value and the threshold value.
Wherein, this device still includes: generating a confrontation network creation module, the module comprising:
an initial-generation-countermeasure-network creating unit configured to create an initial generation countermeasure network, the initial generation countermeasure network including: an encoder network, a generator network, and a discriminator network;
the combined loss function building unit is used for building a combined loss function according to the confrontation loss function, the reconstruction loss function, the corresponding confrontation loss function weight and the corresponding reconstruction loss function weight;
and the weight updating unit is used for updating the weights of the encoder network, the generator network and the discriminator network through the pre-acquired training data and the joint loss function to obtain the generation countermeasure network.
Wherein, the weight updating unit specifically includes:
the first network updating subunit is used for setting fixed default weights for the encoder network and the generator network, and inputting the first original test time-frequency image into the encoder network and the generator network to generate a first reconstructed test time-frequency image; inputting a first original test time-frequency image and a first reconstructed test time-frequency image into a discriminator network, and updating the weight of the discriminator network by using the combined loss function gradient descent method;
the second network updating subunit is used for fixing the weight of the discriminator network, inputting a second original test time-frequency image into the encoder network and the generator network to generate a second reconstruction test time-frequency image, inputting the second original test time-frequency image and the second reconstruction test time-frequency image into the discriminator network, and updating the weight of the encoder network and the generator network by using the combined loss function gradient descent method;
and the execution subunit is used for repeatedly triggering the first network updating subunit and the second network updating subunit for preset iteration times so as to obtain the trained generation countermeasure network.
The device also comprises a threshold value generation module, wherein the module comprises:
a reconstruction test time-frequency image obtaining unit, configured to input each original test time-frequency image in the training data into the generation countermeasure network, and obtain a reconstruction test time-frequency image corresponding to each original test time-frequency image;
the test abnormal value determining unit is used for determining a test abnormal value corresponding to each original test time-frequency image by using the corresponding original test time-frequency image and the reconstruction test time-frequency image;
the sorting unit is used for sorting the test abnormal values in a descending order, and each test abnormal value has a corresponding sorting serial number;
a threshold value determining unit, configured to generate a rule using a threshold value:
determining a threshold value eta; wherein, PFAThe preset false alarm probability, N is the number of the original test time-frequency images,
represents rounding down, A is an abnormal test value, and the threshold value eta is a sequence number
Test outliers of (2).
The reconstructed time-frequency image generation module 300 is specifically configured to:
inputting the original time-frequency image into an encoder network in the generation countermeasure network to obtain encoded data; and inputting the coded data into a generator network in the generation countermeasure network to obtain a reconstructed time-frequency image corresponding to the original time-frequency image.
The anomaly detection result generating module 500 is specifically configured to: by using
Calculating the original time-frequency image x and the reconstructed time-frequency image
Wherein | represents the norm of L1; judging whether the abnormal value is larger than the threshold value; if so, determining that the target signal is abnormal, and determining that no abnormality exists in the abnormal detection result of the target signal; and if not, determining that the target signal is abnormal according to the abnormal detection result.
Wherein, this device still includes:
and the abnormity positioning module is used for subtracting the original time-frequency image and the reconstructed time-frequency image to obtain an abnormity time-frequency image of the abnormal signal when the abnormity detection result is abnormal, so as to position the time-frequency domain of the abnormal signal through the abnormity time-frequency image.
Referring to fig. 8, a schematic structural diagram of an electronic device disclosed in the embodiment of the present invention includes:
a memory 11 for storing a computer program;
a processor 12, configured to implement the steps of the method for unsupervised detection of spectrum anomalies according to the above-mentioned method embodiments when the computer program is executed.
In this embodiment, the device may be a PC (Personal Computer), or may be a terminal device such as a smart phone, a tablet Computer, a palmtop Computer, or a portable Computer.
The device may include a memory 11, a processor 12, and a bus 13.
The memory 11 includes at least one type of readable storage medium, which includes a flash memory, a hard disk, a multimedia card, a card type memory (e.g., SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, and the like. The memory 11 may in some embodiments be an internal storage unit of the device, for example a hard disk of the device. The memory 11 may also be an external storage device of the device in other embodiments, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), etc. provided on the device. Further, the memory 11 may also include both an internal storage unit of the device and an external storage device. The memory 11 may be used not only to store application software installed in the device and various types of data, such as program codes for performing an unsupervised detection method, etc., but also to temporarily store data that has been output or is to be output.
The processor 12 may be, in some embodiments, a Central Processing Unit (CPU), controller, microcontroller, microprocessor or other data Processing chip for executing program codes stored in the memory 11 or Processing data, such as program codes for performing an unsupervised detection method.
The bus 13 may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown in FIG. 8, but this is not intended to represent only one bus or type of bus.
Further, the device may further include a network interface 14, and the network interface 14 may optionally include a wired interface and/or a wireless interface (e.g., WI-FI interface, bluetooth interface, etc.), which are generally used to establish a communication connection between the device and other electronic devices.
Optionally, the device may further comprise a user interface 15, the user interface 15 may comprise a Display (Display), an input unit such as a Keyboard (Keyboard), and the optional user interface 15 may further comprise a standard wired interface, a wireless interface. Alternatively, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, or the like. The display, which may also be referred to as a display screen or display unit, is suitable for displaying information processed in the device and for displaying a visualized user interface.
Fig. 8 shows only the device with the components 11-15, and it will be understood by those skilled in the art that the structure shown in fig. 8 does not constitute a limitation of the device, and may comprise fewer or more components than those shown, or some components may be combined, or a different arrangement of components.
The embodiment of the invention also discloses a computer readable storage medium, wherein a computer program is stored on the computer readable storage medium, and when the computer program is executed by a processor, the steps of the unsupervised detection method for spectrum abnormality in the embodiment of the method are realized.
Wherein the storage medium may include: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (8)
1. An unsupervised detection method of spectrum anomalies, comprising:
acquiring a target signal to be detected;
generating a corresponding original time-frequency image according to the time-frequency graph of the target signal;
inputting the original time-frequency image into a pre-established generation countermeasure network to obtain a reconstructed time-frequency image corresponding to the original time-frequency image;
determining an abnormal value by using the original time-frequency image and the reconstructed time-frequency image, and generating an abnormal detection result of the target signal according to the abnormal value and a threshold value;
the creating method for generating the countermeasure network comprises the following steps:
creating an initial generative countermeasure network, the initial generative countermeasure network comprising: an encoder network, a generator network, and a discriminator network;
constructing a combined loss function according to the countermeasure loss function, the reconstruction loss function, the corresponding countermeasure loss function weight and the reconstruction loss function weight;
updating the weights of the encoder network, the generator network and the discriminator network through pre-acquired training data and the joint loss function to obtain the generation countermeasure network;
wherein, after obtaining the generation countermeasure network, the method further comprises:
inputting each original test time-frequency image in the training data into the generation countermeasure network to obtain a reconstructed test time-frequency image corresponding to each original test time-frequency image;
determining a test abnormal value corresponding to each original test time-frequency image by using the corresponding original test time-frequency image and the reconstructed test time-frequency image;
arranging the test abnormal values in a descending order, wherein each test abnormal value has a corresponding arrangement serial number;
and (3) generating a rule by using a threshold value:
determining a threshold value
(ii) a Wherein,
in order to be a pre-set false alarm probability,
for the number of original test time-frequency images,
which represents the rounding-down of the whole,
to test outliers, the threshold values
To arrange the serial numbers
Test outliers of (2).
2. The unsupervised detection method of claim 1, wherein the obtaining the generation countermeasure network by updating weights of the encoder network, the generator network and the discriminator network through pre-obtained training data and the joint loss function comprises:
the method comprises the following steps: setting fixed default weights for the encoder network and the generator network, and inputting a first original test time-frequency image into the encoder network and the generator network to generate a first reconstructed test time-frequency image; inputting a first original test time-frequency image and a first reconstructed test time-frequency image into a discriminator network, and updating the weight of the discriminator network by using the combined loss function gradient descent method;
step two: fixing the weight of the discriminator network, inputting a second original test time-frequency image into the encoder network and the generator network to generate a second reconstruction test time-frequency image, inputting the second original test time-frequency image and the second reconstruction test time-frequency image into the discriminator network, and updating the weights of the encoder network and the generator network by using the combined loss function gradient descent method;
step three: and repeatedly executing the step one and the step two for preset iteration times to obtain the trained generation countermeasure network.
3. The unsupervised detection method of claim 1, wherein inputting the original time-frequency image into a pre-created generation countermeasure network to obtain a reconstructed time-frequency image corresponding to the original time-frequency image comprises:
inputting the original time-frequency image into an encoder network in the generation countermeasure network to obtain encoded data;
and inputting the coded data into a generator network in the generation countermeasure network to obtain a reconstructed time-frequency image corresponding to the original time-frequency image.
4. The unsupervised detection method of claim 1, wherein the determining an abnormal value using the original time-frequency image and the reconstructed time-frequency image, and generating an abnormal detection result of the target signal according to the abnormal value and a threshold value comprises:
by using
Calculating the original time-frequency image
And said reconstructed time-frequency image
Abnormal value of
Wherein
represents the norm of L1;
judging whether the abnormal value is larger than the threshold value;
if so, determining that the target signal is abnormal, and determining that no abnormality exists in the abnormal detection result of the target signal; and if not, determining that the target signal is abnormal according to the abnormal detection result.
5. The unsupervised detection method according to any one of claims 1 to 4, wherein if the abnormality detection result is that there is an abnormality, the unsupervised detection method further comprises:
and subtracting the reconstructed time-frequency image from the original time-frequency image to obtain an abnormal time-frequency image of the abnormal signal, and positioning the time-frequency domain of the abnormal signal through the abnormal time-frequency image.
6. An unsupervised detection device of spectral anomalies, comprising:
the signal acquisition module is used for acquiring a target signal to be detected;
the original time-frequency image generating module is used for generating a corresponding original time-frequency image according to the time-frequency image of the target signal;
the reconstruction time-frequency image generation module is used for inputting the original time-frequency image into a pre-established generation countermeasure network to obtain a reconstruction time-frequency image corresponding to the original time-frequency image;
an abnormal value determining module, configured to determine an abnormal value by using the original time-frequency image and the reconstructed time-frequency image;
an anomaly detection result generation module, configured to generate an anomaly detection result of the target signal according to the anomaly value and the threshold value;
wherein, the unsupervised detection device further comprises: a generative confrontation network creation module comprising:
an initial-generation-countermeasure-network creating unit configured to create an initial generation countermeasure network, the initial generation countermeasure network including: an encoder network, a generator network, and a discriminator network;
the combined loss function building unit is used for building a combined loss function according to the confrontation loss function, the reconstruction loss function, the corresponding confrontation loss function weight and the corresponding reconstruction loss function weight;
a weight updating unit, configured to update weights of the encoder network, the generator network, and the discriminator network through pre-obtained training data and the joint loss function, so as to obtain the generation countermeasure network;
wherein, the unsupervised detection device further comprises: a threshold value generation module, wherein the threshold value generation module comprises:
a reconstruction test time-frequency image obtaining unit, configured to input each original test time-frequency image in the training data into the generation countermeasure network, and obtain a reconstruction test time-frequency image corresponding to each original test time-frequency image;
the test abnormal value determining unit is used for determining a test abnormal value corresponding to each original test time-frequency image by using the corresponding original test time-frequency image and the reconstruction test time-frequency image;
the sorting unit is used for sorting the test abnormal values in a descending order, and each test abnormal value has a corresponding sorting serial number;
a threshold value determining unit, configured to generate a rule using a threshold value:
determining a threshold value
(ii) a Wherein,
in order to be a pre-set false alarm probability,
for the number of original test time-frequency images,
which represents the rounding-down of the whole,
to test outliers, the threshold values
To arrange the serial numbers
Test outliers of (2).
7. An electronic device, comprising:
a memory for storing a computer program;
a processor for implementing the steps of the method for unsupervised detection of a spectrum anomaly according to any one of claims 1 to 5 when executing said computer program.
8. A computer-readable storage medium, characterized in that a computer program is stored thereon, which computer program, when being executed by a processor, carries out the steps of the method for unsupervised detection of spectral anomalies according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010934719.8A CN111800811B (en) | 2020-09-08 | 2020-09-08 | Unsupervised detection method, unsupervised detection device, unsupervised detection equipment and storage medium for frequency spectrum abnormality |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010934719.8A CN111800811B (en) | 2020-09-08 | 2020-09-08 | Unsupervised detection method, unsupervised detection device, unsupervised detection equipment and storage medium for frequency spectrum abnormality |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111800811A CN111800811A (en) | 2020-10-20 |
| CN111800811B true CN111800811B (en) | 2020-12-08 |
Family
ID=72834292
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010934719.8A Active CN111800811B (en) | 2020-09-08 | 2020-09-08 | Unsupervised detection method, unsupervised detection device, unsupervised detection equipment and storage medium for frequency spectrum abnormality |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111800811B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112327331A (en) * | 2020-11-02 | 2021-02-05 | 中山大学 | GNSS deception jamming detection method, device, equipment and storage medium |
| CN112488294A (en) * | 2020-11-20 | 2021-03-12 | 北京邮电大学 | Data enhancement system, method and medium based on generation countermeasure network |
| CN112308038B (en) * | 2020-11-25 | 2022-09-27 | 哈尔滨工业大学 | Mechanical equipment fault signal identification method based on classroom type generation confrontation network model |
| WO2022119426A1 (en) * | 2020-12-01 | 2022-06-09 | Université Internationale de RABAT | Intelligent system for immediate detection and notification of disturbances in electrical signal quality |
| CN113052080A (en) * | 2021-03-26 | 2021-06-29 | 南京航空航天大学 | Illegal unmanned aerial vehicle detection method based on unsupervised generation countermeasure network |
| CN113572539B (en) * | 2021-06-24 | 2022-08-26 | 西安电子科技大学 | Storage-enhanced unsupervised spectrum anomaly detection method, system, device and medium |
| CN114781460B (en) * | 2022-05-13 | 2023-04-07 | 西安电子科技大学 | Interference signal detection and identification method and device for coupling with communication signal |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108226892A (en) * | 2018-03-27 | 2018-06-29 | 天津大学 | A kind of radar signal restoration methods under complicated noise based on deep learning |
| CN110297218A (en) * | 2019-07-09 | 2019-10-01 | 哈尔滨工程大学 | Based on the unknown modulation system detection method of radar signal for generating confrontation network |
| CN111370020A (en) * | 2020-02-04 | 2020-07-03 | 清华珠三角研究院 | Method, system, device and storage medium for converting voice into lip shape |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106971270A (en) * | 2017-03-30 | 2017-07-21 | 中国人民解放军装备学院 | A kind of electronics use the appraisal procedure of applicability |
| US11115815B2 (en) * | 2018-05-14 | 2021-09-07 | Ppip, Llc | Radio frequency (RF) emitter detector |
| CN110544239B (en) * | 2019-08-19 | 2021-12-17 | 中山大学 | Multi-modal MRI conversion method, system and medium for generating countermeasure network based on conditions |
-
2020
- 2020-09-08 CN CN202010934719.8A patent/CN111800811B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108226892A (en) * | 2018-03-27 | 2018-06-29 | 天津大学 | A kind of radar signal restoration methods under complicated noise based on deep learning |
| CN110297218A (en) * | 2019-07-09 | 2019-10-01 | 哈尔滨工程大学 | Based on the unknown modulation system detection method of radar signal for generating confrontation network |
| CN111370020A (en) * | 2020-02-04 | 2020-07-03 | 清华珠三角研究院 | Method, system, device and storage medium for converting voice into lip shape |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111800811A (en) | 2020-10-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111800811B (en) | Unsupervised detection method, unsupervised detection device, unsupervised detection equipment and storage medium for frequency spectrum abnormality | |
| CN112508044A (en) | Artificial intelligence AI model evaluation method, system and equipment | |
| Wu et al. | Blind image quality assessment based on rank-order regularized regression | |
| EP4024297A1 (en) | Artificial intelligence (ai) model evaluation method and system, and device | |
| CN110458084B (en) | Face age estimation method based on inverted residual error network | |
| CN108509407A (en) | Text semantic similarity calculating method, device and user terminal | |
| CN111260620B (en) | Image anomaly detection method and device and electronic equipment | |
| CN115964670B (en) | Spectrum anomaly detection method | |
| CN110096617A (en) | Video classification methods, device, electronic equipment and computer readable storage medium | |
| CN109583367A (en) | Image text row detection method and device, storage medium and electronic equipment | |
| He et al. | A visual residual perception optimized network for blind image quality assessment | |
| CN111461445A (en) | Short-term wind speed prediction method and device, computer equipment and storage medium | |
| CN111260655A (en) | Image generation method and device based on deep neural network model | |
| CN113205820A (en) | Method for generating voice coder for voice event detection | |
| CN117596011A (en) | Power grid flow anomaly detection method and system based on countermeasure convolutional neural network | |
| CN113408804B (en) | Electricity stealing behavior detection method, system, terminal equipment and storage medium | |
| CN107729947A (en) | A kind of Face datection model training method, device and medium | |
| CN117693754A (en) | Training masked automatic encoders for image restoration | |
| CN114399025A (en) | Graph neural network interpretation method, system, terminal and storage medium | |
| CN113822144A (en) | Target detection method and device, computer equipment and storage medium | |
| CN113393854B (en) | Voice processing method and device, electronic equipment and computer storage medium | |
| CN116467466A (en) | Knowledge graph-based code recommendation method, device, equipment and medium | |
| CN110490824A (en) | The method and device of image denoising | |
| CN115859048A (en) | Noise processing method and device for partial discharge signal | |
| CN115511262A (en) | Transformer quality detection method and device |
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 |