CN112435356B - ETC interference signal identification method and detection system - Google Patents

ETC interference signal identification method and detection system Download PDF

Info

Publication number
CN112435356B
CN112435356B CN202011213489.2A CN202011213489A CN112435356B CN 112435356 B CN112435356 B CN 112435356B CN 202011213489 A CN202011213489 A CN 202011213489A CN 112435356 B CN112435356 B CN 112435356B
Authority
CN
China
Prior art keywords
signal
module
data
signals
interference signal
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
Application number
CN202011213489.2A
Other languages
Chinese (zh)
Other versions
CN112435356A (en
Inventor
曾德国
鲁加战
李志坚
李振宇
刘晓俊
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.)
Nanjing Aerospace Industry Technology Co ltd
Original Assignee
Nanjing Aerospace Industry 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 Nanjing Aerospace Industry Technology Co ltd filed Critical Nanjing Aerospace Industry Technology Co ltd
Priority to CN202011213489.2A priority Critical patent/CN112435356B/en
Publication of CN112435356A publication Critical patent/CN112435356A/en
Application granted granted Critical
Publication of CN112435356B publication Critical patent/CN112435356B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G07—CHECKING-DEVICES
    • G07B—TICKET-ISSUING APPARATUS; FARE-REGISTERING APPARATUS; FRANKING APPARATUS
    • G07B15/00—Arrangements or apparatus for collecting fares, tolls or entrance fees at one or more control points
    • G07B15/06—Arrangements for road pricing or congestion charging of vehicles or vehicle users, e.g. automatic toll systems
    • G07B15/063—Arrangements for road pricing or congestion charging of vehicles or vehicle users, e.g. automatic toll systems using wireless information transmission between the vehicle and a fixed station
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/10—Monitoring; Testing of transmitters
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/20—Monitoring; Testing of receivers
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • General Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Electromagnetism (AREA)
  • Health & Medical Sciences (AREA)
  • Signal Processing (AREA)
  • Business, Economics & Management (AREA)
  • Finance (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to an identification method of an ETC interference signal and a detection system for realizing the same, wherein the identification method of the ETC interference signal is combined with hardware equipment in the detection system to realize identification of the ETC interference signal. Firstly, initializing parameters of a receiver for detecting signals; secondly, directionally receiving electromagnetic signals by combining with a ground induction coil, transmitting the received signals to an individual identification server through a network, and performing classified storage; thirdly, judging the received wireless signals through spectrum analysis; secondly, when the judgment result is the interference signal, performing deep learning identification on the interference signal through a migration learning identification method based on a convolutional neural network, and extracting individual features; and finally, calling a camera to shoot a vehicle picture, transmitting the vehicle picture to an individual recognition server, and storing evidence and providing a basis for later crime tracking and discrimination.

Description

ETC interference signal identification method and detection system
Technical Field
The invention relates to an identification method and a detection system of an ETC interference signal, belonging to the technical field of electric digital data processing G06F.
Background
With the rapid development of intelligent traffic, the electronic toll collection system becomes one of the representative products for reducing the labor cost at the high-speed intersection. The ETC is realized by realizing data interaction processing through microwave communication between a vehicle-mounted unit arranged in front of the inner side of a vehicle windshield and a road test unit of a toll gate, recording the position of a vehicle entering and exiting a high-speed gate by means of a traffic system, and paying for actual mileage.
However, at present, part of lawbreakers use radio jammers to interfere with ETC charging systems, so as to achieve the purpose of avoiding road tolls, cause certain economic loss, and generate great negative effects on the development of highway toll collection, thereby aggravating subsequent processing work such as payment and legal responsibility pursuit. In the prior art, a newly-built deep learning network model is adopted for signal feature recognition, the training time is too long, and the improvement of the accuracy requires massive data training set and longer time cost.
Disclosure of Invention
The purpose of the invention is as follows: an identification method and a detection system for an ETC interference signal are provided to solve the above problems in the prior art. The further purpose is to reduce the occurrence of the subsequent pursuit work progress obstruction when the charging fails by tracking the detected abnormal signals.
The technical scheme is as follows: an ETC interference signal identification method is characterized by comprising the following steps:
initializing hardware parameters of a detection signal receiver, and establishing communication connection between the detection signal receiver and an individual identification server;
step two, acquiring a signal;
step three, receiving the signals obtained in the step two, performing spectrum analysis on the received signals, and identifying whether the signals are abnormal according to spectrum change;
step four, inputting the interference signal image data analyzed in the step three into the constructed signal identification network model to extract the signal characteristics, thereby judging the type of the received interference signal;
after the signal recognition network model finely adjusts the predefined module, the data set is trained again, so that the learning capacity of extracting the signal characteristics is trained;
and step five, storing the vehicle related information with the interference signal.
In a further embodiment. The second step is further as follows: judging the passing state of the vehicle by using a vehicle detector, and triggering a receiver to read a corresponding signal when the judged state is that the passing condition of the vehicle exists; the vehicle detector detects the change of electric quantity of oscillation frequency in a coupling circuit formed by the annular coil by using a detector; the electric quantity change is generated by the magnetic flux change in the ground induction coil when the vehicle reaches the ETC inlet and passes through the upper surface of the ground induction coil; simultaneously the detector outputs two corresponding groups of logic signals; the two groups of logic signals comprise signals entering the ground induction coil and signals leaving the ground induction coil; the signal leaving the ground induction coil is used for triggering a receiver to perform corresponding signal reading.
In a further embodiment, the third step is further: identifying interference signals by combining spectral analysis according to the interactive signals acquired in the step two; the judgment basis of the frequency spectrum analysis is that when an interference signal exists, the noise of the electromagnetic signal around the frequency band of the ETC working frequency band is obviously improved, and the normal ETC working signal is suppressed by the increase of the noise of the ETC working frequency band; when normal ETC communication is not interfered, the effective signal energy is higher in an ETC working frequency band, and the non-effective signal area is lower; after normal ETC communication receives the interference, whole ETC working frequency range and peripheral signal will all be raised to submerge the effective signal.
In a further embodiment, the fourth step is further: the specific steps of the construction process of the signal identification network model further comprise:
4-1, constructing a model training data set;
step 4-2, fine tuning the convolution network;
4-3, determining a difference measurement method between the target field and the source field;
4-4, constructing a loss function for weight adjustment;
and 4-5, receiving a training data set, and training the learning ability of the model for extracting the signal characteristics.
Step 4-2 is further to finely adjust the convolutional neural network after the convolutional network is trained; the fine adjustment mode is realized by fixing parameters of the predefined modules and changing the structures of partial modules.
Step 4-3 further uses data in the source domain and data in the target domain, and after the function random projection, the supremum of the expected difference between the two solves the problem that the target domain has no notes, and the implementation manner is shown in the following expression:
Figure 348238DEST_PATH_IMAGE001
further comprises the following steps:
Figure 838125DEST_PATH_IMAGE002
further comprises the following steps:
Figure 617731DEST_PATH_IMAGE003
in the formula
Figure 370924DEST_PATH_IMAGE004
，
Figure 49030DEST_PATH_IMAGE005
Representing a function
Figure 14712DEST_PATH_IMAGE006
The space in which the device is located is provided with a plurality of grooves,
Figure 461873DEST_PATH_IMAGE007
to represent
Figure 120388DEST_PATH_IMAGE008
The distribution of (a) to (b) is,
Figure 761757DEST_PATH_IMAGE009
to represent
Figure 593447DEST_PATH_IMAGE010
The distribution of (a) to (b) is,
Figure 832799DEST_PATH_IMAGE011
representing source domain functions
Figure 927794DEST_PATH_IMAGE012
In the expectation that the position of the target is not changed,
Figure 580492DEST_PATH_IMAGE013
representing a target domain function
Figure 887976DEST_PATH_IMAGE014
(iii) a desire; for the best-case core, further feature mapping is performed, and the cores are weighted using m different gaussian cores, that is:
Figure 309730DEST_PATH_IMAGE015
in the formula (I), the compound is shown in the specification,
Figure 559315DEST_PATH_IMAGE016
representing the weight, for the coefficient
Figure 433730DEST_PATH_IMAGE016
The resulting k is guaranteed to be unique.
Said step 4-4 further comprising constructing a loss function comprising a two-part loss; wherein, part of the loss is the loss of the traditional convolutional neural network, and the cross entropy is adopted as a loss function; the other part of loss is the MK-MMD distance of the source domain at the output domain of the full connection layer and the target domain at the output of the full connection layer, and the loss function is shown in the following expression:
Figure 76064DEST_PATH_IMAGE017
in the formula (I), the compound is shown in the specification,
Figure 290008DEST_PATH_IMAGE018
a penalty coefficient is expressed for controlling the distribution difference of the source domain and the target domain in the full connection layer, and
Figure 523543DEST_PATH_IMAGE018
satisfy the requirement of
Figure 88517DEST_PATH_IMAGE019
；
Figure 987071DEST_PATH_IMAGE020
Indicating a starting layer for performing domain adaptation;
Figure 383418DEST_PATH_IMAGE021
indicating an end layer for performing domain adaptation;
Figure 725537DEST_PATH_IMAGE022
representing the number of the data with notes in the source domain and the target domain;
Figure 574544DEST_PATH_IMAGE023
parameters representing the entire neural network;
Figure 293102DEST_PATH_IMAGE024
representing the output of the convolutional neural network;
Figure 481638DEST_PATH_IMAGE025
which represents the input image, is,
Figure 56975DEST_PATH_IMAGE026
to represent
Figure 845809DEST_PATH_IMAGE025
A corresponding label;
Figure 102478DEST_PATH_IMAGE027
indicating source domain data in
Figure 207837DEST_PATH_IMAGE028
Outputting the layer;
Figure 891759DEST_PATH_IMAGE029
indicating that the target data field is in
Figure 449780DEST_PATH_IMAGE028
Outputting the layer;
Figure 775719DEST_PATH_IMAGE030
representing a source domain andtarget domain data is in
Figure 922535DEST_PATH_IMAGE028
The MK-MMD distance of the layer output squared.
In a further embodiment, the step five is further: and transmitting the signal which is judged as the interference signal in the third step into a signal identification network model constructed in the fourth step, identifying and analyzing the signal characteristics, performing characteristic matching according to the existing interference signal information in the individual identification server so as to judge whether the signal is the same interference transmitter, and further transmitting the processed data characteristics to the individual identification server for data storage.
An ETC interference signal detection system for implementing the method, comprising:
a first module for detecting a vehicle passing condition;
a second module for determining an interference signal;
a third module for identifying an interfering signal type;
a fourth module for storing data.
In a further embodiment, the first module further comprises a ground induction coil module and a detector module; the ground induction coil module is used for data acquisition, and the detector module further comprises a data judgment module and a logic signal output module; the data in the data acquisition is further a change value of inductance in a ground induction coil loop caused by changing magnetic flux in the ground induction coil when a vehicle passes through the ground induction coil; the data judgment module analyzes the inductance data according to the data acquisition; the logic signal output module sends two groups of relay signals according to the data analyzed by the data judgment module; one group of the two groups of relay signals is a signal entering the ground induction coil, and the other group of the two groups of relay signals is a signal leaving the ground induction coil; the determination of the running speed and the passing state of the vehicle is achieved according to the interval between the start and the end of the signal.
In a further embodiment, the second module further comprises a signal receiving module, a signal processing module, and a signal determining module. The system comprises a signal receiving module, a signal receiving module and a signal processing module, wherein the signal receiving module is used for continuously receiving wireless signals of an ETC working frequency band in a specific direction and an ETC working peripheral frequency band when a vehicle passes through an ETC toll collection channel, and an interference signal detection receiver in the module selects a narrow beam direction antenna; the signal processing module is used for carrying out spectrum analysis on the wireless signals of the ETC working frequency band and the ETC working peripheral frequency band received by the signal receiving module and transmitting the processed result to the signal judging module; the signal judgment module is used for judging whether the signal is an abnormal signal or not according to a judgment rule; wherein, the rule of judging is when connecting interference signal, and ETC operating band has the peripheral electromagnetic signal noise of frequency channel and will obviously improve, and the normal ETC operating signal will be suppressed in the raising of ETC operating band noise.
In a further embodiment, the third module further comprises an interference signal receiving module, a network model for signal identification, and a signal extracting module; the interference signal receiving module is used for receiving the interference signal judged in the second module; the signal extraction module is used for extracting the interference signals received by the interference signal receiving module and presenting the interference signals in a data form similar to an image; the network model of the signal identification is used for realizing the identification of the interference signal type.
In a further embodiment, the fourth module further comprises an image acquisition module, a data storage module; the image acquisition module is triggered after the second module judges the signal, and a camera is used for shooting pictures of passing vehicles; the data storage module is used for storing the data in the third module and the image acquisition module and realizing the reading and storage of the data by each module by utilizing the data interaction of the server.
Has the advantages that: the invention relates to an ETC interference signal identification method and a detection system, which can reduce the requirement on training data by extracting the characteristics of an interference signal through a deep migration learning network, increase the generalization capability of a training model by constructing a loss function, avoid training a complex model with million parameters from a pre-trained model and further enhance the robustness.
Drawings
FIG. 1 is a block diagram of the method of the present invention.
Fig. 2 is a schematic diagram of a signal spectrum in normal ETC communication.
Fig. 3 is a schematic diagram of a signal spectrum when the ETC communication is disturbed.
Fig. 4 is a diagram of the interference detection and individual identification scenario of the present invention.
Fig. 5 is a diagram of a received signal scenario for a receiver of the present invention.
Fig. 6 is a flow chart of the detection receiver operation of the present invention.
Fig. 7 is a diagram of a source domain CNN model architecture.
Fig. 8 is an integrated diagram of an incrementally fine tuned CNN model.
Fig. 9 is a diagram of a deep migration learning network migration method.
Detailed Description
In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the present invention. It will be apparent, however, to one skilled in the art, that the present invention may be practiced without one or more of these specific details. In other instances, well-known features have not been described in order to avoid obscuring the invention. The technical solution of the present invention is further illustrated by the following examples.
In order to solve the problems in the prior art, the invention provides an identification method and a detection system of an ETC interference signal, which are used for processing and judging the collected ETC signal and tracking the obtained abnormal signal so as to reduce the occurrence of the obstruction of the subsequent pursuit payment work progress when the charging fails. The ETC interference signal identification method is characterized by comprising the following steps of:
step one, initializing parameters of a receiver for detecting signals; this step further initializes the means for detecting the receiver with data for hardware parameter configuration and establishes communication with the individual identification server.
Step two, acquiring a signal; as shown in fig. 5, in this step, when a vehicle passes over the upper surface of the coil, the iron of the vehicle changes the magnetic flux in the ground induction coil, so as to cause the change of the inductance in the coil loop, and then the detector detects the change of the electric quantity of the oscillation frequency in the coupling circuit formed by the loop coil, so as to determine the passing state of the vehicle, and simultaneously output two corresponding sets of logic signals, one set of signals enters the ground induction coil, the other set of signals leaves the inductance coil, and finally the two sets of logic signals are received by the detection receiver device, and trigger the signal monitoring device to monitor the interaction signals of the vehicle passing through the ground induction coil.
Step three, identifying interference signals; as shown in fig. 4, the interference detection receiver is deployed on the portal frame in cooperation with the ETC system RSU, receives and detects the wireless transmission signal of the passing vehicle, and transmits the acquired signal to the individual identification server through the network after identifying that the passing vehicle transmits the interference signal outwards. Specifically, according to the interactive signals obtained in the second step, the interference signals are identified by combining with spectrum analysis. According to spectral analysis, when interference signals exist, the noise of the electromagnetic signals around the ETC working frequency band is obviously improved, and the normal ETC working signals are suppressed by the raising of the noise of the ETC working frequency band. As shown in fig. 2, when normal ETC communication is not interfered, the effective signal energy is higher in the working frequency band of the ETC, and the non-effective signal area is lower; as shown in fig. 3, after normal ETC communication is disturbed, the whole working frequency band of the ETC and peripheral signals are raised and the effective signals are submerged.
Step four, judging the type of the interference signal; inputting the interference signal image data extracted in the third step into the constructed signal identification network model to extract the signal characteristics, thereby judging the type of the received interference signal. The specific steps of the construction process of the signal identification network model further comprise:
4-1, constructing a model training data set; the data sets are image data of different interference signal types, such as image information presented in spectral analysis and time-frequency transformation.
Step 4-2, fine tuning the convolution network; the convolutional network is a trained convolutional neural network; the fine adjustment mode is realized by fixing parameters of the predefined modules and changing the structures of partial modules. The preferred embodiment is shown in fig. 9, which divides the convolutional network structure into five modules, wherein the parameters of the first two modules are fixed, and the network structure after the second module is trimmed.
4-3, determining a difference measurement method between the target field and the source field; the step is further a method for limiting the error value of the target domain within the range of the error of the source domain plus a difference value, so as to solve the problem that the target domain data has no notes. Further, by using the data in the source domain and the data in the target domain, after the function random projection, the supremum of the expected difference between the two is shown in the following expression:
Figure 839676DEST_PATH_IMAGE031
in the formula (I), the compound is shown in the specification,
Figure 88254DEST_PATH_IMAGE005
representing a function
Figure 952305DEST_PATH_IMAGE006
The space in which the device is located is provided with a plurality of grooves,
Figure 766677DEST_PATH_IMAGE007
to represent
Figure 57981DEST_PATH_IMAGE008
The distribution of (a) to (b) is,
Figure 590594DEST_PATH_IMAGE009
to represent
Figure 976445DEST_PATH_IMAGE010
The distribution of (a) to (b) is,
Figure 910903DEST_PATH_IMAGE011
representing source domain functions
Figure 638687DEST_PATH_IMAGE012
In the expectation that the position of the target is not changed,
Figure 330700DEST_PATH_IMAGE013
representing a target domain function
Figure 598870DEST_PATH_IMAGE014
The expectation is that. When in use
Figure 591097DEST_PATH_IMAGE032
I.e. by
Figure 207892DEST_PATH_IMAGE008
And
Figure 715097DEST_PATH_IMAGE033
are equally distributed, regardless of function
Figure 459062DEST_PATH_IMAGE006
How to project it, the expected values will all be the same, so the value of the above expression is equal to 0; when in use
Figure 305795DEST_PATH_IMAGE004
And a function space
Figure 172120DEST_PATH_IMAGE005
When it is sufficiently rich, the value of the above expression will not be 0, and the value will be represented by the function space
Figure 838724DEST_PATH_IMAGE005
In (1) is
Figure 448697DEST_PATH_IMAGE008
And
Figure 664784DEST_PATH_IMAGE033
the function that maximizes the difference in the distribution of the projections. But when the function space
Figure 374114DEST_PATH_IMAGE005
If the function space is too rich, the probability of value taking is too high, which leads to the infinite expression, thereby further restricting the function space. When in use
Figure 855911DEST_PATH_IMAGE005
To reproduce a unit sphere in nuclear hilbert space, i.e., a space whose distance from the origin is less than or equal to 1, is preferable, the above expression is further as follows:
Figure 941679DEST_PATH_IMAGE034
further comprises the following steps:
Figure 559742DEST_PATH_IMAGE035
further comprises the following steps:
Figure 705552DEST_PATH_IMAGE036
in the formula
Figure 596017DEST_PATH_IMAGE004
，
Figure 547792DEST_PATH_IMAGE005
Representing a function
Figure 958045DEST_PATH_IMAGE006
The space in which the device is located is provided with a plurality of grooves,
Figure 337074DEST_PATH_IMAGE007
to represent
Figure 731146DEST_PATH_IMAGE008
The distribution of (a) to (b) is,
Figure 158716DEST_PATH_IMAGE009
to represent
Figure 751372DEST_PATH_IMAGE010
The distribution of (a) to (b) is,
Figure 488252DEST_PATH_IMAGE011
representing source domain functions
Figure 900779DEST_PATH_IMAGE012
In the expectation that the position of the target is not changed,
Figure 132040DEST_PATH_IMAGE013
representing a target domain function
Figure 782464DEST_PATH_IMAGE014
The expectation is that. When a single kernel is used, either a gaussian kernel or a linear kernel can be chosen, but this still has the disadvantage that it is not known which kernel is superior, so the feature mapping is further performed and the kernels are weighted using m different gaussian kernels, i.e.:
Figure 503296DEST_PATH_IMAGE037
in the formula (I), the compound is shown in the specification,
Figure 606381DEST_PATH_IMAGE016
representing the weight, for the coefficient
Figure 438071DEST_PATH_IMAGE016
The resulting k is guaranteed to be unique. In a general convolutional neural network, the characteristics are changed into special characteristics from the characteristics before the last layer, and the conversion is increased along with the increase of the network layer number to the full connection layer. Namely, the deep network is specialized by specific tasks in the full-connection layer, and when the target domain does not have enough data to perform supervised learning, the full-connection layer cannot directly perform training on the target domain. Because the source domain is labeled, the whole network is trained on the source domain, the accuracy of the network on the source domain is improved, and meanwhile, the output of the data on the source domain on a full connection layer and the data on the target domain can be output by using the expression calculation methodThe difference of the output on the full connection layer is as small as possible, so that the distribution difference of the data of the source domain and the data of the target domain after feature mapping is reduced, and the purpose of transfer learning is achieved.
4-4, constructing a loss function for weight adjustment; the loss function constructed in this step further comprises two-part loss; part of the loss is the loss of the traditional convolutional neural network, namely the loss of the convolutional neural network classified on a source domain, and the construction of a loss function adopts cross entropy; the other part of loss is the MK-MMD distance of the source domain at the output domain of the full connection layer and the target domain at the output domain of the full connection layer, and the construction of a loss function is shown in the following expression:
Figure 926690DEST_PATH_IMAGE038
in the formula (I), the compound is shown in the specification,
Figure 818422DEST_PATH_IMAGE018
a penalty coefficient is expressed for controlling the distribution difference of the source domain and the target domain in the full connection layer, and
Figure 408804DEST_PATH_IMAGE018
satisfy the requirement of
Figure 981868DEST_PATH_IMAGE019
；
Figure 403622DEST_PATH_IMAGE020
Indicating a starting layer for performing domain adaptation;
Figure 403939DEST_PATH_IMAGE021
indicating an end layer for performing domain adaptation;
Figure 730884DEST_PATH_IMAGE022
representing the number of the data with notes in the source domain and the target domain;
Figure 904376DEST_PATH_IMAGE023
parameters representing the entire neural network；
Figure 118320DEST_PATH_IMAGE024
Representing the output of the convolutional neural network;
Figure 351855DEST_PATH_IMAGE025
which represents the input image, is,
Figure 651249DEST_PATH_IMAGE026
to represent
Figure 628433DEST_PATH_IMAGE025
A corresponding label;
Figure 228041DEST_PATH_IMAGE027
indicating source domain data in
Figure 819428DEST_PATH_IMAGE028
Outputting the layer;
Figure 402856DEST_PATH_IMAGE029
indicating that the target data field is in
Figure 121414DEST_PATH_IMAGE028
Outputting the layer;
Figure 372266DEST_PATH_IMAGE030
indicating source domain and target domain data in
Figure 150867DEST_PATH_IMAGE028
The MK-MMD distance of the layer output squared.
And 4-5, receiving a training data set, and training the learning ability of the model for extracting the signal characteristics.
And step five, storing vehicle related information with interference signals, further transmitting the signals which are judged to be the interference signals in the step three to a signal identification network model constructed in the step four, carrying out identification analysis on signal characteristics, simultaneously matching the extracted individual characteristics with all stored data according to the existing interference signal information in an individual identification server, thereby judging whether the signals are the same interference transmitter, further transmitting the processed data characteristics to the individual identification server, and storing the data, thereby being used as a basis for crime tracking and discrimination.
Based on the above method, there is further provided a detection system for an ETC interference signal transmitter, configured to implement the foregoing solution, including:
a first module for detecting a vehicle passing condition; the module further includes a ground coil module and a detector module. And by analyzing the change of the inductance in the ground inductance module, the running state and the running speed of the vehicle are judged by using the detector module. The ground induction coil module is further used for analyzing a change numerical value of inductance in the ground induction coil when the vehicle passes through the ground induction coil, wherein the change numerical value is changed by iron of the vehicle. The detector module further comprises a data analysis and judgment module for analyzing the data according to the collected inductance data and a logic signal output module for sending out two groups of relay signals according to the data analyzed by the data judgment module. One group of the two groups of relay signals is a signal entering the ground induction coil, and the other group of the two groups of relay signals is a signal leaving the ground induction coil; the determination of the running speed and the passing state of the vehicle is achieved based on the interval between the start and the end of the signal.
A second module for determining an interference signal; the module further comprises a signal receiving module, a signal processing module and a signal judging module. As shown in fig. 5, the signal receiving module is configured to continuously receive wireless signals in an ETC operating frequency band and an ETC operating peripheral frequency band in a specific direction when a vehicle passes through an ETC toll collection channel, wherein an interference signal detecting receiver included in the module selects a narrow-beam directional antenna to accurately identify the vehicle triggering the interference signal analysis; the signal processing module is used for carrying out spectrum analysis on the wireless signals of the ETC working frequency band and the ETC working peripheral frequency band received by the signal receiving module and transmitting the processed result to the signal judging module; the signal judgment module is used for judging whether the signal is an abnormal signal or not according to a judgment rule; wherein, the rule of judging is when connecting interference signal, and ETC operating band has the peripheral electromagnetic signal noise of frequency channel and will obviously improve, and the normal ETC operating signal will be suppressed in the raising of ETC operating band noise.
A third module for identifying a characteristic of the interfering signal; the module comprises an interference signal receiving module, a network model for signal identification and a signal extraction module; the interference signal receiving module is used for receiving the interference signal judged in the second module; the signal extraction module is used for extracting the interference signals received by the interference signal receiving module and presenting the interference signals in a data form similar to an image; the network model of the signal identification is used for receiving the interference signal acquired by the signal extraction module, extracting the characteristics of the interference signal, comparing the extracted characteristics with the characteristics of the existing interference signal, judging the type of the interference signal, and finally outputting the characteristic information to the fourth module for storage. The network model of number identification further comprises a fine tuning module, a domain adaptation module and a loss function module.
And the fine tuning module further deletes and adds a predefined module structure to the source field convolution network model. The preferred embodiment is to adjust the convolutional neural network by using an incremental fine tuning method, for example, fig. 7 is an architecture diagram of a source domain convolutional neural network model, which includes five convolutional layers, three fully-connected layers and one Softmax layer. When an input tensor x is given, the CNN model of the linear structure is used for prediction, which is to perform layer-by-layer operation from left to right, and finally, the posterior probability prediction output of the multi-class labels is performed through a Softmax layer. For the CNN-S, CNN-M and CNN-F models in the source domain, although they are trained with the same CNN architecture, the computational unit settings of each layer are different, including: the number of convolution filters, the size of the receptive field, the convolution step size, the size of the spatial pooling, and the like. Therefore, the parameter values learned on the same training data set by using the CNN models of the three source domains are greatly different from each other, and the number of object types is different from that of the source domains for the recognition task of the target domain. The training data set is an Imagenet data set, and CNN-S, CNN-M and CNN-F models in the source field are obtained through training of the data set.
The domain adaptation module is used for solving the problem that the target domain data has no notes by using a method of supremum boundary of expected difference values of data on a source domain and data on a target domain after function random projection.
The loss function module comprises a traditional convolutional neural network loss module and a distance module; the conventional convolutional neural network loss module is used for further classifying the loss of the convolutional neural network on a source domain, and cross entropy is used as a loss function; the distance module further adopts the MK-MMD distance output by the source domain at the output domain of the full connection layer and the target domain at the full connection layer as a loss function.
In order to enable the model trained in the source field to be used for tasks of the target field, a method for incrementally fine-tuning the CNN is adopted to construct a CNN model factory of the target field. And further, based on the source domain CNN architecture, FC7 and FC8 layers are removed, a domain adaptation module is added, and the added domain adaptation module comprises two New full communication layers New-FC7 and New-FC8, wherein the input of the New-FC7 comes from an FC6 layer in the source domain model. As shown in FIG. 8, migrated parameters refer to copying these parameters obtained from pre-training in the source domain directly into the target domain model. And only updating the weight of the fine-tuned parameter layer when the target domain CNN model is trained. Since the last layer of the target domain CNN architecture is the Softmax layer, the predicted output of CNN is the posterior probability value over the target domain class. In order to further improve the diversity of the trained CNN model, the fourth dimension of the convolution filter in New-FC7 and the third dimension of the convolution filter in New-FC8 are preferably set to be 2048 pixels, 1024 pixels and 512 pixels respectively, and each setting mode corresponds to one type of domain adaptation module. This section uses
Figure 159274DEST_PATH_IMAGE039
To represent n convolution filters, each convolution filter having a size of
Figure 478260DEST_PATH_IMAGE040
。
A fourth module for storing data; the module comprises an image acquisition module and a data storage module. The image acquisition module is triggered after the second module judges the signal, and when the received signal is judged to be the interference signal, the camera is called to take a picture of the passing vehicle, and the picture information is transmitted to the server through the communication network to be stored so as to provide a basis for follow-up accountability. The data storage module is used for storing data in the third module and the image acquisition module and realizing the functional requirements of the modules on data reading and storing by utilizing the data interaction of the server.
As noted above, while the present invention has been shown and described with reference to certain preferred embodiments, it is not to be construed as limited thereto. Various changes in form and detail may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (8)

1. An ETC interference signal identification method is characterized by comprising the following steps:
initializing hardware parameters of a detection signal receiver, and establishing communication connection between the detection signal receiver and an individual identification server;
step two, acquiring a signal;
step three, receiving the signals obtained in the step two, performing spectrum analysis on the received signals, and identifying whether the signals are abnormal according to spectrum change;
step four, inputting the interference signal image data analyzed in the step three into the constructed signal identification network model to extract the signal characteristics, thereby judging the type of the received interference signal;
after the signal recognition network model finely adjusts the predefined module, the data set is trained again, so that the learning capacity of extracting the signal characteristics is trained;
step five, storing the vehicle-related information with the interference signal;
wherein, the fourth step further comprises the construction of a signal identification network model, and the specific steps of the construction process further comprise:
4-1, constructing a model training data set;
step 4-2, fine tuning the convolution network;
4-3, determining a difference measurement method between the target domain and the source domain;
4-4, constructing a loss function for weight adjustment;
4-5, receiving a training data set, and training a learning capacity of a model for extracting signal characteristics;
step 4-2 is further to finely adjust the convolutional neural network after the convolutional network is trained; the fine adjustment mode is realized by fixing the parameters of the predefined module and changing the structure of part of the modules;
step 4-3 further uses data in the source domain and data in the target domain, and after the function random projection, the supremum of the expected difference between the two solves the problem that the target domain has no notes, and the implementation manner is shown in the following expression:
Figure DEST_PATH_IMAGE002
in the formula (I), the compound is shown in the specification,
Figure DEST_PATH_IMAGE004
representing a function
Figure DEST_PATH_IMAGE006
The space in which the device is located is provided with a plurality of grooves,
Figure DEST_PATH_IMAGE008
to represent
Figure DEST_PATH_IMAGE010
The distribution of (a) to (b) is,
Figure DEST_PATH_IMAGE012
to represent
Figure DEST_PATH_IMAGE014
The distribution of (a) to (b) is,
Figure DEST_PATH_IMAGE016
representing source domain functions
Figure DEST_PATH_IMAGE018
In the expectation that the position of the target is not changed,
Figure DEST_PATH_IMAGE020
representing a target domain function
Figure DEST_PATH_IMAGE022
(iii) a desire;
Figure DEST_PATH_IMAGE024
a weighting value representing a Gaussian kernel;
further comprises the following steps:
Figure DEST_PATH_IMAGE026
in the formula (I), the compound is shown in the specification,
Figure 642132DEST_PATH_IMAGE004
representing a function
Figure 762535DEST_PATH_IMAGE006
The space in which the device is located is provided with a plurality of grooves,
Figure 975560DEST_PATH_IMAGE008
to represent
Figure 636349DEST_PATH_IMAGE010
The distribution of (a) to (b) is,
Figure 388404DEST_PATH_IMAGE012
to represent
Figure 617391DEST_PATH_IMAGE014
The distribution of (a) to (b) is,
Figure DEST_PATH_IMAGE028
express expectation
Figure 353266DEST_PATH_IMAGE016
，
Figure DEST_PATH_IMAGE030
Express expectation
Figure 660489DEST_PATH_IMAGE020
，
Figure 594947DEST_PATH_IMAGE024
A weighting value representing a Gaussian kernel;
further comprises the following steps:
Figure DEST_PATH_IMAGE032
in the formula
Figure DEST_PATH_IMAGE034
，
Figure 371666DEST_PATH_IMAGE004
Representing a function
Figure 329258DEST_PATH_IMAGE006
The space in which the device is located is provided with a plurality of grooves,
Figure 597428DEST_PATH_IMAGE008
to represent
Figure 261759DEST_PATH_IMAGE010
The distribution of (a) to (b) is,
Figure 894865DEST_PATH_IMAGE012
to represent
Figure 402070DEST_PATH_IMAGE014
The distribution of (a) to (b) is,
Figure 378991DEST_PATH_IMAGE028
express expectation
Figure 960145DEST_PATH_IMAGE016
，
Figure 436257DEST_PATH_IMAGE030
Express expectation
Figure 696337DEST_PATH_IMAGE020
，
Figure 712835DEST_PATH_IMAGE024
The weighted value of the gaussian kernel is expressed, the feature mapping is further performed for the clearly optimal kernel, and m different gaussian kernels are used for weighting the kernel, that is:
Figure DEST_PATH_IMAGE036
in the formula (I), the compound is shown in the specification,
Figure DEST_PATH_IMAGE038
representing the weight, for the coefficient
Figure 191571DEST_PATH_IMAGE038
The resulting k is guaranteed to be unique;
said step 4-4 further comprising constructing a loss function comprising a two-part loss; wherein, part of the loss is the loss of the traditional convolutional neural network, and the cross entropy is adopted as a loss function; the other part of loss is the MK-MMD distance of the source domain at the output domain of the full connection layer and the target domain at the output of the full connection layer, and the loss function is shown in the following expression:
Figure DEST_PATH_IMAGE040
in the formula (I), the compound is shown in the specification,
Figure DEST_PATH_IMAGE042
a penalty coefficient is expressed for controlling the distribution difference of the source domain and the target domain in the full connection layer, and
Figure 868277DEST_PATH_IMAGE042
satisfy the requirement of
Figure DEST_PATH_IMAGE044
；
Figure DEST_PATH_IMAGE046
Indicating a starting layer for performing domain adaptation;
Figure DEST_PATH_IMAGE048
indicating an end layer for performing domain adaptation;
Figure DEST_PATH_IMAGE050
representing the number of the data with notes in the source domain and the target domain;
Figure DEST_PATH_IMAGE052
parameters representing the entire neural network;
Figure DEST_PATH_IMAGE054
representing the output of the convolutional neural network;
Figure DEST_PATH_IMAGE056
which represents the input image, is,
Figure DEST_PATH_IMAGE058
to represent
Figure 680900DEST_PATH_IMAGE056
A corresponding label;
Figure DEST_PATH_IMAGE060
indicating source domain data in
Figure DEST_PATH_IMAGE062
Outputting the layer;
Figure DEST_PATH_IMAGE064
indicating that the target data field is in
Figure 140569DEST_PATH_IMAGE062
Outputting the layer;
Figure DEST_PATH_IMAGE066
indicating source domain and target domain data in
Figure 332866DEST_PATH_IMAGE062
The MK-MMD distance of the layer output squared.
2. The method according to claim 1, wherein the second step is further:
judging the passing state of the vehicle by using a vehicle detector, and triggering a receiver to read a corresponding signal when the judged state is that the passing condition of the vehicle exists; the vehicle detector detects the change of electric quantity of oscillation frequency in a coupling circuit formed by the annular coil by using a detector; the electric quantity change is generated by the magnetic flux change in the ground induction coil when the vehicle reaches the ETC inlet and passes through the upper surface of the ground induction coil; simultaneously the detector outputs two corresponding groups of logic signals; the two groups of logic signals comprise signals entering the ground induction coil and signals leaving the ground induction coil; the signal leaving the ground induction coil is used for triggering a receiver to perform corresponding signal reading.
3. The method according to claim 1, wherein the third step is further:
according to the interactive signal obtained in the second step, the judgment basis in the process of spectrum analysis is combined, when an interference signal exists, the noise of electromagnetic signals at the working frequency band of the ETC and the periphery of the frequency band is obviously improved, and the normal ETC working signal is suppressed due to the rise of the noise of the working frequency band of the ETC; when normal ETC communication is not interfered, the effective signal energy is higher in an ETC working frequency band, and the non-effective signal area is lower; after normal ETC communication receives the interference, whole ETC working frequency range and peripheral signal will all be raised to submerge the effective signal.
4. The method according to claim 1, wherein the fifth step is further:
and transmitting the signal which is judged as the interference signal in the third step into a signal identification network model constructed in the fourth step, identifying and analyzing the signal characteristics, performing characteristic matching according to the existing interference signal information in the individual identification server so as to judge whether the signal is the same interference transmitter, and further transmitting the processed data characteristics to the individual identification server for data storage.
5. A detection system of ETC interference signals, which is used for realizing the method of any one of the claims 1-4, and is characterized by comprising:
a first module for detecting a vehicle passing condition;
a second module for determining an interference signal;
a third module for identifying an interfering signal type;
a fourth module for storing data.
6. The system according to claim 5, wherein the first module further comprises a ground induction coil module and a detector module; the ground induction coil module is used for data acquisition, and the detector module further comprises a data judgment module and a logic signal output module; the data in the data acquisition is further a change value of inductance in a ground induction coil loop caused by changing magnetic flux in the ground induction coil when a vehicle passes through the ground induction coil; the data judgment module analyzes the inductance data according to the data acquisition; the logic signal output module sends two groups of relay signals according to the data analyzed by the data judgment module; one group of the two groups of relay signals is a signal entering the ground induction coil, and the other group of the two groups of relay signals is a signal leaving the ground induction coil; realizing the determination of the running speed and the passing state of the vehicle according to the interval between the start and the end of the signal;
a second module for determining an interference signal; the module further comprises a signal receiving module, a signal processing module and a signal judging module; the system comprises a signal receiving module, a signal receiving module and a signal processing module, wherein the signal receiving module is used for continuously receiving wireless signals of an ETC working frequency band in a specific direction and an ETC working peripheral frequency band when a vehicle passes through an ETC toll collection channel, and an interference signal detection receiver in the module selects a narrow beam direction antenna; the signal processing module is used for carrying out spectrum analysis on the wireless signals of the ETC working frequency band and the ETC working peripheral frequency band received by the signal receiving module and transmitting the processed result to the signal judging module; the signal judgment module is used for judging whether the signal is an abnormal signal or not according to a judgment rule; wherein, the rule of judging is when connecting interference signal, and ETC operating band has the peripheral electromagnetic signal noise of frequency channel and will obviously improve, and the normal ETC operating signal will be suppressed in the raising of ETC operating band noise.
7. The ETC interference signal detection system according to claim 5, wherein the third module further comprises an interference signal receiving module, a network model for signal identification, and a signal extraction module; the interference signal receiving module is used for receiving the interference signal judged in the second module; the signal extraction module is used for extracting the interference signals received by the interference signal receiving module and presenting the interference signals in the form of image data; the network model of the signal identification is used for realizing the identification of the interference signal type.
8. The ETC interference signal detection system according to claim 5, wherein the fourth module further comprises an image acquisition module, a data storage module; the image acquisition module is triggered after the second module judges the signal, and a camera is used for shooting pictures of passing vehicles; the data storage module is used for storing the data in the third module and the image acquisition module and realizing the reading and storage of the data by each module by utilizing the data interaction of the server.
CN202011213489.2A 2020-11-04 2020-11-04 ETC interference signal identification method and detection system Active CN112435356B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011213489.2A CN112435356B (en) 2020-11-04 2020-11-04 ETC interference signal identification method and detection system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011213489.2A CN112435356B (en) 2020-11-04 2020-11-04 ETC interference signal identification method and detection system

Publications (2)

Publication Number Publication Date
CN112435356A CN112435356A (en) 2021-03-02
CN112435356B true CN112435356B (en) 2021-11-26

Family

ID=74695313

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011213489.2A Active CN112435356B (en) 2020-11-04 2020-11-04 ETC interference signal identification method and detection system

Country Status (1)

Country Link
CN (1) CN112435356B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023214906A1 (en) * 2022-05-06 2023-11-09 Telefonaktiebolaget Lm Ericsson (Publ) Network node and methods for robust kernel-based interference detection in a wireless communications network
CN115642968A (en) * 2022-12-22 2023-01-24 中南大学 Communication interference signal and spectrum cavity joint cognition method, device and medium based on deep learning
CN117272022B (en) * 2023-09-19 2024-07-05 北京中关村集成电路设计园发展有限责任公司 Detection method of MEMS oscillator

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107203782A (en) * 2017-05-23 2017-09-26 哈尔滨工业大学 Communication interference signals recognition methods under Larger Dynamic signal to noise ratio based on convolutional neural networks
CN110197127B (en) * 2019-05-06 2022-10-18 安徽继远软件有限公司 Wireless signal detection and electromagnetic interference classification system and method based on deep learning
CN110288839B (en) * 2019-07-30 2022-05-17 成都安杰联科技有限公司 Detection and positioning device and method for ETC (electronic toll Collection) interference unit
CN110445556B (en) * 2019-07-30 2022-03-22 成都安杰联科技有限公司 Device and method for detecting and positioning ETC (electronic toll Collection) interference unit
CN110764064A (en) * 2019-11-08 2020-02-07 哈尔滨工业大学 Radar interference signal identification method based on deep convolutional neural network integration
CN110927706B (en) * 2019-12-10 2022-05-24 电子科技大学 Convolutional neural network-based radar interference detection and identification method
CN110996343B (en) * 2019-12-18 2023-06-30 中国人民解放军陆军工程大学 Intelligent recognition system and recognition method of interference recognition model based on deep convolutional neural network
CN111783558A (en) * 2020-06-11 2020-10-16 上海交通大学 Satellite navigation interference signal type intelligent identification method and system

Also Published As

Publication number Publication date
CN112435356A (en) 2021-03-02

Similar Documents

Publication Publication Date Title
CN112435356B (en) ETC interference signal identification method and detection system
Won Intelligent traffic monitoring systems for vehicle classification: A survey
CN107862705B (en) Unmanned aerial vehicle small target detection method based on motion characteristics and deep learning characteristics
CN111310583B (en) Vehicle abnormal behavior identification method based on improved long-term and short-term memory network
WO2019241224A1 (en) Architectures for vehicle tolling
KR20180036709A (en) Media classification
CN112733885A (en) Point cloud identification model determining method and point cloud identification method and device
CN111597901A (en) Illegal billboard monitoring method
CN114926825A (en) Vehicle driving behavior detection method based on space-time feature fusion
CN115546608A (en) Unmanned aerial vehicle data link electromagnetic interference classification and threat assessment method
CN114417942B (en) Clutter recognition method, system, device and medium
CN101320520A (en) Vehicle detection method and equipment thereof
CN111597902A (en) Motor vehicle illegal parking monitoring method
Sharma et al. Iot in smart cities: Exploring information theoretic and deep learning models to improve parking solutions
CN110263836A (en) A kind of bad steering state identification method based on multiple features convolutional neural networks
Meduri et al. A haar-cascade classifier based smart parking system
CN116894223A (en) Airborne underwater anomaly detection method based on self-adaptive cancellation and ResNet neural network
CN111597897B (en) High-speed service area parking space recognition method
CN115393743A (en) Vehicle detection method based on double-branch encoding and decoding network, unmanned aerial vehicle and medium
CN114972725A (en) Model training method, readable medium and electronic device
Celso et al. Two-step vehicle classification system for traffic monitoring in the philippines
CN114119382A (en) Image raindrop removing method based on attention generation countermeasure network
Kolli et al. Robust localization of key fob using channel impulse response of ultra wide band sensors for keyless entry systems
Zhao et al. Application of data fusion based on clustering-neural network for ETC gantry flow capacity correction
CN117332299A (en) Millimeter wave Lei Dadian cloud statistical feature-based target classification 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