CN112232329A - Multi-core SVM training and alarming method, device and system for intrusion signal recognition - Google Patents

Multi-core SVM training and alarming method, device and system for intrusion signal recognition Download PDF

Info

Publication number
CN112232329A
CN112232329A CN202011490179.5A CN202011490179A CN112232329A CN 112232329 A CN112232329 A CN 112232329A CN 202011490179 A CN202011490179 A CN 202011490179A CN 112232329 A CN112232329 A CN 112232329A
Authority
CN
China
Prior art keywords
kernel
core
intrusion signal
determining
intrusion
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
CN202011490179.5A
Other languages
Chinese (zh)
Inventor
涂家勇
刘卫华
李源
郭远方
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.)
Optical Valley Technology Co ltd
Original Assignee
Optical Valley 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 Optical Valley Technology Co ltd filed Critical Optical Valley Technology Co ltd
Priority to CN202011490179.5A priority Critical patent/CN112232329A/en
Publication of CN112232329A publication Critical patent/CN112232329A/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
    • G06F18/22—Matching criteria, e.g. proximity measures
    • 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
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G08—SIGNALLING
    • G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
    • G08B13/00—Burglar, theft or intruder alarms
    • G08B13/02—Mechanical actuation
    • G08B13/12—Mechanical actuation by the breaking or disturbance of stretched cords or wires
    • G08B13/122—Mechanical actuation by the breaking or disturbance of stretched cords or wires for a perimeter fence
    • G08B13/124—Mechanical actuation by the breaking or disturbance of stretched cords or wires for a perimeter fence with the breaking or disturbance being optically detected, e.g. optical fibers in the perimeter fence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Signal Processing (AREA)
  • Burglar Alarm Systems (AREA)

Abstract

The invention provides a multi-core SVM training and alarming method, device and system for intrusion signal recognition, which relate to the technical field of intelligent security and comprise the following steps: acquiring an intrusion signal data set and carrying out normalization processing; selecting a plurality of basic kernel functions, determining the similarity between corresponding kernel matrixes of the plurality of basic kernel functions according to the plurality of basic kernel functions and the intrusion signal data set, and determining the kernel weight of each basic kernel function according to the similarity; determining a multi-core function according to the core weight; and determining sample membership according to a fuzzy rough set method, performing multi-core SVM training and optimization according to the sample membership and a multi-core function, determining optimized core weight and Lagrangian multiplier of the optimal solution, and finishing the training of the multi-core SVM. According to the method, the kernel weight is determined through kernel alignment to construct a multi-kernel function, and multi-kernel SVM training and optimization are performed by combining with the calculation of the membership degree of the sample, so that the classification interval is maximized, and the identification of different types of intrusion signals is more accurate.

Description

Multi-core SVM training and alarming method, device and system for intrusion signal recognition
Technical Field
The invention relates to the technical field of intelligent security and protection, in particular to a multi-core SVM training and alarming method, device and system for intrusion signal recognition.
Background
With the continuous development of economy and science and technology, the number of places requiring high-quality security work is also increasing. In order to better protect the security of the country and personal property, the security system is required to be covered fully and alarm accurately as much as possible. In recent years, a series of high and new technologies such as internet of things and cloud processing have been developed rapidly, so that information resources can be shared in a networked manner. Meanwhile, in order to meet the security requirements of places with large area and many places, such as various malls, financial centers and the like, the security alarm system must have the characteristics of remote supervision and control, networking and comprehensive management of alarm information, stability, reliability, difficulty in damage by intruders, low system cost and the like.
The electronic fence is widely applied to the field of public security as a perimeter security defense alarm system, can also serve as an important helper of community public security business, saves a large amount of community public security police strength along with continuous optimization and perfection of the alarm function of the system, and obviously improves the public security. The existing electronic fence system is composed of a front-end physical protection fence and a system host, the host can generate and receive pulse detection signals, when the front-end detection fence is invaded, abnormal states such as cable short circuit and open circuit can be caused, invasion information is sent to the system host, and alarm signals are generated after processing, so that police officers can master the condition of an alarm area in time and make a quick response, and the efficiency of processing the alarm condition is improved.
The identification of pulse echo signals in the intelligent electronic fence system is a core part for embodying humanization and intellectualization of the intelligent electronic fence system, and classification and alarm are carried out on characteristic information according to different intrusion behaviors based on the extraction of signal characteristics. However, the pulse echo signal received by the system is a non-stationary signal, and noise interference of various factors is large, so that difficulty in identifying characteristic information of the intrusion signal is increased, and alarm accuracy is reduced.
Disclosure of Invention
The present invention solves the technical problem in the prior art, and in order to achieve the above object, in a first aspect, the present invention provides a multi-core SVM training method for intrusion signal recognition, including:
acquiring an intrusion signal data set and carrying out normalization processing, wherein the intrusion signal data set is expressed as
Figure 454999DEST_PATH_IMAGE001
Wherein, in the step (A),xandyrespectively representing the random input and output results,lrepresenting the number of samples corresponding to the data, and marking noise, wave crests and/or frequency in the intrusion signal data set according to the characteristics of the intrusion signals;
selecting a plurality of basic kernel functions, determining the similarity between corresponding kernel matrixes of the basic kernel functions according to the basic kernel functions and the intrusion signal data set, and determining the kernel weight of each basic kernel function according to the similarity;
determining a multi-kernel function according to the kernel weight;
determining sample membership according to a fuzzy rough set method, performing multi-core SVM training and optimization according to the sample membership and the multi-core function, determining optimized core weight and Lagrange multiplier of an optimal solution, and finishing the training of the multi-core SVM.
Further, the determining the similarity between the corresponding kernel matrices of the plurality of base kernel functions according to the selected plurality of base kernel functions and the intrusion signal data set comprises:
determining similarity between corresponding kernel matrices of a plurality of said base kernel functions
Figure 569585DEST_PATH_IMAGE002
Wherein, in the step (A),Trepresenting the intrusion signal data in a data stream,
Figure 350460DEST_PATH_IMAGE003
and
Figure 883203DEST_PATH_IMAGE004
respectively representing basic kernel functions
Figure 724120DEST_PATH_IMAGE005
And
Figure 326003DEST_PATH_IMAGE006
the kernel matrix after the mapping is performed,
Figure 910568DEST_PATH_IMAGE007
，
Figure 530774DEST_PATH_IMAGE008
separately representing intrusion signal datasetsTThe sample of (1).
Further, the determining the kernel weight for each of the base kernel functions according to the similarity comprises:
determining a maximized similarity of a combined kernel of a plurality of the base kernel functions to an ideal kernel;
and optimizing according to the similarity and the maximized similarity, and determining the kernel weight of each basic kernel function.
Further, the determining the sample membership according to the fuzzy rough set method includes:
determining the sample membership of an intrusion signal using an approximation operator under a gaussian kernel based fuzzy rough set:
Figure 542592DEST_PATH_IMAGE009
，
wherein the content of the first and second substances,
Figure 631771DEST_PATH_IMAGE010
representation of belonging to a sample set
Figure 770760DEST_PATH_IMAGE011
The degree of membership of (a) is,
Figure 996205DEST_PATH_IMAGE012
，
Figure 178924DEST_PATH_IMAGE013
to represent
Figure 755399DEST_PATH_IMAGE014
Corresponding to the blur value in the blur rough set.
Further, the training and optimizing the multi-core SVM according to the sample membership and the multi-core function, and determining the optimized kernel weight and the lagrangian multiplier of the optimal solution includes:
the optimization problem of the multi-core SVM is determined as follows:
Figure 190754DEST_PATH_IMAGE015
，
wherein the content of the first and second substances,
Figure 270706DEST_PATH_IMAGE016
representing the direction of the classification hyperplane in the high-dimensional feature space,ηthe weight of the kernel is represented by,
Figure 889906DEST_PATH_IMAGE017
the value of the relaxation variable is represented by,
Figure 688098DEST_PATH_IMAGE018
，Ca penalty factor is represented which is a function of,wa generalized set of parameters representing a decision function,bthe amount of offset is indicated by an indication,y i representing a single random output in a fuzzy rough set;
introducing lagrange multipliers
Figure 903309DEST_PATH_IMAGE019
Determining the corresponding Lagrangian function as:
Figure 103347DEST_PATH_IMAGE020
，
wherein the content of the first and second substances,
Figure 159027DEST_PATH_IMAGE021
representing a non-linear mapping that maps the original space to a high-dimensional feature space;
according to the similarity between the combined kernel maximizing the plurality of basic kernel functions and the ideal kernel, determining an optimization function as follows by a Lagrange multiplier method:
Figure 428203DEST_PATH_IMAGE022
wherein the content of the first and second substances,
Figure 227532DEST_PATH_IMAGE023
with penalty factorCIs changed;
and solving the optimization function, and determining the optimized kernel weight and the Lagrangian multiplier of the optimal solution.
In order to achieve the above object, in a second aspect, the present invention provides a multi-core SVM training apparatus for intrusion signal recognition, including:
an acquisition module for acquiring and normalizing an intrusion signal data set expressed as
Figure 360704DEST_PATH_IMAGE024
Wherein, in the step (A),xandyrespectively representing the random input and output results,lrepresenting the number of samples corresponding to the data, and marking noise, wave crests and/or frequency in the intrusion signal data set according to the characteristics of the intrusion signals;
a processing module, configured to select a plurality of basic kernel functions, determine, according to the plurality of basic kernel functions and the intrusion signal data set, similarities between kernel matrices corresponding to the plurality of basic kernel functions, and determine the kernel weight of each basic kernel function according to the similarities; the multi-core function is determined according to the core weight;
and the training module is used for determining sample membership according to a fuzzy rough set method, training and optimizing the multi-core SVM according to the sample membership and the multi-core function, determining optimized core weight and a Lagrangian multiplier of an optimal solution, and finishing the training of the multi-core SVM.
By using the multi-core SVM training method or device for identifying the intrusion signals, disclosed by the invention, the multi-core SVM is trained and optimized by selecting a plurality of basic kernel functions, determining the kernel weight of each basic kernel function based on kernel alignment, constructing the multi-core function according to the kernel weights and combining with the sample membership degree of a calculated noise rough set method to obtain the optimized kernel weights and Lagrange multipliers of optimal solutions, so that the classification interval of the trained multi-core SVM is maximized, and the identification of different types of intrusion signals is more accurate.
In order to achieve the above object, in a third aspect, the present invention provides an alarm method for intrusion signal identification, including:
acquiring an original intrusion signal, performing positive and negative complementary empirical mode decomposition, and acquiring a plurality of IMF components to be input into a multi-core SVM model for recognition, wherein the multi-core SVM model is trained by adopting the multi-core SVM training method for recognizing the intrusion signal;
and alarming according to the recognition result of the multi-core SVM model.
In order to achieve the above object, in a fourth aspect, the present invention provides an alarm device for intrusion signal identification, including:
a memory and a processor; the memory for storing a computer program; the processor is configured to implement the alarm method for intrusion signal identification as described above when executing the computer program.
By using the alarm method or the alarm device for identifying the intrusion signal, which is disclosed by the invention, the signal decomposition is carried out after positive and negative Gaussian white noise is added into the original signal, so that the residual noise in the decomposition process can be effectively neutralized, and the difference of the signal characteristics can be better analyzed. After singular value entropy is obtained, intrusion signal type recognition is carried out according to the trained multi-core SVM, and alarming is carried out according to the corresponding type, so that the accuracy rate of alarming can be effectively improved.
In order to achieve the above object, in a fifth aspect, the present invention provides an intrusion signal recognition alarm system, which includes a plurality of front-end detection electronic fences, an intrusion signal recognition alarm device as described above, and a user operation platform, wherein the plurality of front-end detection electronic fences are used to detect intrusion signals, the intrusion signal recognition alarm device is used to identify and alarm the intrusion signals, and the user operation platform is used to perform management and query.
The alarm system for identifying the intrusion signals is used for detecting echo signals by arranging a plurality of front-end detection fences to obtain possible intrusion signals, analyzing and identifying the signals according to the alarm device for identifying the intrusion signals based on the multi-core SVM to obtain different intrusion signal types, and then alarming according to different intrusion signals to realize efficient and accurate intrusion identification and ensure the safety of users.
To achieve the above object, in a sixth aspect, the present invention provides a non-transitory computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, implements the monitoring camera parameter calibration method according to the first aspect of the present invention or the alarm method for intrusion signal identification according to the third aspect of the present invention.
According to the non-transitory computer readable storage medium and the computing device of the present invention, the method for training the multi-core SVM according to the first aspect of the present invention or the method for alarming the intrusion signal recognition according to the third aspect of the present invention has similar beneficial effects, and will not be described herein again.
Drawings
FIG. 1 is a flowchart illustrating a multi-core SVM training method for intrusion signal recognition according to an embodiment of the present invention;
FIG. 2 is a flow chart illustrating the determination of kernel weights according to an embodiment of the present invention;
FIG. 3 is a schematic flow chart illustrating the determination of kernel weights based on kernel alignment according to an embodiment of the present invention;
FIG. 4 is a schematic flow diagram illustrating training and optimization of a multi-core SVM according to an embodiment of the present invention;
FIG. 5 is a schematic structural diagram of a multi-core SVM training device for intrusion signal recognition according to an embodiment of the present invention;
FIG. 6 is a flow chart illustrating an alarm method for intrusion signal identification according to an embodiment of the present invention;
FIG. 7 is a schematic structural diagram of an alarm device for intrusion signal identification according to an embodiment of the present invention;
fig. 8 is a schematic structural diagram of an alarm system for intrusion signal identification according to an embodiment of the present invention.
Detailed Description
Embodiments in accordance with the present invention will now be described in detail with reference to the drawings, wherein like reference numerals refer to the same or similar elements throughout the different views unless otherwise specified. It is to be noted that the embodiments described in the following exemplary embodiments do not represent all embodiments of the present invention. They are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the claims, and the scope of the present disclosure is not limited in these respects. Features of the various embodiments of the invention may be combined with each other without departing from the scope of the invention.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two, e.g., two, three, etc., unless specifically limited otherwise.
With the development of science and technology, security systems are widely applied to various industries of the society, and people can know things happening in related places in real time through the security systems. For individuals, the security system can ensure that the properties of the individuals are effectively monitored, and the property safety of the individuals is guaranteed. For related departments, the security system can be used as an important means for daily monitoring, and when an emergency occurs, the security system can be used for commanding and scheduling to evacuate and evacuate people. Therefore, the security system plays an important role in the stability of the society and the property safety of the public. While the security system is gradually popularized, people put higher requirements on the system, so that it is necessary to apply a related artificial intelligence algorithm to the security system to form a new intelligent security system.
The perimeter alarm system of the electronic fence is widely applied to the field of public security, and can effectively block intrusion behaviors, position intrusion positions and give an alarm. The current electronic fence protection mostly adopts intellectualization, and mainly has the following two modes: one is to set up physical fences to keep people out of the production area, and the other is to use more grating electronic fences. The grating electronic fence is an intelligent perimeter system formed by combining a pulse generator (host) and a physical fence, is one of active infrared correlation, namely, a plurality of beams of infrared light are used for correlation, an emitter emits infrared light to a receiver in a low-frequency emission and time division detection mode, once a person or an object blocks any two adjacent beams of infrared light emitted by the emitter for more than 30 ms, the receiver immediately outputs an alarm signal, and the grating electronic fence has high sensitivity. However, when the intrusion type is identified, the classification of the characteristic information is insufficient, and the false alarm rate is high.
In recent years, SVM (Support Vector Machine) is widely used in an electronic fence security system, but gradually shows some limitations. Firstly, the SVM is very sensitive to isolated point kernel noise data; secondly, the selection of kernel functions and kernel parameters has a crucial influence on learning performance, but no effective means for selecting kernel functions and kernel parameters exists at present. In recent years, multi-core learning has become a research focus in the field of machine learning, i.e., a combination of multiple kernel functions is used to replace a single kernel function. The multiple kernel functions can more fully describe the similarity among data (especially complex data), so that the similarity of the data can be more accurately expressed. Therefore, there is a need for further improvement in the identification and processing of the intrusion echo signals of the existing intelligent electronic fence.
According to the method, different kernel functions are selected based on the check based on the single-kernel-based function SVM, different parameters are designated, a learning model of the multi-kernel SVM is established for multiple invasion types to train and optimize, and climbing, touch, striking and other false alarm signals are identified. The multi-core SVM learning method is the extension of a single-core SVM and aims to determine a plurality of optimal kernel function combinations, so that the classification space reaches the maximum, and the accuracy of intrusion signal type identification is effectively improved.
Fig. 1 is a flowchart illustrating a multi-core SVM training method for intrusion signal recognition according to an embodiment of the present invention, including steps S1 to S4.
In step S1, an intrusion signal data set, which is expressed as intrusion signal data set, is acquired and normalized
Figure 321707DEST_PATH_IMAGE001
Wherein, in the step (A),xandyrespectively representing the random input and output results,land representing the number of samples corresponding to the data, and marking the noise, the wave crest and/or the frequency according to the characteristics of the intrusion signal in the intrusion signal data set. It is to be understood that the noise, the peak and/or the frequency are labeled in the intrusion signal data set according to the unique characteristics of the pulse signal (i.e., the intrusion signal), and the present invention is not limited thereto.
In the embodiment of the present invention, the acquired intrusion signal is normalized, for example, the pulse signal frequency is normalized to [0, 1] to be a dimensionless pure number. It can be understood that, in the embodiment of the present invention, the acquired intrusion signal data is averagely divided into a plurality of parts, one of the parts is used as a test set each time, and the rest is used as a training set, so as to perform subsequent training and optimization on the multi-core SVM.
In step S2, a plurality of basic kernel functions are selected, similarity between corresponding kernel matrices of the plurality of basic kernel functions is determined according to the plurality of basic kernel functions and the intrusion signal data set, and the kernel weight of each basic kernel function is determined according to the similarity. In the embodiment of the invention, a heuristic learning mode is adopted to determine the weight coefficient, and the multi-core weight coefficient is calculated in a core alignment mode, namely, the core weight of each core function is determined by calculating the similarity between the corresponding core matrixes of the core functions. Fig. 2 is a schematic flow chart illustrating the process of determining kernel weights according to the embodiment of the present invention, which includes steps S21 to S22.
In step S21, similarity between core matrices corresponding to a plurality of the base kernel functions is determined according to the selected plurality of the base kernel functions and the intrusion signal data set. In the embodiment of the present invention, the basic kernel function to be selected has many forms, and may be some commonly used kernel functions, such as a linear kernel function, a polynomial kernel function, a gaussian kernel function, a Sigmoid kernel function, and the like. It can be understood that a plurality of basic kernel functions can be selected according to actual application requirements to select kernel weights, and a multi-kernel function is formed to perform subsequent optimization, which is not limited in the present invention.
In the embodiment of the invention, the similarity between the core matrixes corresponding to the kernel functions is calculated according to each kernel function and each intrusion signal data set so as to determine the kernel weight. For intrusion signal data setsTDetermining a kernel function
Figure 94491DEST_PATH_IMAGE025
And
Figure 681199DEST_PATH_IMAGE026
mapped corresponding kernel matrix
Figure 855828DEST_PATH_IMAGE027
And
Figure 987733DEST_PATH_IMAGE028
the similarities between them are:
Figure 998545DEST_PATH_IMAGE002
，
wherein the content of the first and second substances,
Figure 874097DEST_PATH_IMAGE029
，
Figure 903233DEST_PATH_IMAGE030
separately representing intrusion signal datasetsTThe sample of (1).
In step S22, the kernel weight of each of the base kernel functions is determined according to the similarity. Fig. 3 is a schematic flowchart illustrating a process of determining kernel weights based on kernel alignment according to an embodiment of the present invention, which includes steps S221 to S222.
In step S221, a maximum similarity of a combined kernel of a plurality of the base kernel functions to an ideal kernel is determined. In the embodiment of the present invention, the multi-core combination formula based on the check-alignment weighted sum is:
Figure 455306DEST_PATH_IMAGE031
wherein the content of the first and second substances,
Figure 733841DEST_PATH_IMAGE032
the weight of the kernel is represented by,
Figure 147504DEST_PATH_IMAGE033
，
Figure 781879DEST_PATH_IMAGE034
a base kernel matrix representing a certain set of,
Figure 990007DEST_PATH_IMAGE035
to obtain the maximum similarity between the combined kernel and the ideal kernel, the following formula is adopted:
Figure 224679DEST_PATH_IMAGE036
，
Figure 697160DEST_PATH_IMAGE037
wherein the content of the first and second substances,
Figure 700889DEST_PATH_IMAGE038
is based on an intrusion signal data setTThe ideal kernel of the two-classification task of (1),
Figure 345497DEST_PATH_IMAGE039
the inner product is represented by the sum of the two,
Figure 801886DEST_PATH_IMAGE040
this means that the first row equation is satisfied and the following condition needs to be satisfied.
In step S222, optimization is performed according to the similarity and the maximized similarity, and the kernel weight of each of the basic kernel functions is determined. In the embodiment of the present invention, the denominator in the above formula is taken as a constant, so as to optimize the problem, that is, the above formula is equivalent to:
Figure 308084DEST_PATH_IMAGE041
wherein the content of the first and second substances,Ca penalty factor is indicated. It can be understood that, by solving the above optimization problem, the kernel weight corresponding to each basic kernel function can be obtained.
In step S3, a multi-core function is determined according to the core weights. In the embodiment of the invention, a linear combination mode, namely a mode of single-core addition and weighted linear combination is adopted to determine the multi-core function:
Figure 431898DEST_PATH_IMAGE042
wherein the content of the first and second substances,ηthe weight of the kernel is represented by,η i a simple single-core addition is achieved when both are 1,pthe number of kernel functions is indicated. It is understood that the multi-core function may be determined in different manners according to the actual application requirement, and the invention is not limited thereto.
In step S4, a sample membership is determined according to a fuzzy rough set method, multi-core SVM training and optimization are performed according to the sample membership and the multi-core function, an optimized core weight and a lagrangian multiplier of an optimal solution are determined, and training of the multi-core SVM is completed. Fig. 4 is a schematic flowchart illustrating the process of performing multi-core SVM training and optimization according to an embodiment of the present invention, including steps S41 to S45.
In step S41, the sample membership of the intrusion signal is determined using an approximation operator under a fuzzy rough set based on gaussian kernels. In the embodiment of the invention, the traditional support vector machine is very sensitive to samples, so that the application effect in a noise environment is poor. In order to overcome the defect of the traditional support vector machine, the embodiment of the invention assigns membership to each intrusion echo signal sample, wherein the important sample has larger membership and the secondary sample and the noise have smaller membership, thereby effectively reducing the influence of the noise on the intrusion echo signals and improving the robustness during the detection of the echo signal characteristics. In an embodiment of the invention, approximation operators under a gaussian kernel based Fuzzy Rough Set (FRS) are used as membership of the intrusion signal. Take k (x, x’) For a Gaussian kernel function, for sample points
Figure 247408DEST_PATH_IMAGE043
Let us order
Figure 174781DEST_PATH_IMAGE044
To belong to a sample set
Figure 733938DEST_PATH_IMAGE045
Degree of membership of
Figure 446680DEST_PATH_IMAGE044
The lower limit function is obtained:
Figure 698669DEST_PATH_IMAGE046
，
wherein the content of the first and second substances,
Figure 880383DEST_PATH_IMAGE047
representation of belonging to a sample set
Figure 977652DEST_PATH_IMAGE048
The degree of membership of (a) is,
Figure 544900DEST_PATH_IMAGE049
to represent
Figure 217058DEST_PATH_IMAGE050
Corresponding to the blur value in the blur rough set.
In step S42, the optimization problem of the multi-core SVM is determined as:
Figure 869757DEST_PATH_IMAGE051
wherein the content of the first and second substances,
Figure 770716DEST_PATH_IMAGE052
representing the direction of the classification hyperplane in the high-dimensional feature space,ηthe weight of the kernel is represented by,
Figure 458050DEST_PATH_IMAGE053
a slack variable is indicated to allow for the presence of misclassified samples,
Figure 802575DEST_PATH_IMAGE054
，Ca penalty factor is represented which is a function of,wa generalized set of parameters representing a decision function,bthe amount of offset is indicated by an indication,y i representing a single random output that blurs the coarse concentration.
In step S43, a Lagrangian multiplier is introduced
Figure 942569DEST_PATH_IMAGE055
Determining the corresponding Lagrangian function as:
Figure 381640DEST_PATH_IMAGE056
，
wherein the content of the first and second substances,
Figure 923480DEST_PATH_IMAGE057
representing a non-linear mapping that maps the original space to a high-dimensional feature space. In the embodiment of the invention, the parameters of the Gaussian kernel function are selected and solved to obtain the sample pointsx i Of lagrange multiplier
Figure 671862DEST_PATH_IMAGE058
And offsetb。
In step S44, an optimization function is determined by the lagrange multiplier method according to the similarity between the combined kernel maximizing the plurality of basic kernel functions and the ideal kernel. In the embodiment of the invention, the optimization problem of the multi-core support vector machine based on the core alignment is solved to obtain the optimized core weightηAnd lagrange multiplier of the optimal solution
Figure 299153DEST_PATH_IMAGE058
。
In the embodiment of the invention, the multi-core function combination mode based on the core alignment weighted summation is as follows:
Figure 541915DEST_PATH_IMAGE059
wherein the content of the first and second substances,
Figure 203841DEST_PATH_IMAGE060
the weight of the kernel is represented by,
Figure 624589DEST_PATH_IMAGE061
expressing a basic kernel matrix, and determining the maximum similarity of the combined kernel and the ideal kernel to obtain the following formula:
Figure 473596DEST_PATH_IMAGE062
to solve the optimization problem, in the embodiment of the present invention, let
Figure 520050DEST_PATH_IMAGE063
The denominator in (1) is constant, i.e. the above formula is equivalent to:
Figure 36482DEST_PATH_IMAGE064
converting the above formula into a dual problem thereof by a Lagrange multiplier method, and determining an optimization function as follows:
Figure 144244DEST_PATH_IMAGE065
wherein the content of the first and second substances,
Figure 480548DEST_PATH_IMAGE066
with penalty factorCMay vary.
In step S45, the optimization function is solved, and the optimized kernel weight and the lagrangian multiplier of the optimal solution are determined. In the embodiment of the invention, the current core weight
Figure 65113DEST_PATH_IMAGE068
By making a linear change, the value of the kernel alignment remains unchanged, so here will be
Figure 436051DEST_PATH_IMAGE069
With 1, the above optimization function can be simplified as follows:
Figure 198602DEST_PATH_IMAGE070
in the embodiment of the invention, the simplified optimization function is solved, so that the optimized kernel weight and the Lagrangian multiplier of the optimal solution can be determined.
It will be appreciated that for the sample to be tested, the decision function is calculated:
Figure 287781DEST_PATH_IMAGE071
in the embodiment of the invention, the positive and negative values of the post-sgn calculation are obtained so as to obtain the decision value. The calculation of the sgn function is:
Figure 410458DEST_PATH_IMAGE072
thereby deriving decision weights. It can be understood that training and optimization of the multi-core SVM are performed by combining specific parameters summarized in different scenes and different training sets and test sets with test accuracy and test time as evaluation indexes, and prediction results are predicted and output.
By adopting the multi-core SVM training method for identifying the intrusion signals, disclosed by the embodiment of the invention, the multi-core SVM is trained and optimized by selecting a plurality of basic kernel functions, determining the kernel weight of each basic kernel function based on kernel alignment, constructing the multi-core function according to the kernel weights and combining the sample membership degree of the calculated noise rough set method, so that the optimized kernel weights and the Lagrangian multiplier of the optimal solution are obtained, the classification interval of the trained multi-core SVM is maximized, and the identification of different types of intrusion signals is more accurate.
The embodiment of the second aspect of the invention also provides a multi-core SVM training device for intrusion signal recognition. Fig. 5 is a schematic structural diagram of a multi-core SVM training apparatus 500 for intrusion signal recognition according to an embodiment of the present invention, which includes an obtaining module 501, a processing module 502, and a training module 503.
The obtaining module 501 is configured to obtain an intrusion signal data set, which is expressed as intrusion signal data set, and perform normalization processing on the intrusion signal data set
Figure DEST_PATH_IMAGE073
Wherein, in the step (A),xandyrespectively representing the random input and output results,land representing the number of samples corresponding to the data, and marking the noise, the wave crest and/or the frequency according to the characteristics of the intrusion signal in the intrusion signal data set.
The processing module 502 is configured to select a plurality of basic kernel functions, determine similarity between corresponding kernel matrices of the plurality of basic kernel functions according to the plurality of basic kernel functions and the intrusion signal data set, and determine the kernel weight of each basic kernel function according to the similarity; and is further configured to determine a multi-kernel function based on the kernel weights.
The judging module 503 is configured to determine a sample membership degree according to a fuzzy rough set method, perform multi-core SVM training and optimization according to the sample membership degree and the multi-core function, determine an optimized core weight and a lagrangian multiplier of an optimal solution, and complete training of the multi-core SVM.
In this embodiment of the present invention, the processing module 502 is further configured to determine, according to the selected plurality of basic kernel functions and the intrusion signal data set, similarities between kernel matrices corresponding to the plurality of basic kernel functions; and further for determining the kernel weight for each of the base kernel functions based on the similarity.
In this embodiment of the present invention, the determining module 503 is further configured to determine the sample membership of the intrusion signal using an approximation operator under a fuzzy rough set based on a gaussian kernel.
For a more specific implementation manner of each module of the multi-core SVM training device 500 for intrusion signal recognition, reference may be made to the description of the multi-core SVM training method for intrusion signal recognition of the present invention, and similar beneficial effects are obtained, and no further description is given here.
The embodiment of the third aspect of the invention also provides an alarm method for intrusion signal identification. Fig. 6 is a flowchart illustrating an alarm method for intrusion signal identification according to an embodiment of the present invention, including steps S61 to S62, where:
in step S61, an original intrusion signal is obtained, positive-negative complementary empirical mode decomposition is performed, a plurality of IMF components are obtained and input to a multi-core SVM model for recognition, wherein the multi-core SVM model is trained by using the above described multi-core SVM training method for intrusion signal recognition. In the embodiment of the invention, under the multi-core function mapping, a plurality of feature spaces are combined into a high-dimensional space, and different feature components of heterogeneous data can be solved by corresponding core functions respectively due to different feature mapping capabilities of each basic core function of the combined space. The basic kernel functions may first be linearly combined, as shown in the following expression:
Figure 681908DEST_PATH_IMAGE074
wherein the content of the first and second substances,
Figure 864627DEST_PATH_IMAGE075
the original multi-core matrix is directly normalized and summed, or weighted and summed, but the biggest problem of the method is that some characteristic information of the original data may be lost, for example, a certain block area of the data information contains a lot of variable information, when an averaging or weighted averaging method is used, the data representing the variable information may be smoothed by the core function, the integrity of the final characteristic information is damaged, and the classification accuracy of the classifier is affected. The IMF components from empirical mode decomposition are constructed into a composite matrix of the form:
Figure 175523DEST_PATH_IMAGE076
as can be seen from the above equation, the kernel matrix of the original intrusion signal component is on the diagonal of the composite matrix, and other elements in the matrix are defined as
Figure 101891DEST_PATH_IMAGE077
Of two different kernel matrices. Because the gaussian kernel function only has a width parameter σ, which reflects the corresponding width of the internal dot product to the input, and the value of σ affects the test sample time and the classification accuracy of the SVM, a proper parameter needs to be selected to transform the data set, and meanwhile, the punishment degree of the data set, the complexity of a balanced classifier model and the error are controlled by a punishment coefficient C, taking two gaussian kernel functions as an example, the following formula can be used for solving:
Figure 198154DEST_PATH_IMAGE078
wherein d represents the maximum distance corresponding to all points in one classification datasetThe value of the one or more of,m i 、m j respectively representing datax i 、x j White gaussian noise is present.
It will be appreciated that the method of synthesizing kernel functions can better process data sets when there is a local change in the distribution of these data sets. In the multi-core SVM algorithm, different kernel function combinations are selected according to different intrusion and false alarm signal characteristics, the kernel combination method depends on a training data set to a large extent, some weight coefficients can be obtained through training and learning to identify the importance of each kernel and make up the problem that sub-kernel functions are selected to be non-optimal, and linear kernel functions and radial basis kernel functions are selected in the embodiment of the invention to enable classification to be more accurate.
In an embodiment of the invention, a set of empirical mode decomposition components is selectedIMF i The singular value features of the multi-core SVM are input as the feature vectors of the trained multi-core SVM, and the state and the intrusion behavior of the electronic fence are recognized. In the current research on classification methods of SVM, there are one-to-one, one-to-many methods and binary tree-based methods, and according to the nonlinearity of system pulse signals and the diversity of intrusion behaviors, if a basic binary classifier or a basic three-class classifier is selected, only classification can be performed to judge whether the signal characteristics are intrusion behaviors or false alarm conditions to determine whether alarming is needed, and various types of intrusion behaviors cannot be identified.
In the embodiment of the invention, for various noises of an electronic fence system under the conditions of short circuit, open circuit (shearing), climbing, knocking, strong wind, rain and the like in a fault state, each type of noise sample is used for training and testing, and the idea is to design an SVM between any two types of samples. When one type of samples is classified and predicted, all noise classifiers are used for matching the type of samples, the category with the most votes is the final category of the type of samples, and different weight values are assigned to different types of noise. Respectively sampling the signals for N times according to the 12.5 MHz sampling frequency in the high-speed AD sampling ranging, and obtaining 4N original signals as sample data, wherein the 4N original signals are respectively used for simulating a strong wind interference condition, a rainfall interference test, a touch and climbing behavior and a wearing insulating material touch behavior; performing positive-negative complementary empirical mode decomposition on the signal samples in each group of states to obtain a plurality of IMF components, selecting IMF components with the amplitude larger than 3 as source characteristic signals for characteristic vector calculation due to different quantities of IMF components in different states, selecting n IMF components containing invasive characteristic information as research objects, calculating singular value entropies of the IMF components, constructing energy characteristic vectors of singular values as input of a multi-core SVM (support vector machine) for training, and finally determining the working state and the fault condition through the output of an SVM classifier. The singular value entropies of the various intrusion signals and false alarm states are shown in tables 1 and 2 below.
TABLE 1 ComponentsIMF 1 ~ IMF 8Entropy of singular value of
Figure 551775DEST_PATH_IMAGE079
TABLE 2 ComponentsIMF 9~ IMF 16Entropy of singular value of
Figure 349967DEST_PATH_IMAGE080
As can be seen from tables 1 and 2, the singular value entropies of the high-frequency parts of the two intrusion behaviors and the two false alarm signals are higher than those of the low-frequency parts, which indicates that the energy distribution in the high-frequency component part is relatively even, the singular value entropies are not greatly different, and the signal characteristics and behavior distinguishing properties cannot be distinguished; the singular value entropy of the low-frequency part is smaller, the magnitude becomes smaller after the 10 th IMF component and the 11 th IMF component, the difference between the singular value entropies is gradually shown, the main frequency components of the signals are represented, the singular value entropies of the low-frequency part of each signal are different due to different intrusion behaviors and different frequency distributions of the false alarm state signals, and then the signal identification can be carried out.
In step S62, an alarm is given according to the recognition result of the multi-core SVM model. In the embodiment of the present invention, after the identification components of each state are determined, 80 sets of signal sample data are selected as training samples, 60 sets of test identification samples are tested, the accuracy of classification identification is checked, and the test results are shown in table 3 below. It can be understood that after the multi-core SVM model identifies different types of intrusion signals, an alarm may be given according to different preset manners, such as sound, light warning, or user operating system information prompt, which is not limited in the present invention.
TABLE 3 test false counts and classifier recognition rates
Figure 80025DEST_PATH_IMAGE081
The method has the advantages that the positive and negative Gaussian white noises are added into the original signal and then the signal decomposition is carried out, the residual noises in the decomposition process can be effectively neutralized, the difference of signal characteristics can be better analyzed, the false alarm rate is obviously reduced by carrying out test classification comparison on the single-core SVM and the multi-core SVM, and the recognition efficiency of the multi-core SVM method is superior to that of the single-core SVM. As can be seen from table 3, after the classification, identification and detection of 80 groups of training samples and 60 groups of test samples, the number of false alarms in strong wind and rainy days is reduced, wherein the number of false alarms in strong wind is reduced by 2, the effective identification rate is increased from 90% to 97%, and the false alarm rate is reduced; meanwhile, when the climbing behavior is identified, the number of false alarms is reduced by 3, the identification rate is improved to 88% from 82%, the number of false alarms of the knocking behavior is reduced by three, and the identification rate is improved to 87% from 79%. Therefore, the identification accuracy can be effectively improved and the noise influence and the false alarm rate are reduced by calculating the singular value entropy and the classification identification method of the multi-core SVM.
It can be understood that, before identifying the echo signal, the echo signal is preprocessed by an adaptive filtering method, a voltage comparison circuit and a pulse peak detection method, so as to obtain an identification signal to be detected which meets the requirement, and then the extraction of the relevant signal characteristics is carried out, in order to remove the noise influence of the pulse signal in a transmission cable, a device and the external environment before identifying, the peak threshold value is determined, the signal needs to be subjected to adaptive filtering, normalization processing and the like, so that the change range of the signal characteristic information is narrowed. Meanwhile, parameters of time domain and frequency domain related to echo signals with typical noise characteristics, such as strong wind, rain, climbing, knocking and the like, need to be estimated in advance, and reliable characteristic parameters are provided for signal identification. In addition, the signal characteristics generated by the corresponding intrusion and disturbance behaviors are extracted, classified and identified, and compared with the signal characteristics in a normal state, so that the threshold range is determined, and finally, effective intrusion signals are identified.
By adopting the alarm method for identifying the intrusion signal, which is disclosed by the embodiment of the invention, the signal decomposition is carried out after positive and negative white Gaussian noises are added into the original signal, so that the residual noises in the decomposition process can be effectively neutralized, and the difference of the signal characteristics can be better analyzed. After singular value entropy is obtained, intrusion signal type recognition is carried out according to the trained multi-core SVM, and alarming is carried out according to the corresponding type, so that the accuracy rate of alarming can be effectively improved.
The embodiment of the fourth aspect of the invention also provides an alarm device for intrusion signal identification. Fig. 7 is a schematic structural diagram of an intrusion signal recognition alarm device 700 according to an embodiment of the present invention, which includes a memory 701 and a processor 702.
The memory 701 is used to store computer programs.
The processor 702 is adapted to implement the alarm method of intrusion signal identification as described above when executing the computer program.
For a more specific implementation manner of each module of the alarm device 700 for intrusion signal identification, reference may be made to the description of the alarm method for intrusion signal identification of the present invention, and similar beneficial effects are obtained, and no further description is given here.
The embodiment of the fifth aspect of the invention also provides an alarm system for intrusion signal identification. Fig. 8 is a schematic structural diagram of an intrusion signal recognition alarm system 800 according to an embodiment of the present invention, which includes a plurality of front-end detection electronic fences 801, an intrusion signal recognition alarm device 700 as described above, and a user operation platform 802.
A plurality of front-end detection electronic fences 801 are used to detect intrusion signals. In the embodiment of the present invention, the front-end detection electronic fence 801 transmits a frequency modulated pulse signal to detect an intrusion signal, and is further used to implement functions such as fault location.
The alarm device 700 for identifying the intrusion signal is used for identifying and alarming the intrusion signal. In the embodiment of the present invention, the alarm device 700 for intrusion signal recognition is used for signal recognition, including empirical mode decomposition and SVM classification as described above, and the alarm device 700 for intrusion signal recognition is also used for preprocessing the intrusion signal, such as effective echo detection and pulse peak detection.
The user operation platform 802 is used for management and query. In the embodiment of the present invention, the user operation platform 802 may implement functions such as user management, video monitoring, electronic maps, voice alarm, and alert inquiry. It is understood that the user operation platform 802 can perform function expansion according to practical application requirements, and the invention is not limited thereto.
By adopting the alarm system for intrusion signal identification provided by the embodiment of the invention, a plurality of front-end detection fences are arranged to detect echo signals so as to obtain possible intrusion signals, the alarm device for intrusion signal identification based on the multi-core SVM is used for signal analysis and identification so as to obtain different intrusion signal types, and then the alarm device alarms according to different intrusion signals, so that high-efficiency and accurate intrusion identification is realized, and the safety of users is ensured.
An embodiment of the sixth aspect of the present invention proposes a non-transitory computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, implements the multi-core SVM training method for intrusion signal recognition according to the first aspect of the present invention or implements the alarm method for intrusion signal recognition according to the third aspect of the present invention.
Generally, computer instructions for carrying out the methods of the present invention may be carried using any combination of one or more computer-readable storage media. Non-transitory computer readable storage media may include any computer readable medium except for the signal itself, which is temporarily propagating.
A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
Computer program code for carrying out operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C + +, and conventional procedural programming languages, such as the "C" programming language or similar programming languages, and in particular may employ Python languages suitable for neural network computing and TensorFlow, PyTorch-based platform frameworks. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider).
The non-transitory computer-readable storage medium according to the sixth aspect of the present invention may be implemented with reference to the contents specifically described in the embodiments of the first aspect or the third aspect of the present invention, and has similar beneficial effects to the multi-core SVM training method for intrusion signal recognition according to the embodiments of the first aspect of the present invention or the alarm method for intrusion signal recognition according to the embodiments of the third aspect of the present invention, and is not described herein again.
Although embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are illustrative and not to be construed as limiting the present invention, and that changes, modifications, substitutions and alterations can be made in the above embodiments by those of ordinary skill in the art within the scope of the present invention.

Claims (10)

1. A multi-core SVM training method for intrusion signal recognition is characterized by comprising the following steps:
acquiring an intrusion signal data set and carrying out normalization processing, wherein the intrusion signal data set is expressed as
Figure 621013DEST_PATH_IMAGE001
Wherein, in the step (A),xandyrespectively representing the random input and output results,lrepresenting the number of samples corresponding to the data, and marking noise, wave crests and/or frequency in the intrusion signal data set according to the characteristics of the intrusion signals;
selecting a plurality of basic kernel functions, determining the similarity between corresponding kernel matrixes of the basic kernel functions according to the basic kernel functions and the intrusion signal data set, and determining the kernel weight of each basic kernel function according to the similarity;
determining a multi-kernel function according to the kernel weight;
determining sample membership according to a fuzzy rough set method, performing multi-core SVM training and optimization according to the sample membership and the multi-core function, determining the optimized core weight and the Lagrange multiplier of the optimal solution, and finishing the training of the multi-core SVM.
2. The method of claim 1, wherein determining similarity between corresponding kernel matrices of a plurality of the base kernel functions according to the selected plurality of the base kernel functions and the intrusion signal data set comprises:
determining a phase between a plurality of corresponding kernel matrices of the base kernel functionSimilarity of characters
Figure 94720DEST_PATH_IMAGE002
Wherein, in the step (A),Trepresenting the intrusion signal data in a data stream,
Figure 188447DEST_PATH_IMAGE003
and
Figure 140222DEST_PATH_IMAGE004
respectively representing basic kernel functions
Figure 284896DEST_PATH_IMAGE005
And
Figure 54138DEST_PATH_IMAGE006
the kernel matrix after the mapping is performed,
Figure 510527DEST_PATH_IMAGE007
，
Figure 406939DEST_PATH_IMAGE008
separately representing intrusion signal datasetsTThe sample of (1).
3. The multi-core SVM training method for intrusion signal recognition according to claim 2, wherein the determining the core weight of each of the base kernel functions according to the similarity comprises:
determining a maximized similarity of a combined kernel of a plurality of the base kernel functions to an ideal kernel;
and optimizing according to the similarity and the maximized similarity, and determining the kernel weight of each basic kernel function.
4. The multi-core SVM training method for intrusion signal recognition according to any one of claims 1-3, wherein the determining sample membership according to a fuzzy rough set method comprises:
determining the sample membership of an intrusion signal using an approximation operator under a gaussian kernel based fuzzy rough set:
Figure 999594DEST_PATH_IMAGE009
，
wherein the content of the first and second substances,
Figure 939737DEST_PATH_IMAGE010
representation of belonging to a sample set
Figure 883422DEST_PATH_IMAGE011
The degree of membership of (a) is,
Figure 317946DEST_PATH_IMAGE012
，
Figure 30687DEST_PATH_IMAGE013
to represent
Figure 876152DEST_PATH_IMAGE014
Corresponding to the blur value in the blur rough set.
5. The method of claim 4, wherein the performing the multi-core SVM training and optimization according to the sample membership and the multi-core function, and determining the optimized kernel weight and the Lagrangian multiplier of the optimal solution comprises:
the optimization problem of the multi-core SVM is determined as follows:
Figure 307133DEST_PATH_IMAGE015
wherein the content of the first and second substances,
Figure 279768DEST_PATH_IMAGE016
representing the direction of the classification hyperplane in the high-dimensional feature space,ηthe weight of the kernel is represented by,
Figure 112595DEST_PATH_IMAGE017
the value of the relaxation variable is represented by,
Figure 128962DEST_PATH_IMAGE018
，Ca penalty factor is represented which is a function of,wa generalized set of parameters representing a decision function,bthe amount of offset is indicated by an indication,y i representing a single random output in a fuzzy rough set;
introducing lagrange multipliers
Figure 188185DEST_PATH_IMAGE019
Determining the corresponding Lagrangian function as:
Figure 89144DEST_PATH_IMAGE020
，
wherein the content of the first and second substances,
Figure 635532DEST_PATH_IMAGE021
representing a non-linear mapping that maps the original space to a high-dimensional feature space;
according to the similarity between the combined kernel maximizing the plurality of basic kernel functions and the ideal kernel, determining an optimization function as follows by a Lagrange multiplier method:
Figure 963745DEST_PATH_IMAGE022
wherein the content of the first and second substances,
Figure 979106DEST_PATH_IMAGE023
with penalty factorCIs changed;
and solving the optimization function, and determining the optimized kernel weight and the Lagrangian multiplier of the optimal solution.
6. A multi-core SVM training device for intrusion signal recognition is characterized by comprising:
an acquisition module for acquiring and normalizing an intrusion signal data set expressed as
Figure 683757DEST_PATH_IMAGE024
Wherein, in the step (A),xandyrespectively representing the random input and output results,lrepresenting the number of samples corresponding to the data, and marking noise, wave crests and/or frequency in the intrusion signal data set according to the characteristics of the intrusion signals;
the processing module is used for selecting a plurality of basic kernel functions, determining the similarity between corresponding kernel matrixes of the basic kernel functions according to the basic kernel functions and the intrusion signal data set, and determining the kernel weight of each basic kernel function according to the similarity; the multi-core function is determined according to the core weight;
and the training module is used for determining sample membership according to a fuzzy rough set method, training and optimizing the multi-core SVM according to the sample membership and the multi-core function, determining the optimized core weight and the Lagrangian multiplier of the optimal solution, and finishing the training of the multi-core SVM.
7. An alarm method for intrusion signal identification, comprising:
acquiring an original intrusion signal, performing positive-negative complementary empirical mode decomposition, and inputting a plurality of IMF components into a multi-core SVM model for recognition, wherein the multi-core SVM model is trained by adopting the multi-core SVM training method for intrusion signal recognition according to any one of claims 1-5;
and alarming according to the recognition result of the multi-core SVM model.
8. An alarm device for intrusion signal recognition is characterized by comprising a memory and a processor; the memory for storing a computer program; the processor, when executing the computer program, is configured to implement the intrusion signal recognition alarm method according to claim 7.
9. An alarm system of intrusion signal recognition, characterized by, including a plurality of front end detection electronic fences, the alarm device of intrusion signal recognition and user operation platform of claim 8, wherein, a plurality of front end detection electronic fences are used for detecting the intrusion signal, the alarm device of intrusion signal recognition is used for right the intrusion signal discerns and reports to the police, user operation platform is used for managing and inquiring.
10. A non-transitory computer-readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements the multicore SVM training method for intrusion signal recognition according to any one of claims 1 to 5 or the alarm method for intrusion signal recognition according to claim 7.
CN202011490179.5A 2020-12-17 2020-12-17 Multi-core SVM training and alarming method, device and system for intrusion signal recognition Pending CN112232329A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011490179.5A CN112232329A (en) 2020-12-17 2020-12-17 Multi-core SVM training and alarming method, device and system for intrusion signal recognition

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011490179.5A CN112232329A (en) 2020-12-17 2020-12-17 Multi-core SVM training and alarming method, device and system for intrusion signal recognition

Publications (1)

Publication Number Publication Date
CN112232329A true CN112232329A (en) 2021-01-15

Family

ID=74124168

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011490179.5A Pending CN112232329A (en) 2020-12-17 2020-12-17 Multi-core SVM training and alarming method, device and system for intrusion signal recognition

Country Status (1)

Country Link
CN (1) CN112232329A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114143164A (en) * 2021-11-11 2022-03-04 北京银盾泰安网络科技有限公司 Data alarm linkage platform

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106127135A (en) * 2016-06-21 2016-11-16 长江大学 A kind of Ling Qu invasion vibration signal characteristics extracts and classification and identification algorithm
CN108506736A (en) * 2018-03-19 2018-09-07 浙江理工大学 A kind of invasion signal diagnostic system and method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106127135A (en) * 2016-06-21 2016-11-16 长江大学 A kind of Ling Qu invasion vibration signal characteristics extracts and classification and identification algorithm
CN108506736A (en) * 2018-03-19 2018-09-07 浙江理工大学 A kind of invasion signal diagnostic system and method

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
何强 等: "《核对齐多核模糊支持向量机》", 《智能***学报》 *
蒋立辉 等: "《光纤周界入侵信号特征提取与识别方法的研究》", 《激光与红外》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114143164A (en) * 2021-11-11 2022-03-04 北京银盾泰安网络科技有限公司 Data alarm linkage platform

Similar Documents

Publication Publication Date Title
CN106874833B (en) Vibration event pattern recognition method
CN103617684B (en) Interference-type optical fiber circumference vibrating intruding recognizer
CN110298235B (en) Hyperspectral anomaly detection method and system based on manifold constraint self-coding network
CN110120230B (en) Acoustic event detection method and device
CN111859010B (en) Semi-supervised audio event identification method based on depth mutual information maximization
CN109117641A (en) A kind of network security risk evaluation method based on I-HMM
CN114023354A (en) Guidance type acoustic event detection model training method based on focusing loss function
CN111209853B (en) Optical fiber sensing vibration signal mode identification method based on AdaBoost-ESN algorithm
CN113571067A (en) Voiceprint recognition countermeasure sample generation method based on boundary attack
CN112560596A (en) Radar interference category identification method and system
CN115811440B (en) Real-time flow detection method based on network situation awareness
CN114218998A (en) Power system abnormal behavior analysis method based on hidden Markov model
CN113707175B (en) Acoustic event detection system based on feature decomposition classifier and adaptive post-processing
CN113780094B (en) Perimeter intrusion identification method based on Gaussian mixture model and hidden Markov model
CN111476102A (en) Safety protection method, central control equipment and computer storage medium
CN117197726B (en) Important personnel accurate management and control system and method
CN112232329A (en) Multi-core SVM training and alarming method, device and system for intrusion signal recognition
Heda et al. Performance evaluation of YOLOv3, YOLOv4 and YOLOv5 for real-time human detection
CN117236699A (en) Network risk identification method and system based on big data analysis
Sun et al. Vehicle acoustic and seismic synchronization signal classification using long-term features
CN116318877A (en) Method for defending countersamples by using intrusion detection system with various characteristic manifold vectors
CN115277159A (en) Industrial Internet security situation assessment method based on improved random forest
CN115643153A (en) Alarm correlation analysis method based on graph neural network
CN118010848B (en) Intelligent anchorage device ponding detection method and system
CN117521042B (en) High-risk authorized user identification method based on ensemble learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20210115