CN114254673A - Denoising countermeasure self-encoder-based spectrum anomaly detection method - Google Patents

Denoising countermeasure self-encoder-based spectrum anomaly detection method Download PDF

Info

Publication number
CN114254673A
CN114254673A CN202111544290.2A CN202111544290A CN114254673A CN 114254673 A CN114254673 A CN 114254673A CN 202111544290 A CN202111544290 A CN 202111544290A CN 114254673 A CN114254673 A CN 114254673A
Authority
CN
China
Prior art keywords
signal data
module
abnormal
encoder
spectrum
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111544290.2A
Other languages
Chinese (zh)
Inventor
张卫锋
张希宁
张明根
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiaxing Shenzhi Technology Co ltd
Original Assignee
Jiaxing Shenzhi Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Jiaxing Shenzhi Technology Co ltd filed Critical Jiaxing Shenzhi Technology Co ltd
Priority to CN202111544290.2A priority Critical patent/CN114254673A/en
Publication of CN114254673A publication Critical patent/CN114254673A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Signal Processing (AREA)
  • Maintenance And Management Of Digital Transmission (AREA)

Abstract

The invention discloses a frequency spectrum anomaly detection method based on a denoising countermeasure self-encoder, and relates to the technical field of anomaly detection. The invention provides a frequency spectrum anomaly detection method based on a denoising countermeasure self-encoder, which enhances the robustness of a frequency spectrum anomaly detection model to noise by arranging a denoising layer in an encoder module, so that the frequency spectrum anomaly detection model obtained by training has an accurate detection effect under a low signal-to-noise ratio, the encoder module maps the power spectral density of an input signal to a hidden space characteristic and can classify the frequency spectrum anomaly type distribution, a network of the encoder module adopts a semi-supervised learning mode for training, and the frequency spectrum anomaly detection model can have the frequency spectrum anomaly type prediction capability under the training of a small amount of labeled samples; the invention also adopts counterstudy, can control the distribution of the coded hidden space feature vector and the abnormal type distribution vector compared with the existing self-encoder, has more stable and accurate acquisition of the signal feature and has good spectrum abnormality detection capability.

Description

Denoising countermeasure self-encoder-based spectrum anomaly detection method
Technical Field
The invention relates to the technical field of anomaly detection, in particular to a frequency spectrum anomaly detection method based on a denoising countermeasure self-encoder.
Background
The frequency spectrum abnormity detection is a process of judging whether the current frequency spectrum is abnormally changed according to the previous normal operation state of the target frequency spectrum, and is used as an important method for monitoring the state of a radio system, and the signal frequency spectrum abnormity detection has important significance for judging whether the radio system normally operates.
The traditional spectrum anomaly detection method is a classification algorithm based on a decision tree, the method is simple in design and easy to implement, however, the extraction and selection of spectrum features need expert intervention, the quality of threshold setting in the decision tree has great influence on the final detection effect, and the noise resistance performance is poor. With the development of deep learning technology in recent years, researchers provide a spectrum anomaly detection model based on a deep neural network, and the model has the advantages of high detection accuracy, adaptive feature selection and the like, but has poor anomaly detection effect under the condition of low signal-to-noise ratio. In addition, the methods all adopt a supervised learning method, a large number of normal and abnormal frequency spectrum samples need to be collected and labeled, and the abnormal type of the frequency spectrum cannot be judged, so that the effect of the methods in practical application is limited.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a spectrum anomaly detection method based on a denoising countermeasure autoencoder, which combines the denoising autoencoder and the countermeasure learning technology to enable a spectrum anomaly detection model to have more robust spectrum feature learning capability. The technical scheme provided by the invention is as follows:
according to an aspect of the embodiments of the present invention, there is provided a method for detecting spectrum abnormality of an anti-self-encoder based on denoising, the method comprising:
calculating the signal power spectral density of each IQ signal data, and dividing each IQ signal data into a training sample set D according to the abnormal state of the signal power spectral density corresponding to each IQ signal datatAnd validating the sample set DvWherein the training sample setDtComprising a non-anomalous training sample set D consisting of unlabeled IQ signal datat1And an abnormal training sample set D composed of IQ signal data including an abnormal type tagt2The verification sample set DvAre all IQ signal data including an abnormal type tag;
using the training sample set DtTraining each IQ signal data to obtain a spectrum anomaly detection model based on a denoising countermeasure self-encoder, and verifying a sample set D according to the spectrum anomaly detection modelvDetermining a reconstruction error threshold, an abnormal type judgment loss threshold and a characteristic judgment loss threshold of the spectrum abnormal detection model by each IQ signal data;
inputting IQ signal data to be detected into the spectrum anomaly detection model, calculating a reconstruction error, an anomaly type judgment loss and a characteristic judgment loss corresponding to the IQ signal data to be detected, comparing the reconstruction error, the anomaly type judgment loss and the characteristic judgment loss with the reconstruction error threshold, the anomaly type judgment loss threshold and the characteristic judgment loss threshold respectively, and determining the anomaly state of the IQ signal data to be detected according to the comparison result.
Preferably, the non-abnormal training sample set
Figure BDA0003415341260000021
The abnormal training sample set
Figure BDA0003415341260000022
The verification sample set
Figure BDA0003415341260000023
Wherein n1 is the non-abnormal training sample set Dt1The number of IQ signal data, n2 being the abnormal training sample set Dt2Number of intermediate IQ signal data, xiIs the signal power spectral density, yiIs an exception type tag.
Preferably, the spectrum abnormal model is composed of an encoder module E, a decoder module D and a characteristic discrimination module QFAnd an abnormality type discriminating module QCThe encoder module E is used for corresponding IQ signal dataThe decoder module D is used for reconstructing the signal power spectral density corresponding to the IQ signal data according to the hidden space eigenvector and the abnormal type distribution vector corresponding to the IQ signal data, and the characteristic discrimination module QFThe abnormal type judging module Q is used for judging whether the characteristic vector comes from the output of the encoder or the normal distribution samplingCThe distribution vector for discriminating between anomaly types is from the encoder output or the classification distribution sample.
Preferably, the training sample set D is utilizedtThe step of training each IQ signal data to obtain a spectrum anomaly detection model based on a denoising countermeasure self-encoder includes:
randomly initializing the encoder module E, the decoder module D, and the feature discriminating module QFThe abnormality type discriminating module QCDetermining a preset training termination condition;
from the non-abnormal training sample set Dt1Randomly extracting a preset number of IQ signal data as first training samples, and calculating hidden space feature vectors and abnormal type distribution vectors corresponding to the first training samples through the encoder module E:
Figure BDA0003415341260000024
reconstructing a signal power spectral density corresponding to each first training sample by the decoder module D
Figure BDA0003415341260000025
And calculating the reconstruction error corresponding to each first training sample
Figure BDA0003415341260000026
From each reconstruction error LrUpdating network parameters corresponding to the encoder module E and the decoder module D, wherein B is the extraction quantity of the first training samples;
freezing the encoder module E, and sampling from the classification distribution to obtain the abnormal type label corresponding to each first training sample
Figure BDA0003415341260000027
Obtaining the characteristic vector corresponding to each first training sample by sampling from normal distribution
Figure BDA0003415341260000028
And recalculating the hidden space feature vector and the abnormal type distribution vector corresponding to each first training sample by using the encoder module E:
Figure BDA0003415341260000031
and through the abnormal type judging module QCJudging the distribution source of the abnormal type corresponding to each first training sample:
Figure BDA0003415341260000032
by the feature discrimination module QFJudgment of ziAnd
Figure BDA0003415341260000033
probability from prior sampling
Figure BDA0003415341260000034
Module for judging the type of the abnormality QCLoss of power
Figure BDA0003415341260000035
And updating the abnormal type discriminating module QCThe network parameter of (2), calculate the characteristic discrimination module QFLoss of power
Figure BDA0003415341260000036
Figure BDA0003415341260000037
And updating the feature discrimination module QFThe network parameter of (2);
unfreezing the encoder module E and recalculating
Figure BDA0003415341260000038
Computing the encoderLoss of module E
Figure BDA0003415341260000039
Updating the network parameters of the encoder module E;
from the abnormal training sample set Dt2Randomly extracting a preset number of IQ signal data as second training samples, and predicting the abnormal class of the input signal by the encoder module E
Figure BDA00034153412600000310
Calculating the encoder module E classification loss
Figure BDA00034153412600000311
Updating the network parameters of the encoder module E, wherein C is the extraction quantity of the second training samples;
repeatedly executing the steps until a preset training termination condition is met, and storing the encoder module E, the decoder module D and the feature discrimination module QFThe abnormality type discriminating module QCThe spectrum anomaly detection model based on the denoising countermeasure self-encoder is obtained through the structure and the network parameters.
Preferably, the verification sample set D is used as the basis of the verification sample setvDetermining a reconstruction error threshold, an abnormal type judgment loss threshold and a characteristic judgment loss threshold of the spectrum abnormal detection model by each IQ signal data in the IQ signal data, wherein the steps comprise:
set the verification samples DvTaking IQ signal data as verification sample xiInputting the vector into the encoder module E to obtain a hidden space feature vector and an abnormal type distribution vector:
Figure BDA00034153412600000312
reconstructing each verification sample x by the decoder module DiCorresponding signal power spectral density
Figure BDA00034153412600000313
And calculating a reconstruction error:
Figure BDA00034153412600000314
determining a reconstruction error threshold T of the spectrum anomaly detection modelr＝μr+α×σrAlpha is a hyperparameter, murAnd σrRespectively for all verification samples xiAverage reconstruction error and variance of;
calculate each validation sample xiAnd (3) judging loss of the corresponding abnormal type:
Figure BDA00034153412600000315
Figure BDA00034153412600000316
wherein,
Figure BDA00034153412600000317
determining an anomaly type judgment loss threshold T of the spectrum anomaly detection model from classified distribution samplingC＝μC+β×σCBeta is a hyperparameter, muCAnd σCRespectively for all verification samples xiJudging loss and variance according to the average abnormal type;
calculate each validation sample xiThe corresponding characteristics judge the loss:
Figure BDA00034153412600000318
wherein,
Figure BDA00034153412600000319
determining a characteristic judgment loss threshold T of the spectrum anomaly detection model from normal distribution samplingF＝μF+γ×σF. Gamma is a hyperparameter, muFAnd σFThe loss and variance are separately identified for the mean features of all validation samples.
Preferably, the step of inputting the IQ signal data to be detected to the spectrum anomaly detection model, calculating a reconstruction error, an anomaly type judgment loss and a feature judgment loss corresponding to the IQ signal data to be detected, comparing the reconstruction error, the anomaly type judgment loss and the feature judgment loss with the reconstruction error threshold, the anomaly type judgment loss threshold and the feature judgment loss threshold, and determining the anomaly state of the IQ signal data to be detected according to the comparison result includes:
calculating power spectrum density x of IQ signal data to be detected, inputting the power spectrum density x corresponding to the IQ signal data to be detected into the spectrum anomaly detection model, and calculating to obtain reconstruction error corresponding to the IQ signal data to be detected
Figure BDA0003415341260000041
Figure BDA0003415341260000042
Outputting the IQ signal data to be detected to the decoder module D, and further calculating to obtain the abnormal type judgment loss l corresponding to the IQ signal data to be detectedCAnd feature judgment loss lF；
When A ═ l is detectedr＞Tr)∧(lC＞TC)∧(lF＞TF) And if the detected IQ signal data is true, determining that the IQ signal data to be detected is abnormal, and determining the class corresponding to the maximum value in the abnormal type distribution vector output by the encoder module E as the abnormal type of the IQ signal data to be detected.
Compared with the prior art, the spectrum anomaly detection method based on the denoising countermeasure self-encoder has the following advantages:
the invention provides a frequency spectrum anomaly detection method based on a denoising countermeasure self-encoder, which enhances the robustness of a frequency spectrum anomaly detection model to noise by arranging a denoising layer in an encoder module, so that the frequency spectrum anomaly detection model obtained by training has an accurate detection effect under a low signal-to-noise ratio, the encoder module maps the power spectral density of an input signal to a hidden space characteristic and can classify the frequency spectrum anomaly type distribution, a network of the encoder module adopts a semi-supervised learning mode for training, and the frequency spectrum anomaly detection model can have the frequency spectrum anomaly type prediction capability under the training of a small amount of labeled samples; the invention also adopts counterstudy, can control the distribution of the coded hidden space feature vector and the abnormal type distribution vector compared with the existing self-encoder, has more stable and accurate acquisition of the signal feature and has good spectrum abnormality detection capability.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.
FIG. 1 is a schematic diagram of an implementation environment provided by one embodiment of the invention.
FIG. 2 is a schematic diagram of another implementation environment provided by an embodiment of the invention.
Fig. 3 is a flow chart illustrating a method for spectral anomaly detection based on denoising counterautoencoder according to an exemplary embodiment.
Fig. 4 is a schematic diagram illustrating detection of a spectrum anomaly detection model corresponding to a spectrum anomaly detection method based on a denoising countermeasure self-encoder according to an embodiment of the present invention.
Fig. 5 is a schematic diagram of a network structure of an encoder module in a spectrum anomaly detection model according to an embodiment of the present invention.
Fig. 6 is a schematic network structure diagram of a decoder module in a spectrum anomaly detection model according to an embodiment of the present invention.
Fig. 7 is a block diagram illustrating an apparatus for implementing a denoising-based countering self-encoder spectral anomaly detection method according to an exemplary embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in detail below with reference to specific embodiments (but not limited to) and the accompanying drawings, and it is apparent that the described embodiments are only a part of the embodiments of the present invention, rather than 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.
In one possible implementation, a schematic diagram of an implementation environment related to an embodiment of the present invention may be as shown in fig. 1. In the schematic diagram of the implementation environment shown in fig. 1, the implementation environment includes at least one IQ signal transmission source 10 and a server 30.
Wherein, the IQ signal emission source 10 is used for generating IQ signal data;
the server 30 is configured to receive IQ signal data sent by each IQ signal emission source 10, and determine an abnormal state of each IQ signal data through a trained spectrum abnormality detection model; further, the server 30 may display the abnormal state of each IQ signal data through a display device such as a display screen.
In another possible implementation, a schematic diagram of an implementation environment related to an embodiment of the present invention may be as shown in fig. 2. In the schematic diagram of the implementation environment shown in fig. 2, the implementation environment includes at least one IQ signal transmission source 10, at least one receiving terminal 30, and a server 30.
Wherein, the IQ signal emission source 10 is used for generating IQ signal data;
the receiving terminal 20 is configured to collect IQ signal data generated by each IQ signal emission source 10, send each IQ signal data to the server 30, and then receive abnormal state data of each IQ signal data fed back by the server 30, and the receiving terminal 20 is in electrical signal connection with the server 30 in a wired or wireless manner;
the server 30 is configured to receive the IQ signal data sent by each receiving terminal 20, determine an abnormal state of each IQ signal data through the trained spectrum abnormality detection model, and feed back the abnormal state data of each IQ signal data to the receiving terminal 20; further, the server 20 may display the abnormal state of each IQ signal data through a display device such as a display screen.
Fig. 3 is a flowchart illustrating a method for detecting spectrum abnormality of an anti-self-encoder based on denoising according to an exemplary embodiment, and as shown in fig. 2, there is provided a method for detecting spectrum abnormality of an anti-self-encoder based on denoising, the method comprising:
step 100: calculating the signal power spectral density of each IQ signal data, and dividing each IQ signal data into a training sample set D according to the abnormal state of the signal power spectral density corresponding to each IQ signal datatAnd validating the sample set DvWherein the training sample set DtComprising a non-anomalous training sample set D consisting of unlabeled IQ signal datat1And an abnormal training sample set D composed of IQ signal data including an abnormal type tagt2The verification sample set DvAre IQ signal data including an exception type tag.
Step 200: using the training sample set DtTraining each IQ signal data to obtain a spectrum anomaly detection model based on a denoising countermeasure self-encoder, and verifying a sample set D according to the spectrum anomaly detection modelvDetermining a reconstruction error threshold, an abnormal type judgment loss threshold and a characteristic judgment loss threshold of the spectrum abnormal detection model by the IQ signal data in the spectrum abnormal detection model.
Step 300: inputting IQ signal data to be detected into the spectrum anomaly detection model, calculating a reconstruction error, an anomaly type judgment loss and a characteristic judgment loss corresponding to the IQ signal data to be detected, comparing the reconstruction error, the anomaly type judgment loss and the characteristic judgment loss with the reconstruction error threshold, the anomaly type judgment loss threshold and the characteristic judgment loss threshold respectively, and determining the anomaly state of the IQ signal data to be detected according to the comparison result.
Preferably, the non-abnormal training sample set
Figure BDA0003415341260000061
The abnormal training sample set
Figure BDA0003415341260000062
The verification sample set
Figure BDA0003415341260000063
Wherein n1 is the non-abnormal training sample set Dt1The number of IQ signal data, n2 being the abnormal training sample set Dt2Number of intermediate IQ signal data, xiIs the signal power spectral density, yiIs an exception type tag.
Preferably, the spectrum abnormal model is composed of an encoder module E, a decoder module D and a characteristic discrimination module QFAnd abnormalityType discrimination module QCThe decoder module D is used for reconstructing the signal power spectral density corresponding to the IQ signal data according to the hidden space eigenvector corresponding to the IQ signal data and the abnormal type distribution vector, and the characteristic discrimination module Q is used for determining the characteristic of the IQ signal dataFThe abnormal type judging module Q is used for judging whether the characteristic vector comes from the output of the encoder or the normal distribution samplingCThe distribution vector for discriminating between anomaly types is from the encoder output or the classification distribution sample.
Preferably, the training sample set D is utilizedtThe step of training each IQ signal data to obtain a spectrum anomaly detection model based on a denoising countermeasure self-encoder includes:
step 210: randomly initializing the encoder module E, the decoder module D, and the feature discriminating module QFThe abnormality type discriminating module QCAnd determining the number of training iterations.
Step 220: from the non-abnormal training sample set Dt1Randomly extracting a preset number of IQ signal data as first training samples, and calculating hidden space feature vectors and abnormal type distribution vectors corresponding to the first training samples through the encoder module E:
Figure BDA0003415341260000071
reconstructing a signal power spectral density corresponding to each first training sample by the decoder module D
Figure BDA0003415341260000072
And calculating the reconstruction error corresponding to each first training sample
Figure BDA0003415341260000073
From each reconstruction error LrAnd updating the network parameters corresponding to the encoder module E and the decoder module D, wherein B is the extraction number of the first training samples.
Step (ii) of230: freezing the encoder module E, and sampling from the classification distribution to obtain the abnormal type label corresponding to each first training sample
Figure BDA0003415341260000074
Obtaining the characteristic vector corresponding to each first training sample by sampling from normal distribution
Figure BDA0003415341260000075
And recalculating the hidden space feature vector and the abnormal type distribution vector corresponding to each first training sample by using the encoder module E:
Figure BDA0003415341260000076
and through the abnormal type judging module QCJudging the distribution source of the abnormal type corresponding to each first training sample:
Figure BDA0003415341260000077
by the feature discrimination module QFJudgment of ziAnd
Figure BDA0003415341260000078
probability from prior sampling
Figure BDA0003415341260000079
Module for judging the type of the abnormality QCLoss of power
Figure BDA00034153412600000710
And updating the abnormal type discriminating module QCThe network parameter of (2), calculate the characteristic discrimination module QFLoss of power
Figure BDA00034153412600000711
And updating the feature discrimination module QFThe network parameter of (2).
Step 240: unfreezing the encoder module E and recalculating
Figure BDA00034153412600000712
Calculating the codeLoss of device module E
Figure BDA00034153412600000713
And updating the network parameters of the encoder module E.
Step 250: from the abnormal training sample set Dt2Randomly extracting a preset number of IQ signal data as second training samples, and predicting the abnormal class of the input signal by the encoder module E
Figure BDA00034153412600000714
Calculating the encoder module E classification loss
Figure BDA00034153412600000715
And updating the network parameters of the encoder module E, wherein C is the number of second training samples extracted.
Step 260: repeating the steps 210 to 250 until a preset training termination condition is satisfied, and saving the encoder module E, the decoder module D, and the feature determination module QFThe abnormality type discriminating module QCThe spectrum anomaly detection model based on the denoising countermeasure self-encoder is obtained through the structure and the network parameters.
Preferably, the verification sample set D is used as the basis of the verification sample setvDetermining a reconstruction error threshold, an abnormal type judgment loss threshold and a characteristic judgment loss threshold of the spectrum abnormal detection model by each IQ signal data in the IQ signal data, wherein the steps comprise:
step 270: set the verification samples DvTaking IQ signal data as verification sample xiInputting the vector into the encoder module E to obtain a hidden space feature vector and an abnormal type distribution vector:
Figure BDA0003415341260000081
reconstructing each verification sample x by the decoder module DiCorresponding signal power spectral density
Figure BDA0003415341260000082
And calculating a reconstruction error:
Figure BDA0003415341260000083
determining a reconstruction error threshold T of the spectrum anomaly detection modelr＝μr+α×σrAlpha is a hyperparameter, murAnd σrRespectively for all verification samples xiAverage reconstruction error and variance.
Step 280: calculate each validation sample xiAnd (3) judging loss of the corresponding abnormal type:
Figure BDA0003415341260000084
Figure BDA0003415341260000085
wherein,
Figure BDA0003415341260000086
determining an anomaly type judgment loss threshold T of the spectrum anomaly detection model from classified distribution samplingC＝μC+β×σCBeta is a hyperparameter, muCAnd σCRespectively for all verification samples xiThe average anomaly type of (2) discriminates the loss and variance.
Step 290: calculate each validation sample xiThe corresponding characteristics judge the loss:
Figure BDA0003415341260000087
Figure BDA0003415341260000088
wherein,
Figure BDA0003415341260000089
determining a characteristic judgment loss threshold T of the spectrum anomaly detection model from normal distribution samplingF＝μF+γ×σF. Gamma is a hyperparameter, muFAnd σFThe loss and variance are separately identified for the mean features of all validation samples.
Preferably, the step of inputting the IQ signal data to be detected to the spectrum anomaly detection model, calculating a reconstruction error, an anomaly type judgment loss and a feature judgment loss corresponding to the IQ signal data to be detected, comparing the reconstruction error, the anomaly type judgment loss and the feature judgment loss with the reconstruction error threshold, the anomaly type judgment loss threshold and the feature judgment loss threshold, and determining the anomaly state of the IQ signal data to be detected according to the comparison result includes:
step 310: calculating power spectrum density x of IQ signal data to be detected, inputting the power spectrum density x corresponding to the IQ signal data to be detected into the spectrum anomaly detection model, and calculating to obtain reconstruction error corresponding to the IQ signal data to be detected
Figure BDA00034153412600000810
Figure BDA00034153412600000811
Outputting the IQ signal data to be detected to the decoder module D, and further calculating to obtain the abnormal type judgment loss l corresponding to the IQ signal data to be detectedCAnd feature judgment loss lF。
Step 320: when A ═ l is detectedr＞Tr)∧(lC＞TC)∧(lF＞TF) And if the detected IQ signal data is true, determining that the IQ signal data to be detected is abnormal, and determining the class corresponding to the maximum value in the abnormal type distribution vector output by the encoder module E as the abnormal type of the IQ signal data to be detected.
Note that when a ═ is detected (l)r＞Tr)∧(lC＞TC)∧(lF＞TF) If the IQ signal data to be detected is true, determining that the IQ signal data to be detected is abnormal, and determining that the IQ signal data to be detected is not abnormal.
The embodiment of the invention can also display the abnormal state on the display screen for the user to know the specific abnormal content of the IQ signal to be detected.
In a possible implementation manner, a detection schematic diagram of a spectrum anomaly detection model corresponding to a spectrum anomaly detection method based on a denoising countermeasure self-encoder provided by an embodiment of the present invention may be shown in fig. 4.
In a possible implementation manner, a schematic diagram of a network structure of an encoder module in a spectrum anomaly detection model provided by an embodiment of the present invention may be as shown in fig. 5.
In a possible implementation manner, a schematic diagram of a network structure of a decoder module in a spectrum anomaly detection model provided by an embodiment of the present invention may be as shown in fig. 6.
In summary, the method for detecting spectrum anomaly based on a denoising countermeasure self-encoder provided by the invention enhances the robustness of a spectrum anomaly detection model to noise by arranging a denoising layer in an encoder module, so that the trained spectrum anomaly detection model has an accurate detection effect under a low signal-to-noise ratio, the encoder module maps the power spectral density of an input signal to a hidden space characteristic and can classify the distribution of spectrum anomaly types, a network of the encoder module is trained in a semi-supervised learning manner, and the spectrum anomaly detection model can have the prediction capability of the spectrum anomaly types under the training of a small amount of labeled samples; the invention also adopts counterstudy, can control the distribution of the coded hidden space feature vector and the abnormal type distribution vector compared with the existing self-encoder, has more stable and accurate acquisition of the signal feature and has good spectrum abnormality detection capability.
In one possible implementation, fig. 7 is a block diagram of an apparatus for implementing a denoising-based countering self-encoder spectrum anomaly detection method according to an exemplary embodiment. For example, the apparatus 700 may be provided as a server. Referring to fig. 7, apparatus 700 includes a processing component 722 that further includes one or more processors and memory resources, represented by memory 732, for storing instructions, such as applications, that are executable by processing component 722. The application programs stored in memory 732 may include one or more modules that each correspond to a set of instructions. Further, the processing component 722 is configured to execute instructions to perform the above-described denoising-based spectrum anomaly detection method against a self-encoder.
The apparatus 700 may also include a power component 726 configured to perform power management of the apparatus 700, a wired or wireless network interface 750 configured to connect the apparatus 700 to a network, and an input output (I/O) interface 758. The apparatus 700 may operate based on an operating system stored in memory 732, such as Windows Server, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, or the like.
While the invention has been described in detail in the foregoing by way of general description, and specific embodiments and experiments, it will be apparent to those skilled in the art that modifications and improvements can be made thereto based on the invention. Accordingly, such modifications and improvements are intended to be within the scope of the invention as claimed.
Other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the invention and including such departures from the present disclosure as come within known or customary practice within the art to which the invention pertains. It will be understood that the invention is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof.

Claims (6)

1. A spectrum anomaly detection method for a self-encoder based on denoising countermeasure is characterized by comprising the following steps:
calculating the signal power spectral density of each IQ signal data, and dividing each IQ signal data into a training sample set D according to the abnormal state of the signal power spectral density corresponding to each IQ signal datatAnd validating the sample set DvWherein the training sample set DtComprising a non-anomalous training sample set D consisting of unlabeled IQ signal datat1And an abnormal training sample set D composed of IQ signal data including an abnormal type tagt2The verification sample set DvAre all IQ signal data including an abnormal type tag;
using the training sample set DtIn (1)Training each IQ signal data to obtain a spectrum anomaly detection model based on a denoising countermeasure autoencoder, and verifying a sample set D according to the spectrum anomaly detection modelvDetermining a reconstruction error threshold, an abnormal type judgment loss threshold and a characteristic judgment loss threshold of the spectrum abnormal detection model by each IQ signal data;
inputting IQ signal data to be detected into the spectrum anomaly detection model, calculating a reconstruction error, an anomaly type judgment loss and a characteristic judgment loss corresponding to the IQ signal data to be detected, comparing the reconstruction error, the anomaly type judgment loss and the characteristic judgment loss with the reconstruction error threshold, the anomaly type judgment loss threshold and the characteristic judgment loss threshold respectively, and determining the anomaly state of the IQ signal data to be detected according to the comparison result.
2. The method of claim 1, wherein the non-abnormal training sample set
Figure FDA0003415341250000011
The abnormal training sample set
Figure FDA0003415341250000012
The verification sample set
Figure FDA0003415341250000013
Wherein n1 is the non-abnormal training sample set Dt1The number of IQ signal data, n2 being the abnormal training sample set Dt2Number of intermediate IQ signal data, xiIs the signal power spectral density, yiIs an exception type tag.
3. The method of claim 2, wherein the spectral anomaly model is generated by an encoder module E, a decoder module D, and a feature discrimination module QFAnd an abnormality type discriminating module QCThe encoder module E is used for mapping the signal power spectral density corresponding to the IQ signal data to a hidden space to obtain a hidden space feature vector and judging an abnormal type distribution vector of the IQ signal data, and the decoder moduleThe block D is used for reconstructing signal power spectral density corresponding to IQ signal data according to the hidden space feature vector and the abnormal type distribution vector corresponding to the IQ signal data, and the feature discrimination module QFThe abnormal type judging module Q is used for judging whether the characteristic vector comes from the output of the encoder or the normal distribution samplingCThe distribution vector for discriminating between anomaly types is from the encoder output or the classification distribution sample.
4. The method of claim 3, wherein the utilizing the training sample set DtThe step of training each IQ signal data to obtain a spectrum anomaly detection model based on a denoising countermeasure self-encoder includes:
randomly initializing the encoder module E, the decoder module D, and the feature discriminating module QFThe abnormality type discriminating module QCDetermining a preset training termination condition;
from the non-abnormal training sample set Dt1Randomly extracting a preset number of IQ signal data as first training samples, and calculating hidden space feature vectors and abnormal type distribution vectors corresponding to the first training samples through the encoder module E:
Figure FDA00034153412500000218
reconstructing a signal power spectral density corresponding to each first training sample by the decoder module D
Figure FDA0003415341250000022
And calculating the reconstruction error corresponding to each first training sample
Figure FDA0003415341250000023
From each reconstruction error LrUpdating network parameters corresponding to the encoder module E and the decoder module D, wherein B is the extraction quantity of the first training samples;
freezing the encoder module E, and sampling from the classification distribution to obtain the abnormal type label corresponding to each first training sample
Figure FDA0003415341250000024
Obtaining the characteristic vector corresponding to each first training sample by sampling from normal distribution
Figure FDA0003415341250000025
And recalculating the hidden space feature vector and the abnormal type distribution vector corresponding to each first training sample by using the encoder module E:
Figure FDA00034153412500000219
and through the abnormal type judging module QCJudging the distribution source of the abnormal type corresponding to each first training sample:
Figure FDA0003415341250000027
by the feature discrimination module QFJudgment of ziAnd
Figure FDA0003415341250000028
probability from prior sampling
Figure FDA0003415341250000029
Module for judging the type of the abnormality QCLoss of power
Figure FDA00034153412500000210
And updating the abnormal type discriminating module QCThe network parameter of (2), calculate the characteristic discrimination module QFLoss of power
Figure FDA00034153412500000211
Figure FDA00034153412500000212
And updating the feature discrimination module QFThe network parameter of (2);
unfreezing the encoder module E and recalculating
Figure FDA00034153412500000213
Calculating the encoder module E loss
Figure FDA00034153412500000214
Updating the network parameters of the encoder module E;
from the abnormal training sample set Dt2Randomly extracting a preset number of IQ signal data as second training samples, and predicting the abnormal class of the input signal by the encoder module E
Figure FDA00034153412500000215
Calculating the encoder module E classification loss
Figure FDA00034153412500000216
Updating the network parameters of the encoder module E, wherein C is the extraction quantity of the second training samples;
repeatedly executing the steps until a preset training termination condition is met, and storing the encoder module E, the decoder module D and the feature discrimination module QFThe abnormality type discriminating module QCThe spectrum anomaly detection model based on the denoising countermeasure self-encoder is obtained through the structure and the network parameters.
5. The method of claim 4, wherein the verifying the sample set DvDetermining a reconstruction error threshold, an abnormal type judgment loss threshold and a characteristic judgment loss threshold of the spectrum abnormal detection model by each IQ signal data in the IQ signal data, wherein the steps comprise:
set the verification samples DvTaking IQ signal data as verification sample xiInputting the vector into the encoder module E to obtain a hidden space feature vector and an abnormal type distribution vector:
Figure FDA00034153412500000217
reconstructing the experience by the decoder module DCertificate sample xiCorresponding signal power spectral density
Figure FDA0003415341250000031
And calculating a reconstruction error:
Figure FDA0003415341250000032
determining a reconstruction error threshold T of the spectrum anomaly detection modelr＝μr+α×σrAlpha is a hyperparameter, murAnd σrRespectively for all verification samples xiAverage reconstruction error and variance of;
calculate each validation sample xiAnd (3) judging loss of the corresponding abnormal type:
Figure FDA0003415341250000033
Figure FDA0003415341250000034
wherein,
Figure FDA0003415341250000035
determining an anomaly type judgment loss threshold T of the spectrum anomaly detection model from classified distribution samplingC＝μC+β×σCBeta is a hyperparameter, muCAnd σCRespectively for all verification samples xiJudging loss and variance according to the average abnormal type;
calculate each validation sample xiThe corresponding characteristics judge the loss:
Figure FDA0003415341250000036
wherein,
Figure FDA0003415341250000037
determining a characteristic judgment loss threshold T of the spectrum anomaly detection model from normal distribution samplingF＝μF+γ×σF. Gamma is a hyperparameter, muFAnd σFRespectively average characteristics of all verification samplesAnd characterizing discriminant loss and variance.
6. The method according to claim 5, wherein the step of inputting the IQ signal data to be detected to the spectrum anomaly detection model, calculating a reconstruction error, an anomaly type judgment loss and a characteristic judgment loss corresponding to the IQ signal data to be detected, comparing the reconstruction error, the anomaly type judgment loss and the characteristic judgment loss with the reconstruction error threshold, the anomaly type judgment loss threshold and the characteristic judgment loss threshold respectively, and determining the anomaly state of the IQ signal data to be detected according to the comparison result comprises:
calculating power spectrum density x of IQ signal data to be detected, inputting the power spectrum density x corresponding to the IQ signal data to be detected into the spectrum anomaly detection model, and calculating to obtain reconstruction error corresponding to the IQ signal data to be detected
Figure FDA0003415341250000038
Figure FDA0003415341250000039
Outputting the IQ signal data to be detected to the decoder module D, and further calculating to obtain the abnormal type judgment loss l corresponding to the IQ signal data to be detectedCAnd feature judgment loss lF；
When A ═ l is detectedr>Tr)∧(lC>TC)∧(lF>TF) And if the detected IQ signal data is true, determining that the IQ signal data to be detected is abnormal, and determining the class corresponding to the maximum value in the abnormal type distribution vector output by the encoder module E as the abnormal type of the IQ signal data to be detected.
CN202111544290.2A 2021-12-16 2021-12-16 Denoising countermeasure self-encoder-based spectrum anomaly detection method Pending CN114254673A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111544290.2A CN114254673A (en) 2021-12-16 2021-12-16 Denoising countermeasure self-encoder-based spectrum anomaly detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111544290.2A CN114254673A (en) 2021-12-16 2021-12-16 Denoising countermeasure self-encoder-based spectrum anomaly detection method

Publications (1)

Publication Number Publication Date
CN114254673A true CN114254673A (en) 2022-03-29

Family

ID=80795317

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111544290.2A Pending CN114254673A (en) 2021-12-16 2021-12-16 Denoising countermeasure self-encoder-based spectrum anomaly detection method

Country Status (1)

Country Link
CN (1) CN114254673A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115856425A (en) * 2022-11-21 2023-03-28 中国人民解放军32802部队 Spectrum anomaly detection method and device based on hidden space probability prediction
CN116664000A (en) * 2023-06-13 2023-08-29 无锡物联网创新中心有限公司 Industrial equipment abnormality detection method and related device based on long-short-term memory network
CN117292717A (en) * 2023-11-27 2023-12-26 广东美的制冷设备有限公司 Abnormal sound identification method, device, electronic equipment and storage medium

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115856425A (en) * 2022-11-21 2023-03-28 中国人民解放军32802部队 Spectrum anomaly detection method and device based on hidden space probability prediction
CN115856425B (en) * 2022-11-21 2023-10-17 中国人民解放军32802部队 Spectrum anomaly detection method and device based on hidden space probability prediction
CN116664000A (en) * 2023-06-13 2023-08-29 无锡物联网创新中心有限公司 Industrial equipment abnormality detection method and related device based on long-short-term memory network
CN117292717A (en) * 2023-11-27 2023-12-26 广东美的制冷设备有限公司 Abnormal sound identification method, device, electronic equipment and storage medium
CN117292717B (en) * 2023-11-27 2024-03-22 广东美的制冷设备有限公司 Abnormal sound identification method, device, electronic equipment and storage medium

Similar Documents

Publication Publication Date Title
CN114254673A (en) Denoising countermeasure self-encoder-based spectrum anomaly detection method
CN109408389B (en) Code defect detection method and device based on deep learning
CN105740707B (en) The recognition methods of malicious file and device
US20220053010A1 (en) System and method for determining a communication anomaly in at least one network
CN112597864A (en) Monitoring video abnormity detection method and device
CN103310235B (en) A kind of steganalysis method based on parameter identification and estimation
CN111967535B (en) Fault diagnosis method and device for temperature sensor of grain storage management scene
CN114372493B (en) Computer cable electromagnetic leakage characteristic analysis method
CN114861172B (en) Data processing method and system based on government affair service system
CN113328908B (en) Abnormal data detection method and device, computer equipment and storage medium
CN109525994A (en) High energy efficiency frequency spectrum sensing method based on support vector machines
CN116910752B (en) Malicious code detection method based on big data
CN116894985B (en) Semi-supervised image classification method and semi-supervised image classification system
CN112131575A (en) Concept drift detection method based on classification error rate and consistency prediction
CN111522736A (en) Software defect prediction method and device, electronic equipment and computer storage medium
CN114169398A (en) Photovoltaic direct-current arc fault identification method and device based on random forest algorithm
CN114328942A (en) Relationship extraction method, apparatus, device, storage medium and computer program product
CN111259926A (en) Meat freshness detection method and device, computing equipment and storage medium
CN116484184A (en) Method and device for enhancing partial discharge defect sample of power equipment
CN113033817B (en) OOD detection method and device based on hidden space, server and storage medium
CN111209567B (en) Method and device for judging perceptibility of improving robustness of detection model
CN110401639B (en) Method and device for judging abnormality of network access, server and storage medium thereof
US20220405585A1 (en) Training device, estimation device, training method, and training program
CN108052987B (en) Method for detecting image classification output result
CN110728615B (en) Steganalysis method based on sequential hypothesis testing, terminal device and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20220329

WD01 Invention patent application deemed withdrawn after publication