CN109034088A - A kind of unmanned plane signal detection method and device - Google Patents

A kind of unmanned plane signal detection method and device Download PDF

Info

Publication number
CN109034088A
CN109034088A CN201810885975.5A CN201810885975A CN109034088A CN 109034088 A CN109034088 A CN 109034088A CN 201810885975 A CN201810885975 A CN 201810885975A CN 109034088 A CN109034088 A CN 109034088A
Authority
CN
China
Prior art keywords
signal
sample
indicate
observed
subsignal
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
CN201810885975.5A
Other languages
Chinese (zh)
Inventor
冯志勇
张轶凡
于轩
黄赛
宁帆
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.)
Beijing University of Posts and Telecommunications
Original Assignee
Beijing University of Posts and Telecommunications
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 Beijing University of Posts and Telecommunications filed Critical Beijing University of Posts and Telecommunications
Priority to CN201810885975.5A priority Critical patent/CN109034088A/en
Publication of CN109034088A publication Critical patent/CN109034088A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • 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
    • 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)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Signal Processing (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

The embodiment of the invention provides a kind of unmanned plane signal detection method and devices.Concrete scheme is as follows: wireless signal in available frequency range to be detected, as signal to be detected, according to predeterminated frequency interval, divide signal to be detected, obtain the wireless subsignal in multiple frequency ranges, as signal to be observed, for each signal to be observed, based on preset energy characteristic formula, extract the energy feature of the signal to be observed, based on preset cumulative measure feature formula, extract the accumulation measure feature of the signal to be observed, for each observation signal, using the energy feature of the observation signal and accumulation measure feature as input data, input SVM classifier trained in advance, utilize the categorised decision function in SVM classifier, determine in signal to be detected whether include unmanned plane signal.Using scheme provided in an embodiment of the present invention, the detection to unmanned plane signal in wireless environment may be implemented.

Description

A kind of unmanned plane signal detection method and device
Technical field
The present invention relates to the signal detection and recognition technical fields of wireless communication field, believe more particularly to a kind of unmanned plane Number detection method and device.
Background technique
With the continuous development of unmanned air vehicle technique, the application of unmanned plane is also more extensive.Unmanned air vehicle technique take photo by plane, agricultural Plant protection, express transportation and disaster relief etc. various aspects field be production and life bring convenience, at the same time, also along with Occur taking on the sly, transport the problems such as violation article, interference civil aviaton's landing signal.Therefore, for unmanned plane signal in wireless environment Detection be increasingly becoming an important research direction.
Previous unmanned plane detection method is mostly in the mixed signal for be directed to unmanned plane signal and noise signal to unmanned plane Signal carries out signal detection and signal identification, but as wireless communications environment is increasingly complicated, for example, unmanned plane can work 2.4GHz ISM (Industrial Scientific Medical, industry, science, medical treatment) frequency range, but the frequency range is simultaneously There are Wi-Fi (Wireless Fidelity, Wireless Fidelity) signal, Bluetooth signal, wireless phone issue signal and adopt The signal etc. carried out wireless communication with Zigbee (purple honeybee) technology, therefore, previous detection method has been no longer desirable for nobody The detection of machine signal.
It mainly include signal detection and two parts of signal identification about signal detection.Signal detection, which refers to, to be intercepted and captured Or the signal received carries out specific analysis processing, judges the process whether target frequency bands have signal to occur.And signal identification Refer to and deeper analysis processing carried out to signal, the signal that further identification occurs whether be echo signal process.
During signal detection, generally first with sequence structure test statistics is received, later by statistic and door Limit or sorting criterion are made comparisons.However, signal identification due to wireless environment there are multi-signal aliasings the phenomenon that, electromagnetic environment compared with For complexity, there is presently no the recognition methods of relative maturity.In general, existing signal detection recognizer mainly include with Lower four classes: matched filter detection, waveforms detection, correlation matrix detection and energy measuring.Wherein, matched filter detection and wave The realization of shape detection needs enough prior information, and it is actually detected in be generally difficult to meet this condition；Though correlation matrix detects It can be used for blind Detecting, but calculate complexity, and the stability of algorithm performance under circumstances is poor.Energy measuring is several compared to above Kind method has the advantages that realize simple, algorithm complexity is low etc., is most common signal detection algorithm in actually detected.
In the prior art, according to unmanned plane signal on frequency domain be class trapezoidal wave characteristic, can based on energy measuring with Recognition methods indicates the energy of signal using the variation of the Signal gradient value at each Frequency point of frequency-region signal, to nobody Machine signal is identified.But during the detection of energy and recognition methods, due to only focusing on energy feature, signal is caused to be visited It is sensitive to noise signal during survey.That is, in the case that unmanned plane signal is in low signal-to-noise ratio, due to wireless environment In signal mix, wherein the difference between signals and associated noises and pure noise cancellation signal will become smaller, and the detection of unmanned plane signal also will go out Existing deviation.Meanwhile when the interference signal in wireless environment is there are when the non-unmanned plane signal of other class trapezoidal waves, use is existing Energy measuring and recognition methods are identified only for class trapezoidal wave or Signal gradient, also lead to the detection of unmanned plane signal There is deviation.
Summary of the invention
The embodiment of the present invention is designed to provide a kind of unmanned plane signal detection method and device, to realize to wireless ring The detection of unmanned plane signal in border.Specific technical solution is as follows:
The embodiment of the invention provides a kind of unmanned plane signal detection method, the methods, comprising:
The wireless signal in frequency range to be detected is obtained, as signal to be detected；
According to predeterminated frequency interval, divide the signal to be detected, obtain the wireless subsignal in multiple frequency ranges, as to Observation signal；
For each signal to be observed, it is based on preset energy characteristic formula, the energy for extracting the signal to be observed is special Sign；
For each signal to be observed, it is based on preset cumulative measure feature formula, extracts the accumulation of the signal to be observed Measure feature；
For each signal to be observed, the energy feature of the signal to be observed and the accumulation measure feature are made For input data, input support vector machines classifier trained in advance utilizes the categorised decision letter in the SVM classifier Number, determines in the signal to be detected whether include unmanned plane signal, wherein the SVM classifier is based on sample signal Characteristic training obtain, the characteristic of the sample signal includes by sample subsignal each in the sample signal Energy feature and accumulation measure feature the sample supporting vector and the corresponding mark value of the sample subsignal that collectively form, The mark value indicates to whether there is unmanned plane signal in the sample subsignal.
Further, for each signal to be observed, it is based on preset energy characteristic formula, extracts the signal to be observed Energy feature, comprising:
For each signal to be observed, it is based on Parseval's theorem, calculates the corresponding energy value of the signal to be observed, As energy feature, wherein be directed to the corresponding signal y to be observed of i-th of frequency rangei(t), the energy feature EiIt indicates Are as follows:
Yi(f) the signal y to be observed is indicatedi(t) corresponding frequency-region signal, f1Indicate the corresponding lower boundary frequency of i-th of frequency range Rate, f2Indicate the corresponding coboundary frequency of i-th of frequency range, Δ f indicates the frequency sampling interval in i-th of frequency range, when t is indicated Between, f indicates frequency.
Further, for each signal to be observed, it is based on preset cumulative measure feature formula, extracts the letter to be observed Number accumulation measure feature, comprising:
For each signal to be observed, the second-order cumulant, fourth order cumulant and six ranks of the signal to be observed are calculated Cumulant, wherein be directed to each signal to be observed, the second-order cumulant C21It indicates are as follows:
C21=E (| y (k)2|)-|Ey(k)2|
Y (k) indicates that k-th of signal to be observed, E () indicate mathematic expectaion, | | indicate modulus symbol, C21In 1 indicates to introduce a conjugate complex number；
For each signal to be observed, the fourth order statistic C42It indicates are as follows:
C42=E (| y (k)4|)-|Ey(k)2|2-2(E(|y(k)2|))2
Y (k) indicates that k-th of signal to be observed, E () indicate mathematic expectaion, | | indicate modulus symbol, C42In 2 indicate to introduce two conjugate complex numbers；
For each signal to be observed, the six ranks cumulant C63It indicates are as follows:
C63=E (| y (k)6|)-9E(|y(k)|4)E(|y(k)|2)-3E(y*(k)3y(k))E(y(k)2)-3E(y*(k)y (k)3)E(y*(k)2)-18E(y*(k)2)E(y(k)2)E(|y(k)|2)-12(E(|y(k)|2))3
Y (k) indicates that k-th of signal to be observed, E () indicate mathematic expectaion, | | indicate that modulus symbol, * indicate Conjugate of symbol, C63In 3 indicate introduce three conjugate complex numbers；
For each signal to be observed, according to the second-order cumulant of the signal to be observed, fourth order cumulant and Six rank cumulants determine the first accumulative measure feature and the second accumulation of the signal to be observed according to preset cumulative measure feature formula Measure feature, as accumulation measure feature, wherein be directed to each signal to be observed, the cumulant character representation are as follows:
γ1Indicate the first accumulation measure feature, γ2Indicate that the second accumulation measure feature, y (k) indicate described in k-th Signal to be observed, C21Indicate the second-order cumulant, C42Indicate the fourth order cumulant, C63Indicate the six ranks cumulant, C21 In 1 indicate introduce a conjugate complex number, C42In 2 indicate introduce two conjugate complex numbers, C63In 3 indicate introduce three altogether Yoke plural number.
Further, for each signal to be observed, the energy feature of the signal to be observed is tired out with described Accumulated amount feature is as input data, input support vector machines classifier trained in advance, using in the SVM classifier Categorised decision function determines in the signal to be detected whether include unmanned plane signal, comprising:
For each signal to be observed, the energy feature of the signal to be observed and the accumulation measure feature are made For vector element, a corresponding supporting vector is generated, wherein the corresponding supporting vector of n-th of signal to be observed It indicates are as follows:
EnIndicate the energy feature of the signal to be observed, γn1Indicate first cumulant of the signal to be observed Feature, γn2Indicate the second accumulation measure feature of the observation signal；
In the SVM classifier that the supporting vector is trained in advance as input data, input；
According to the categorised decision function in the SVM classifier, the corresponding value of the signal to be observed is determined, as decision Value, wherein the categorised decision function f (mn) indicate are as follows:
Categorised decision function f (mn) value indicate that the corresponding decision value of n-th of signal to be observed, sign () indicate Sign function, N indicate the quantity of the sample subsignal,Indicate the corresponding Lagrange multiplier of i-th of sample supporting vector Allelomorph,<>indicate inner product operation, b*Indicate the value of amount of bias, mn TIndicate that the energy of n-th of signal to be observed is special It seeks peace and accumulates the transposition of the corresponding supporting vector of measure feature, xiIndicate that the energy feature and cumulant of i-th of sample subsignal are special Levy corresponding sample supporting vector, yiIndicate the corresponding mark value of i-th of sample subsignal；
Based on the decision value of each signal to be observed, determine in the signal to be detected whether include nobody Machine signal, when at least one value is 1 in the decision value, then it represents that it include unmanned plane signal in the signal to be detected, when The decision value it is all -1 when, then it represents that not comprising unmanned plane signal in the signal to be detected.
Further, the training process of SVM classifier, comprising:
According to preset condition, the wireless signal of preset quantity is obtained in frequency range to be detected, as sample signal；
According to predeterminated frequency interval, the sample signal is divided, the wireless subsignal in multiple frequency ranges is obtained, as sample Subsignal；
For each sample subsignal, the energy feature and accumulation measure feature of the sample subsignal are extracted；
It, will be by the energy feature of sample subsignal in the sample signal and tired based on the sample signal of the preset quantity The sample supporting vector and the corresponding mark value of the sample subsignal that accumulated amount feature is constituted as input data, adopt by input With current class device parameter and the SVM classifier of preset structure, using current class decision function, to the preset quantity Sample signal is classified, and a wheel training is completed, wherein when training for the first time, the current class device parameter is default initial Classifier parameters, the current class decision function are determined according to the default preliminary classification device parameter；
For each round training, based on the sample signal of the preset quantity, according to default loss function, determine described in The penalty values of SVM classifier；
When determining that the SVM classifier reaches preset standard based on the penalty values, training is completed, determines the SVM The corresponding categorised decision function of classifier；
When determining that the SVM classifier is not up to preset standard based on the penalty values, according to default adjustment mode, adjust The whole classifier parameters obtain new categorised decision function, and using new categorised decision function and the preset quantity Sample signal completes new round training.
Further, for each round training, based on the sample signal of the preset quantity, according to default loss function, Determine the penalty values of the SVM classifier, comprising:
After each round training is completed, it is based on hinge loss function, determines the SVM classifier penalty values, wherein institute State hinge loss function P (x) expression are as follows:
The value of the hinge loss function P (x) indicates the penalty values, and λ indicates parameter to be adjusted, and w indicates the svm classifier The normal vector of divisional plane in device, | | | | expression takes norm, and N indicates the quantity of the sample subsignal, wTIndicate the transposition of w, xi Indicate the energy feature sample supporting vector corresponding with accumulation measure feature by i-th of sample subsignal, yiIndicate i-th of sample The corresponding mark value of subsignal, b indicate intercept.
Further, when determining that the SVM classifier is not up to preset standard based on the penalty values, according to default tune Perfect square formula adjusts the classifier parameters, obtains new categorised decision function, and using new categorised decision function and described pre- If the sample signal of quantity, new round training is completed, comprising:
When determining that the SVM classifier is not up to preset standard based on the penalty values, letter is lost based on the hinge Number, using gradient descent method, determines the value of the parameter to be adjusted in the hinge loss function；
According to the value of the parameter to be adjusted, the corresponding categorised decision function of the SVM classifier is redefined；
Classified according to the categorised decision function redefined to the sample signal of the preset quantity, completes a new round Training.
Further, according to the value of the parameter to be adjusted, the corresponding categorised decision letter of the SVM classifier is redefined Number, comprising:
According to the value of the parameter to be adjusted, the punishment parameter of the SVM classifier is determined, wherein the punishment parameter table Show and adjusts the weight of function interval and accuracy of classifying in optimization direction, the punishment parameter D and described wait adjust between parameter lambda Relationship be expressed as:
Based on the punishment parameter, using convex optimization problem, determine that the SVM classifier corresponds to the division side of divisional plane Formula, wherein what the division mode expression of the divisional plane was classified using sample signal of the hyperplane to the preset quantity Method is embodied as:
s.t.yi(wTxi+b)≥1-ξi, i=1,2 ..., N
ξi>=0, i=1,2 ..., N
S.t. constraint condition is indicated, w indicates the normal vector of the hyperplane, | | | | expression takes norm, and D indicates punishment ginseng Number, N indicate the quantity of the sample subsignal, ξiIndicate the corresponding slack variable of i-th of sample supporting vector, yi(wTxi+b) Indicate spacing distance of i-th of sample supporting vector to divisional plane, yiIndicate the corresponding mark value of i-th of sample subsignal, wTTable Show the transposition of w, xiIndicate i-th of sample supporting vector, b indicates intercept；
Based on Lagrangian and function dualization method, the Solve problems of the division mode are converted, after conversion The division mode indicates are as follows:
0≤αi≤ D, i=1,2 ..., N
S.t. constraint condition, α are indicatediIndicate the value of the corresponding Lagrange multiplier of i-th of sample supporting vector, αjIt indicates The value of the corresponding Lagrange multiplier of j-th of sample supporting vector, yiIndicate the corresponding mark value of i-th of sample subsignal, yjTable Show the corresponding mark value of i-th of sample subsignal,The transposition of the corresponding sample supporting vector of i-th of sample subsignal, xjTable Show the corresponding sample supporting vector of j-th of sample subsignal,<>indicates that inner product operation, D indicate punishment parameter, and N indicates sample The quantity of subsignal；
Based on SMO algorithm, the value of Lagrange multiplier is determined；
Using the property of sample supporting vector, the value of amount of bias is determined, wherein the value b of the amount of bias*It indicates Are as follows:
N indicates the quantity of sample subsignal, ynIndicate the corresponding mark value of n-th of sample supporting vector, yiIt indicates i-th The corresponding mark value of sample supporting vector,Indicate the allelomorph of the corresponding Lagrange multiplier of i-th of sample supporting vector,Indicate the transposition of i-th of sample supporting vector, xnIndicate the corresponding sample supporting vector of n-th of sample subsignal,<>table Show inner product operation；
The value of value and the amount of bias based on the Lagrange multiplier, redefines categorised decision function.
The embodiment of the invention also provides one kind to be used for unmanned plane signal detecting device, described device, comprising:
Signal acquisition module to be detected, for obtaining the wireless signal in frequency range to be detected, as signal to be detected；
Signal acquisition module to be observed, for dividing the signal to be detected, obtaining multiple frequencies according to predeterminated frequency interval Wireless subsignal in section, as signal to be observed；
Power feature extraction module is based on preset energy characteristic formula, extracts for being directed to each signal to be observed The energy feature of the signal to be observed；
Cumulant characteristic extracting module, for being based on preset cumulative measure feature formula for each signal to be observed, Extract the accumulation measure feature of the signal to be observed；
Unmanned plane signal determining module, for being directed to each signal to be observed, by the energy of the signal to be observed As input data, input SVM classifier trained in advance utilizes the SVM classifier for measure feature and the accumulation measure feature In categorised decision function, determine in the signal to be detected whether include unmanned plane signal, wherein the SVM classifier It is that the characteristic training based on sample signal obtains, the characteristic of the sample signal includes by the sample signal The sample supporting vector and the sample subsignal that the energy feature and accumulation measure feature of each sample subsignal collectively form Corresponding mark value, the mark value indicate to whether there is unmanned plane signal in the sample subsignal.
Further, described device, further includes:
Sample signal obtains module, for obtaining the wireless communication of preset quantity in frequency range to be detected according to preset condition Number, as sample signal；
Sample subsignal obtains module, for dividing the sample signal, obtaining multiple frequency ranges according to predeterminated frequency interval Interior wireless subsignal, as sample subsignal；
Characteristic obtains module, and for being directed to each sample subsignal, the energy for extracting the sample subsignal is special It seeks peace and accumulates measure feature；
Sample signal categorization module will be by sample in the sample signal for the sample signal based on the preset quantity The sample supporting vector and the corresponding label of the sample subsignal that the energy feature and accumulation measure feature of this subsignal are constituted Value is used as input data, and input utilizes current class decision using current class device parameter and the SVM classifier of preset structure Function classifies to the sample signal of the preset quantity, completes a wheel training, wherein described current when training for the first time Classifier parameters are default preliminary classification device parameter, and the current class decision function is joined according to the default preliminary classification device Number determination；
Penalty values determining module, for being trained for each round, based on the sample signal of the preset quantity, according to default Loss function determines the penalty values of the SVM classifier；
Grader generation module, it is complete for when determining that the SVM classifier reaches preset standard based on the penalty values At training, the corresponding categorised decision function of the SVM classifier is determined；
Classifier parameters adjust module, determine the not up to pre- bidding of the SVM classifier based on the penalty values for working as On time, according to default adjustment mode, the classifier parameters are adjusted, obtain new categorised decision function, and use new classification The sample signal of decision function and the preset quantity completes new round training.
A kind of unmanned plane signal detection method and device provided in an embodiment of the present invention, nothing in available frequency range to be detected Line signal divides signal to be detected according to predeterminated frequency interval as signal to be detected, obtains the wireless son in multiple frequency ranges Signal is based on preset energy characteristic formula, extracts the signal to be observed as signal to be observed for each signal to be observed Energy feature, be based on preset cumulative measure feature formula, extract the accumulation measure feature of the signal to be observed, for each observation letter Number, using the energy feature of the observation signal and accumulation measure feature as input data, input SVM classifier trained in advance, benefit With the categorised decision function in SVM classifier, determine in signal to be detected whether include unmanned plane signal.Pass through the above method The detection to unmanned plane signal in wireless environment may be implemented.
Certainly, implement any of the products of the present invention or method it is not absolutely required at the same reach all the above excellent Point.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with It obtains other drawings based on these drawings.
Fig. 1 is a kind of one of flow chart of unmanned plane signal detection method provided in an embodiment of the present invention；
Fig. 2 is a kind of SVM classifier training method flow chart of unmanned plane signal detection provided in an embodiment of the present invention；
Fig. 3 is a kind of schematic diagram that SVM classifier classifies to sample signal provided in an embodiment of the present invention；
Fig. 4 is a kind of flow chart of unmanned plane signal detection method provided in an embodiment of the present invention with two；
Fig. 5 is a kind of one of structural schematic diagram of unmanned plane signal detecting device provided in an embodiment of the present invention；
Fig. 6 is a kind of second structural representation of unmanned plane signal detecting device provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other Embodiment shall fall within the protection scope of the present invention.
The embodiment of the invention provides a kind of unmanned plane signal detection methods, as shown in Figure 1, may comprise steps of:
Step S101 obtains the wireless signal in frequency range to be detected, as signal to be detected.
In this step, using antenna, signal receiver etc. in the target environment of required detection, according between preset time Every the wireless signal in frequency range to be detected being obtained, as signal to be detected.
Specifically, according to a kind of regulation for the unmanned plane working frequency being currently known, it can be for the frequency range of unmanned plane work There is 840.5MHz to 845MHz, 1430MHz to 1444MHz and 2408MHz to 2440MHz, it is assumed that now to need to 2408MHz Unmanned plane signal in this frequency range of 2440MHz is detected, then frequency range to be detected is exactly 2408MHz to 2440MHz, every 2 minutes, the wireless signal in the target environment of required detection is acquired using signal receiver, by collected wireless communication Number be used as signal to be detected.
Step S102 divides signal to be detected according to predeterminated frequency interval, obtains the wireless subsignal in multiple frequency ranges, As signal to be observed.
In this step, the frequency intervals such as detectable signal use can be treated according to the spectrogram of above-mentioned signal to be detected It divides, the corresponding wireless subsignal of each frequency range is obtained, as signal to be observed.
Step S103 is based on preset energy characteristic formula, extracts the energy of the signal to be observed for each signal to be observed Measure feature.
In this step, for each signal to be observed, it is based on Parseval's theorem, it is corresponding to calculate the signal to be observed Energy value, as energy feature.
Specifically, being directed to i-th of signal y to be observedi(t), using Parseval's theorem, its energy spy can be calculated Levy Ei, wherein EiIt can indicate are as follows:
Yi(f) the signal y to be observed is indicatedi(t) corresponding frequency-region signal, f1Indicate the corresponding lower boundary frequency of i-th of frequency range Rate, f2Indicate the corresponding coboundary frequency of i-th of frequency range, Δ f indicates the frequency sampling interval in i-th of frequency range, when t is indicated Between, f indicates frequency.
Step S104 is based on preset cumulative measure feature formula, extracts the signal to be observed for each signal to be observed Accumulate measure feature.
In this step, for each signal to be observed, the corresponding second-order cumulant of the observation signal, fourth-order cumulant are based on Amount and six rank cumulants extract the first accumulation measure feature and second of the signal to be observed according to preset cumulative amount formula Accumulate measure feature.
Specifically, being directed to k-th of signal y (k) to be observed, second-order cumulant C21It can indicate are as follows:
C21=E (| y (k)2|)-|Ey(k)2|
Its fourth order cumulant C42It can indicate are as follows:
C42=E (| y (k)4|)-|Ey(k)2|2-2(E(|y(k)2|))2
Its six ranks cumulant C63It can indicate are as follows:
C63=E (| y (k)6|)-9E(|y(k)|4)E(|y(k)|2)-3E(y*(k)3y(k))E(y(k)2)-3E(y*(k)y (k)3)E(y*(k)2)-18E(y*(k)2)E(y(k)2)E(|y(k)|2)-12(E(|y(k)|2))3
In the calculation formula of above-mentioned second-order cumulant, fourth order cumulant and six rank cumulants, y (k) indicate k-th to Observation signal, E () indicate mathematic expectaion, | | indicate that modulus symbol, * indicate conjugate of symbol, C21In 1 indicate introduce one Conjugate complex number, C42In 2 indicate introduce two conjugate complex numbers, C63In 3 indicate introduce three conjugate complex numbers.
According to above-mentioned second-order cumulant, fourth order cumulant and six rank cumulants, can be mentioned according to preset cumulative amount formula Take the accumulation measure feature of each signal to be observed.Wherein, preset cumulative measure feature can indicate are as follows:
γ1Indicate the first accumulation measure feature, γ2Indicate that the second accumulation measure feature, y (k) indicate k-th of signal to be observed, C21Indicate second-order cumulant, C42Indicate fourth order cumulant, C63Indicate six rank cumulants, C21In 1 indicate introduce a conjugate complex Number, C42In 2 indicate introduce two conjugate complex numbers, C63In 3 indicate introduce three conjugate complex numbers.
It can be seen from the above, since the single order cumulant of Gaussian random variable is the mean value of its stochastic variable, second-order cumulant Its variance of a random variable, and cumulants more than its Third-order cumulants or three ranks be equal to zero, therefore, Higher Order Cumulants To can have good inhibitory effect to Gaussian noise.It is all to use the embodiment of the present invention, making an uproar in wireless environment can be inhibited Acoustical signal.
In embodiments of the present invention, for each signal to be observed, the accumulation measure feature of the signal to be observed is primarily referred to as The the first accumulation measure feature and the second cumulant obtained according to its second-order cumulant, fourth order cumulant and six rank cumulants is special Sign, therefore, can similarly introduce the cumulant of higher order, obtain more accumulating measure feature, example according to preset cumulative amount formula Such as, eight rank cumulants can be introduced, third accumulation measure feature etc. is calculated.
Step S105, for each signal to be observed, using the energy feature of the signal to be observed and accumulation measure feature as Input data, input SVM classifier trained in advance determine signal to be detected using the categorised decision function in SVM classifier In whether include unmanned plane signal.
In this step, for each signal to be observed, the energy feature of the signal to be observed and accumulation measure feature are made For vector element, corresponding supporting vector is generated, inputs in SVM classifier trained in advance, utilizes the categorised decision of classifier Function determines whether there is unmanned plane signal in signal to be detected.
Specifically, be directed to each signal to be observed, using the energy feature of the signal to be observed and accumulation measure feature as to Secondary element generates a corresponding supporting vector, wherein the corresponding supporting vector of n-th of signal to be observedIt indicates are as follows:
EnIndicate the energy feature of the signal to be observed, γn1Indicate the first accumulation measure feature of the signal to be observed, γn2 Indicate the second accumulation measure feature of the observation signal.
In the SVM classifier that supporting vector is trained in advance as input data, input.
According to the categorised decision function in SVM classifier, the corresponding value of signal to be observed is determined, as decision value, wherein Categorised decision function f (mn) indicate are as follows:
Categorised decision function f (mn) value indicate that the corresponding decision value of n-th of signal to be observed, sign () indicate symbol Function, N indicate the quantity of sample subsignal,Indicate the antithesis of the corresponding Lagrange multiplier of i-th of sample supporting vector Value,<>indicate inner product operation, b*Indicate the value of amount of bias, mn TIndicate the energy feature and cumulant of n-th of signal to be observed The transposition of the corresponding supporting vector of feature, xiIndicate the energy feature of i-th of sample subsignal sample corresponding with accumulation measure feature This supporting vector, yiIndicate the corresponding mark value of i-th of sample subsignal.
Based on the decision value of each signal to be observed, determine in signal to be detected whether include unmanned plane signal, when certainly When at least one value is 1 in plan value, then it represents that include unmanned plane signal in the signal to be detected, when decision value all -1 When, then it represents that not comprising unmanned plane signal in the signal to be detected.
Further, due to each signal to be observed be by being obtained after signal subsection to be detected, it is every by judging Whether include unmanned plane signal in one signal to be observed, can determine in signal to be detected whether include unmanned plane letter Number.Therefore using the energy feature of each signal to be observed and accumulation measure feature as vector element, corresponding support is generated Vector.Using categorised decision function in SVM classifier categorised decision function trained in advance to each signal to be observed into Row classification, since categorised decision function belongs to sign function, output valve only uses 1 and -1, therefore can in SVM classifier training To indicate whether or not there is man-machine signal with 1, -1 indicates do not have unmanned plane signal.When there is unmanned plane signal in signal to be observed, SVM points The decision value of class device should be 1, and when not having unmanned plane signal in signal to be observed, the decision value of SVM classifier should be -1. So, when at least one value is 1 in its decision value, being indicated for all signals to be observed that signal to be detected is included In the signal to be detected include unmanned plane signal, when its decision value it is all -1 when, then it represents that do not wrapped in the signal to be detected Signal containing unmanned plane.
In embodiments of the present invention, the structure of above-mentioned SVM classifier can use existing network structure, for example, feedforward Type network structure.By the way that each signal to be observed to be connect entirely with sample subsignal when training, determine according to above-mentioned classification Parameter-calculation sequence in plan function is realized the identification for treating unmanned plane signal in observation signal, and then is realized to letter to be detected The identification of unmanned plane signal in number.
In conclusion in embodiments of the present invention, for signal to be observed, noise can be inhibited to believe using Higher Order Cumulants Number characteristic, the noise signal treated in observation signal inhibited, and SVM classifier is recycled to identify in signal to be detected Interference signal and unmanned plane signal realize the detection to unmanned plane signal in wireless environment.
In one embodiment in a kind of above-mentioned unmanned plane signal detection method, SVM classifier in above-mentioned steps S105 Training process, as shown in Fig. 2, may comprise steps of:
Step S201 obtains the wireless signal of preset quantity according to preset condition in frequency range to be detected, believes as sample Number.
In this step, the nothing of preset quantity can be obtained in frequency range to be detected in the case where there is unmanned plane signal Line signal, as positive sample signal；In the case where no unmanned plane signal, the sample of preset quantity is obtained in frequency range to be detected This signal, as negative sample signal.
Specifically, can be in above-mentioned target environment, using the presence or absence of unmanned plane simulation unmanned plane signal, and to be detected The positive sample signal and negative sample signal of preset quantity are obtained in frequency range.For example, it is assumed that frequency range to be detected be 2408MHz extremely 2440MHz, in target environment, using unmanned plane send 2410MHz unmanned plane signal, with 30 seconds for time interval, to The wireless signal for detecting 1000 parts of acquisition in frequency range, as positive sample signal；Similarly, under the same conditions, it is sent out without unmanned plane Unmanned plane signal is sent, 1000 parts of wireless signal is obtained, as negative sample signal.
Step S202 divides sample signal, obtains the wireless subsignal in multiple frequency ranges according to predeterminated frequency interval, makees For sample subsignal.
In this step, sample signal use etc. is treated according to the spectrogram of the sample signal for each sample signal Frequency interval divides, and the corresponding wireless subsignal of each frequency range is obtained, as sample subsignal.
Step S203 extracts the energy feature and accumulation measure feature of the sample subsignal for each sample subsignal.
In this step, each sample letter can be determined according to above-mentioned steps S103 and the identical mode of step S104 Number energy feature and accumulation measure feature.
Specifically, being directed to each sample subsignal, it is based on preset energy characteristic formula, the energy for extracting sample subsignal is special Sign, wherein for the sample subsignal y in i-th of frequency rangeiThe energy feature E of ' (t)i' indicate are as follows:
Yi' (f) indicates sample subsignal y in i-th of frequency rangeiThe corresponding frequency-region signal of ' (t), f1Indicate i-th of frequency range pair The lower boundary frequency answered, f2Indicate the corresponding coboundary frequency of i-th of frequency range, Δ f indicates adopting for the frequency in i-th of frequency range Sample interval, t indicate the time, and f indicates frequency.
For each sample subsignal, according to preset cumulative measure feature formula, determine that the first of the sample subsignal is accumulative Measure feature and the second accumulation measure feature, as accumulation measure feature, wherein k-th of sample subsignal y ' (k), the first cumulant Feature γ1' and the second accumulation measure feature γ2' indicate are as follows:
C21' indicate the second-order cumulant of the sample subsignal, C42The fourth order cumulant of ' expression sample subsignal, C63' table Show six rank cumulants of the sample subsignal, C21' in 1 indicate introduce a conjugate complex number, C42' in 2 indicate introduce two Conjugate complex number, C63' in 3 indicate introduce three conjugate complex numbers.
Step S204, the sample signal based on preset quantity, by by the energy feature of sample subsignal in sample signal and As input data, input uses the sample supporting vector and the corresponding mark value of sample subsignal for accumulating measure feature composition The SVM classifier of current class device parameter and preset structure believes the sample of preset quantity using current class decision function Number classify, completes a wheel training.
In step, the sample signal based on preset quantity will be by the energy feature of the sample subsignal in sample signal Corresponding sample supporting vector is generated with accumulation measure feature, by each sample supporting vector and its corresponding sample subsignal Mark value inputted in the SVM classifier with current class device parameter of preset structure as input data, using current Categorised decision function, classify to the sample signal of preset quantity, complete one wheel training.
Specifically, a wheel training of SVM classifier, needs the sample signal using SVM classifier to above-mentioned preset quantity Corresponding all sample subsignals are classified.As an example it is assumed that above-mentioned positive sample signal has 1000, negative sample signal has 1000, that is, the quantity of sample signal is 2000.After step S202 processing, each sample signal, corresponding life It is exactly 20000 sample subsignals at 10 sample subsignals, existing wheel training, is exactly to be completed using categorised decision function Classification to this 20000 sample subsignals.
In embodiments of the present invention, when training for the first time, the current class device parameter of above-mentioned SVM classifier is to preset , the parameter in categorised decision function is determined according to current class device parameter.
Step S205, for each round training, the sample signal based on preset quantity is determined according to default loss function The penalty values of SVM classifier.
In this step, after the completion of each round training, it can be based on hinge loss function, determine that SVM classifier is lost Value, wherein hinge loss function P (x) is indicated are as follows:
The value of hinge loss function P (x) indicates penalty values, and λ indicates parameter to be adjusted, and w indicates divisional plane in SVM classifier Normal vector, | | | | expression takes norm, and N indicates the quantity of sample subsignal, wTIndicate the transposition of w, xiIt indicates by i-th of sample The energy feature of subsignal sample supporting vector corresponding with accumulation measure feature, yiIndicate the corresponding mark of i-th of sample subsignal Note value, b indicate intercept.
Step S206 completes training, determines svm classifier when determining that SVM classifier reaches preset standard based on penalty values The corresponding categorised decision function of device.
In this step, when penalty values are less than or equal to preset threshold, it is believed that SVM classifier reaches preset standard, can To terminate training, the SVM classifier of preset structure is exactly SVM classifier in above-mentioned steps S105 at this time, and current class is determined Plan function is exactly the categorised decision function in above-mentioned steps S105.
Step S207, when determining that SVM classifier is not up to preset standard based on penalty values, according to default adjustment mode, Classifier parameters are adjusted, obtain new categorised decision function, and believe using the sample of new categorised decision function and preset quantity Number, complete new round training.
In this step, when penalty values are greater than preset threshold, it is believed that SVM classifier is not up to preset standard, can be by Above-mentioned current class device parameter is adjusted, is obtained new by calculating above-mentioned hinge loss function to the gradient of λ according to gradient descent method Categorised decision function, and using new categorised decision function and preset quantity sample signal, complete a new round training.
Specifically, the training process of above-mentioned SVM classifier is exactly to classify to sample signal, as shown in figure 3, x1And x1 The characteristic of signal is indicated, for example, the first accumulation measure feature and the second accumulation measure feature, "+" indicate positive sample signal, "-" Indicate that negative sample signal classifies to positive sample signal and negative sample signal using hyperplane in higher dimensional space.
Wherein, hyperplane can indicate are as follows:
wTX+b=0
W indicates the normal vector of hyperplane, wTIndicate that the transposition of w, b indicate intercept.
So, the selection of hyperplane should choose the middle position to positive sample signal and negative sample signal, that is, away from The positive sample signal and negative sample signal farthest from hyperplane can be indicated to the equal of hyperplane are as follows:
s.t.yi(wTxi+ b) >=1, i=1,2 ..., N
For convenience of the solution of hyperplane problem, can be translated into:
s.t.yi(wTxi+ b) >=1, i=1,2 ..., N
When SVM classifier classifies to sample signal, each round training is completed, and needs the value according to above-mentioned parameter to be adjusted, Plane is redefined, therefore, introduces punishment parameter D and slack variable ξ, wherein punishment parameter indicates to adjust letter in optimization direction The weight of number interval and classification accuracy, slack variable indicate that sample signal allows to deviate the variable quantity in the limit of hyperplane.
Wherein, punishment parameter D and slack variable ξ and the relationship of above-mentioned hinge loss function can indicate are as follows:
And hyperplane problem can indicate are as follows:
s.t.yi(wTxi+b)≥1-ξi, i=1,2 ..., N
ξi>=0, i=1,2 ..., N
S.t. constraint condition is indicated, w indicates the normal vector of hyperplane, | | | | expression takes norm, and D indicates punishment parameter, N Indicate the quantity of sample subsignal, ξiIndicate the corresponding slack variable of i-th of sample supporting vector, yi(wTxi+ b) it indicates i-th Spacing distance of the sample supporting vector to divisional plane, yiIndicate the corresponding mark value of i-th of sample subsignal, wTIndicate turning for w It sets, xiIndicate i-th of sample supporting vector, b indicates intercept.
Further, LagrangianL (w, b, ξ, α, μ) can be constructed, can be indicated are as follows:
Wherein, αi>=0, μi>=0, w indicate the normal vector of hyperplane, | | | | expression takes norm, and D indicates punishment parameter, N Indicate the quantity of sample subsignal, ξiIndicate the corresponding slack variable of i-th of sample supporting vector, αiAnd μiIndicate two glugs The value of bright day multiplier, yiIndicate the corresponding mark value of i-th of sample subsignal, wTIndicate the transposition of w, xiIndicate i-th of sample The corresponding sample supporting vector of signal, b indicate intercept, ξiIndicate the corresponding slack variable of i-th of sample subsignal.
Using dual mode, hyperplane problem is converted, can specifically be indicated are as follows:
0≤αi≤ D, i=1,2 ..., N
S.t. constraint condition, α are indicatediIndicate the value of the corresponding Lagrange multiplier of i-th of sample subsignal, αjIndicate jth The value of the corresponding Lagrange multiplier of a sample subsignal, yiIndicate the corresponding mark value of i-th of sample subsignal, yjIndicate i-th The corresponding mark value of a sample subsignal,The transposition of the corresponding sample supporting vector of i-th of sample subsignal, xjIndicate jth The corresponding sample supporting vector of a sample subsignal,<>indicate that inner product operation, D indicate punishment parameter, and N indicates sample subsignal Quantity.
Based on SMO algorithm, the value of Lagrange multiplier is determined.
Using the property of sample supporting vector, the value of amount of bias is determined, wherein the value b of amount of bias*It indicates are as follows:
N indicates the quantity of sample subsignal, ynIndicate the corresponding mark value of n-th of sample supporting vector, yiIt indicates i-th The corresponding mark value of sample supporting vector,Indicate the allelomorph of the corresponding Lagrange multiplier of i-th of sample supporting vector,Indicate the transposition of i-th of sample supporting vector, xnIndicate the corresponding sample supporting vector of n-th of sample subsignal,<>table Show inner product operation.
The value of value and amount of bias based on Lagrange multiplier, redefines categorised decision function.
In embodiments of the present invention, the property of above-mentioned sample supporting vector indicates to be directed to each sample subsignal, sample The decision function mapping value of supporting vector and the product of its mark value are 1.
In conclusion since sample signal only includes positive sample signal and negative sample signal, utilize sample signal pair SVM classifier is trained, and can preferably determine the accuracy of categorised decision function.So using in the embodiment of the present invention Whether there is unmanned plane signal in the determination signal to be detected that SVM classifier can be more accurate.
Using the embodiment of the present invention, the detection to the unmanned plane signal in wireless environment may be implemented, as shown in figure 4, The SVM classifier training stage, by the energy feature and accumulation measure feature of the original signal in the above-mentioned target environment of extraction, and Energy special medical treatment and accumulation measure feature are extracted to unmanned plane signal and original signal, construct training using obtained characteristic is extracted Sample set is trained SVM classifier using training sample set.And in identification process, according to the energy of signal to be detected Feature and accumulation measure feature, utilization trained SVM classifier, can determine in signal to be detected with the presence or absence of unmanned plane Signal.
Based on same inventive concept, according to a kind of unmanned plane signal detection method that the embodiments of the present invention provide, The embodiment of the invention also provides a kind of devices of unmanned plane signal detection, as shown in figure 5, may include:
Signal acquisition module 501 to be detected, for obtaining the wireless signal in frequency range to be detected, as signal to be detected.
Signal acquisition module 502 to be observed, for dividing signal to be detected, obtaining multiple frequencies according to predeterminated frequency interval Wireless subsignal in section, as signal to be observed.
Power feature extraction module 503 is based on preset energy characteristic formula, extracts for being directed to each signal to be observed The energy feature of the signal to be observed.
Cumulant characteristic extracting module 504, for being based on preset cumulative measure feature formula for each signal to be observed, Extract the accumulation measure feature of the signal to be observed.
Unmanned plane signal determining module 505, for being directed to each signal to be observed, by the energy feature of the signal to be observed With accumulation measure feature as input data, input support vector machines classifier trained in advance, using in SVM classifier Categorised decision function determines in signal to be detected whether include unmanned plane signal.
Further, power feature extraction module 503 is specifically used for being directed to each signal to be observed, is based on Pa Saiwaer Theorem calculates the corresponding energy value of the signal to be observed, as energy feature, wherein corresponding to be observed for i-th of frequency range Signal yi(t), energy feature EiIt indicates are as follows:
Yi(f) the signal y to be observed is indicatedi(t) corresponding frequency-region signal, f1Indicate the corresponding lower boundary frequency of i-th of frequency range Rate, f2Indicate the corresponding coboundary frequency of i-th of frequency range, Δ f indicates the frequency sampling interval in i-th of frequency range, when t is indicated Between, f indicates frequency.
Further, above-mentioned cumulant characteristic extracting module 504, comprising:
Cumulant computational submodule, for be directed to each signal to be observed, calculate the signal to be observed second-order cumulant, Fourth order cumulant and six rank cumulants, wherein be directed to each signal to be observed, second-order cumulant C21It can indicate are as follows:
C21=E (| y (k)2|)-|Ey(k)2|
Y (k) indicates that k-th of signal to be observed, E () indicate mathematic expectaion, | | indicate modulus symbol, C21In 1 table Show and introduces a conjugate complex number；
For each signal to be observed, fourth order statistic C42It can indicate are as follows:
C42=E (| y (k)4|)-|Ey(k)2|2-2(E(|y(k)2|))2
Y (k) indicates that k-th of signal to be observed, E () indicate mathematic expectaion, | | indicate modulus symbol, C42In 2 tables Show and introduces two conjugate complex numbers；
For each signal to be observed, cumulant C63It can indicate are as follows:
C63=E (| y (k)6|)-9E(|y(k)|4)E(|y(k)|2)-3E(y*(k)3y(k))E(y(k)2)-3E(y*(k)y (k)3)E(y*(k)2)-18E(y*(k)2)E(y(k)2)E(|y(k)|2)-12(E(|y(k)|2))3
Y (k) indicates that k-th of signal to be observed, E () indicate mathematic expectaion, | | indicate that modulus symbol, * indicate conjugation Symbol, C63In 3 indicate introduce three conjugate complex numbers.
Cumulant feature calculation submodule, it is tired according to the second order of the signal to be observed for being directed to each signal to be observed Accumulated amount, fourth order cumulant and six rank cumulants determine that the first of the signal to be observed is accumulative according to preset cumulative measure feature formula Measure feature and the second accumulation measure feature, as accumulation measure feature, wherein be directed to each signal to be observed, cumulant character representation Are as follows:
γ1Indicate the first accumulation measure feature, γ2Indicate that the second accumulation measure feature, y (k) indicate k-th of signal to be observed, C21Indicate second-order cumulant, C42Indicate fourth order cumulant, C63Indicate six rank cumulants, C21In 1 indicate introduce a conjugate complex Number, C42In 2 indicate introduce two conjugate complex numbers, C63In 3 indicate introduce three conjugate complex numbers.
Further, above-mentioned unmanned plane signal determining module 505, comprising:
Supporting vector generates submodule, for being directed to each signal to be observed, by the energy feature of the signal to be observed and It accumulates measure feature and is used as vector element, one corresponding supporting vector of generation, wherein the corresponding support of n-th of signal to be observed VectorIt indicates are as follows:
EnIndicate the energy feature of the signal to be observed, γn1Indicate the first accumulation measure feature of the signal to be observed, γn2 Indicate the second accumulation measure feature of the observation signal.
Supporting vector input submodule, for using supporting vector as input data, input SVM classifier trained in advance In.
Decision value determines submodule, for determining that signal to be observed is corresponding according to the categorised decision function in SVM classifier Value, as decision value, wherein categorised decision function f (mn) indicate are as follows:
Categorised decision function f (mn) value indicate that the corresponding decision value of n-th of signal to be observed, sign () indicate symbol Function, N indicate the quantity of sample subsignal,Indicate the antithesis of the corresponding Lagrange multiplier of i-th of sample supporting vector Value,<>indicate inner product operation, b*Indicate the value of amount of bias, mn TIndicate the energy feature and cumulant of n-th of signal to be observed The transposition of the corresponding supporting vector of feature, xiIndicate the energy feature of i-th of sample subsignal sample corresponding with accumulation measure feature This supporting vector, yiIndicate the corresponding mark value of i-th of sample subsignal.
Unmanned plane signal determines submodule, for the decision value based on each signal to be observed, determines in signal to be detected It whether include unmanned plane signal, when at least one value is 1 in decision value, then it represents that include nobody in the signal to be detected Machine signal, when decision value it is all -1 when, then it represents that not comprising unmanned plane signal in the signal to be detected.
Further, a kind of unmanned plane signal detecting device that the embodiments of the present invention provide, as shown in fig. 6, may be used also To include:
Sample signal obtains module 601, for obtaining the wireless of preset quantity in frequency range to be detected according to preset condition Signal, as sample signal.
Sample subsignal obtains module 602, for dividing sample signal, obtaining multiple frequency ranges according to predeterminated frequency interval Interior wireless subsignal, as sample subsignal.
Characteristic obtains module 603, for being directed to each sample subsignal, extracts the energy feature of the sample subsignal With accumulation measure feature.
Sample signal categorization module 604 will be believed for the sample signal based on preset quantity by sample in sample signal Number the sample supporting vector that constitutes of energy feature and accumulation measure feature and the corresponding mark value of sample subsignal as input Data, input is using current class device parameter and the SVM classifier of preset structure, using current class decision function, to pre- If the sample signal of quantity is classified, a wheel training is completed, wherein when training for the first time, current class device parameter is default Preliminary classification device parameter, current class decision function are determined according to default preliminary classification device parameter.
Penalty values determining module 605, for being trained for each round, the sample signal based on preset quantity, according to default Loss function determines the penalty values of SVM classifier.
Grader generation module 606, for completing instruction when determining that SVM classifier reaches preset standard based on penalty values Practice, determines the corresponding categorised decision function of SVM classifier.
Classifier parameters adjust module 607, for when determining that SVM classifier is not up to preset standard based on penalty values, According to default adjustment mode, adjust classifier parameters, obtain new categorised decision function, and using new categorised decision function and The sample signal of preset quantity completes new round training.
Further, penalty values determining module 605 is specifically used for after each round training is completed, is lost based on hinge Function determines SVM classifier penalty values, wherein hinge loss function P (x) is indicated are as follows:
The value of hinge loss function P (x) indicates penalty values, and λ indicates parameter to be adjusted, and w indicates divisional plane in SVM classifier Normal vector, | | | | expression takes norm, and N indicates the quantity of sample subsignal, wTIndicate the transposition of w, xiIt indicates by i-th of sample The energy feature of subsignal sample supporting vector corresponding with accumulation measure feature, yiIndicate the corresponding mark of i-th of sample subsignal Note value, b indicate intercept.
Further, above-mentioned classifier parameters adjust module 607, comprising:
Parameter determination submodule to be adjusted, for being based on when determining that SVM classifier is not up to preset standard based on penalty values Hinge loss function determines the value of the parameter to be adjusted in hinge loss function using gradient descent method.
Categorised decision function determines submodule, for the value according to parameter to be adjusted, redefines corresponding point of SVM classifier Class decision function.
Further, above-mentioned categorised decision function determines submodule, further includes:
Punishment parameter determines submodule, for the value according to parameter to be adjusted, determines the punishment parameter of SVM classifier, wherein Punishment parameter indicates to adjust the weight of function interval and accuracy of classifying in optimization direction, punishment parameter D and parameter lambda to be adjusted it Between relationship be expressed as:
Hyperplane determines submodule, for being based on punishment parameter, using convex optimization problem, determines corresponding point of SVM classifier The division mode of face, wherein the division mode expression of divisional plane is divided using sample signal of the hyperplane to preset quantity The method of class, is embodied as:
s.t.yi(wTxi+b)≥1-ξi, i=1,2 ..., N
ξi>=0, i=1,2 ..., N
S.t. constraint condition is indicated, w indicates the normal vector of hyperplane, | | | | expression takes norm, and D indicates punishment parameter, N Indicate the quantity of sample subsignal, ξ i indicates the corresponding slack variable of i-th of sample subsignal, yi(wTxi+ b) indicate i-th of sample This supporting vector arrives the spacing distance of divisional plane, yiIndicate the corresponding mark value of i-th of sample subsignal, wTIndicate the transposition of w, xiIndicate i-th of sample supporting vector, b indicates intercept.
Problem transform subblock converts division mode for being based on Lagrangian and function dualization method Solve problems, division mode indicates after conversion are as follows:
0≤αi≤ D, i=1,2 ..., N
S.t. constraint condition, α are indicatediIndicate the value of the corresponding Lagrange multiplier of i-th of sample subsignal, αjIndicate jth The value of the corresponding Lagrange multiplier of a sample subsignal, yiIndicate the corresponding mark value of i-th of sample subsignal, yjIndicate i-th The corresponding mark value of a sample subsignal,The transposition of the corresponding sample supporting vector of i-th of sample subsignal, xjIndicate jth The corresponding sample supporting vector of a sample subsignal,<>indicate that inner product operation, D indicate punishment parameter, and N indicates sample subsignal Quantity；
Lagrange multiplier determines submodule, for being based on SMO algorithm, determines the value of Lagrange multiplier.
Amount of bias determines submodule, for the property using sample supporting vector, determines the value of amount of bias, wherein biasing The value b of amount*It indicates are as follows:
N indicates the quantity of sample subsignal, ynIndicate the corresponding mark value of n-th of sample supporting vector, yiIt indicates i-th The corresponding mark value of sample supporting vector,Indicate the allelomorph of the corresponding Lagrange multiplier of i-th of sample supporting vector,Indicate the transposition of i-th of sample supporting vector, xnIndicate the corresponding sample supporting vector of n-th of sample subsignal,<>table Show inner product operation；
The value of value and amount of bias based on Lagrange multiplier, redefines categorised decision function.
It should be noted that, in this document, relational terms such as first and second and the like are used merely to a reality Body or operation are distinguished with another entity or operation, are deposited without necessarily requiring or implying between these entities or operation In any actual relationship or order or sequence.Moreover, the terms "include", "comprise" or its any other variant are intended to Non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or equipment Intrinsic element.In the absence of more restrictions, the element limited by sentence " including one ... ", it is not excluded that There is also other identical elements in the process, method, article or apparatus that includes the element.
Each embodiment in this specification is all made of relevant mode and describes, same and similar portion between each embodiment Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for device reality For applying example, since it is substantially similar to the method embodiment, so being described relatively simple, related place is referring to embodiment of the method Part explanation.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the scope of the present invention.It is all Any modification, equivalent replacement, improvement and so within the spirit and principles in the present invention, are all contained in protection scope of the present invention It is interior.

Claims (10)

1. a kind of unmanned plane signal detection method characterized by comprising
The wireless signal in frequency range to be detected is obtained, as signal to be detected；
According to predeterminated frequency interval, the signal to be detected is divided, the wireless subsignal in multiple frequency ranges is obtained, as to be observed Signal；
For each signal to be observed, it is based on preset energy characteristic formula, extracts the energy feature of the signal to be observed；
For each signal to be observed, it is based on preset cumulative measure feature formula, the cumulant for extracting the signal to be observed is special Sign；
For each signal to be observed, using the energy feature of the signal to be observed with the accumulation measure feature as defeated Enter data, inputs support vector machines classifier trained in advance, using the categorised decision function in the SVM classifier, really It whether include unmanned plane signal in the fixed signal to be detected, wherein the SVM classifier is the feature based on sample signal Data training obtains, and the characteristic of the sample signal includes by the energy of sample subsignal each in the sample signal The sample supporting vector and the corresponding mark value of the sample subsignal that feature and accumulation measure feature collectively form, the mark Note value indicates to whether there is unmanned plane signal in the sample subsignal.
2. the method according to claim 1, wherein described be directed to each signal to be observed, based on default Energy feature formula extracts the energy feature of the signal to be observed, comprising:
For each signal to be observed, it is based on Parseval's theorem, calculates the corresponding energy value of the signal to be observed, as Energy feature, wherein be directed to the corresponding signal y to be observed of i-th of frequency rangei(t), the energy feature EiIt indicates are as follows:
Yi(f) the signal y to be observed is indicatedi(t) corresponding frequency-region signal, f1Indicate the corresponding lower boundary frequency of i-th of frequency range, f2 Indicate the corresponding coboundary frequency of i-th of frequency range, Δ f indicates the frequency sampling interval in i-th of frequency range, and t indicates time, f Indicate frequency.
3. the method according to claim 1, wherein described be directed to each signal to be observed, based on default Cumulant characteristic formula extracts the accumulation measure feature of the signal to be observed, comprising:
For each signal to be observed, second-order cumulant, fourth order cumulant and six the ranks accumulation of the signal to be observed are calculated Amount, wherein be directed to each signal to be observed, the second-order cumulant C21It indicates are as follows:
C21=E (| y (k)2|)-|Ey(k)2|
Y (k) indicates that k-th of signal to be observed, E () indicate mathematic expectaion, | | indicate modulus symbol, C21In 1 table Show and introduces a conjugate complex number；
For each signal to be observed, the fourth order statistic C42It indicates are as follows:
C42=E (| y (k)4|)-|Ey(k)2|2-2(E(|y(k)2|))2
Y (k) indicates that k-th of signal to be observed, E () indicate mathematic expectaion, | | indicate modulus symbol, C42In 2 tables Show and introduces two conjugate complex numbers；
For each signal to be observed, the six ranks cumulant C63It indicates are as follows:
C63=E (| y (k)6|)-9E(|y(k)|4)E(|y(k)|2)-3E(y*(k)3y(k))E(y(k)2)
-3E(y*(k)y(k)3)E(y*(k)2)-18E(y*(k)2)E(y(k)2)E(|y(k)|2)
-12(E(|y(k)|2))3
Y (k) indicates that k-th of signal to be observed, E () indicate mathematic expectaion, | | indicate that modulus symbol, * indicate conjugation Symbol, C63In 3 indicate introduce three conjugate complex numbers；
For each signal to be observed, according to the second-order cumulant, fourth order cumulant and six ranks of the signal to be observed Cumulant determines that the first accumulative measure feature of the signal to be observed and the second cumulant are special according to preset cumulative measure feature formula Sign, as accumulation measure feature, wherein be directed to each signal to be observed, the cumulant character representation are as follows:
γ1Indicate the first accumulation measure feature, γ2Indicate the second accumulation measure feature, k-th of y (k) expression described wait see Survey signal, C21Indicate the second-order cumulant, C42Indicate the fourth order cumulant, C63Indicate the six ranks cumulant, C21In 1 indicates to introduce a conjugate complex number, C42In 2 indicate introduce two conjugate complex numbers, C63In 3 indicate introduce three conjugate complexes Number.
4. this is waited seeing the method according to claim 1, wherein described be directed to each signal to be observed The energy feature and the accumulation measure feature for surveying signal are as input data, input support vector machines trained in advance Classifier determines in the signal to be detected whether include unmanned plane using the categorised decision function in the SVM classifier Signal, comprising:
For each signal to be observed, using the energy feature of the signal to be observed and the accumulation measure feature as to Secondary element generates a corresponding supporting vector, wherein the corresponding supporting vector of n-th of signal to be observedIt indicates Are as follows:
EnIndicate the energy feature of the signal to be observed, γn1Indicate the first accumulation measure feature of the signal to be observed, γn2Indicate the second accumulation measure feature of the observation signal；
In the SVM classifier that the supporting vector is trained in advance as input data, input；
According to the categorised decision function in the SVM classifier, the corresponding value of the signal to be observed is determined, as decision value, Wherein, the categorised decision function f (mn) indicate are as follows:
Categorised decision function f (mn) value indicate that the corresponding decision value of n-th of signal to be observed, sign () indicate symbol Function, N indicate the quantity of the sample subsignal,Indicate pair of the corresponding Lagrange multiplier of i-th of sample supporting vector Even value,<>indicate inner product operation, b*Indicate the value of amount of bias, mn TIndicate n-th of signal to be observed energy feature and Accumulate the transposition of the corresponding supporting vector of measure feature, xiIndicate the energy feature and accumulation measure feature pair of i-th of sample subsignal The sample supporting vector answered, yiIndicate the corresponding mark value of i-th of sample subsignal；
Based on the decision value of each signal to be observed, determine in the signal to be detected whether include unmanned plane letter Number, when at least one value is 1 in the decision value, then it represents that include unmanned plane signal in the signal to be detected, when described Decision value it is all -1 when, then it represents that not comprising unmanned plane signal in the signal to be detected.
5. according to the method described in claim 1, the training process of the SVM classifier, comprising:
According to preset condition, the wireless signal of preset quantity is obtained in frequency range to be detected, as sample signal；
According to predeterminated frequency interval, the sample signal is divided, obtains the wireless subsignal in multiple frequency ranges, is believed as sample Number；
For each sample subsignal, the energy feature and accumulation measure feature of the sample subsignal are extracted；
It, will be by the energy feature and cumulant of sample subsignal in the sample signal based on the sample signal of the preset quantity As input data, input uses works as the sample supporting vector and the corresponding mark value of the sample subsignal that feature is constituted The SVM classifier of preceding classifier parameters and preset structure, using current class decision function, to the sample of the preset quantity Signal is classified, and a wheel training is completed, wherein when training for the first time, the current class device parameter is default preliminary classification Device parameter, the current class decision function are determined according to the default preliminary classification device parameter；
For each round training, the SVM points are determined according to default loss function based on the sample signal of the preset quantity The penalty values of class device；
When determining that the SVM classifier reaches preset standard based on the penalty values, training is completed, determines the svm classifier The corresponding categorised decision function of device；
When determining that the SVM classifier is not up to preset standard based on the penalty values, according to default adjustment mode, institute is adjusted Classifier parameters are stated, obtain new categorised decision function, and using the sample of new categorised decision function and the preset quantity Signal completes new round training.
6. according to the method described in claim 5, it is characterized in that, described train for each round, based on the preset quantity Sample signal determine the penalty values of the SVM classifier according to default loss function, comprising:
After each round training is completed, it is based on hinge loss function, determines the SVM classifier penalty values, wherein the hinge Chain loss function P (x) is indicated are as follows:
The value of the hinge loss function P (x) indicates the penalty values, and λ indicates parameter to be adjusted, and w is indicated in the SVM classifier The normal vector of divisional plane, | | | | expression takes norm, and N indicates the quantity of the sample subsignal, wTIndicate the transposition of w, xiIt indicates By the energy feature of i-th of sample subsignal sample supporting vector corresponding with accumulation measure feature, yiIndicate i-th of sample letter Number corresponding mark value, b indicate intercept.
7. according to the method described in claim 5, it is characterized in that, described ought determine the svm classifier based on the penalty values When device is not up to preset standard, according to default adjustment mode, the classifier parameters are adjusted, obtain new categorised decision function, And using the sample signal of new categorised decision function and the preset quantity, new round training is completed, comprising:
When determining that the SVM classifier is not up to preset standard based on the penalty values, it is based on the hinge loss function, benefit With gradient descent method, the value of the parameter to be adjusted in the hinge loss function is determined；
According to the value of the parameter to be adjusted, the corresponding categorised decision function of the SVM classifier is redefined；
Classified according to the categorised decision function redefined to the sample signal of the preset quantity, completes new round instruction Practice.
8. the method according to the description of claim 7 is characterized in that the value of the parameter to be adjusted according to, redefines institute State the corresponding categorised decision function of SVM classifier, comprising:
According to the value of the parameter to be adjusted, the punishment parameter of the SVM classifier is determined, wherein the punishment parameter indicates to adjust The weight of function interval and accuracy of classifying in section optimization direction, the punishment parameter D and the pass wait adjust between parameter lambda System indicates are as follows:
Based on the punishment parameter, using convex optimization problem, determine that the SVM classifier corresponds to the division mode of divisional plane, In, the division mode of the divisional plane indicates the method classified using sample signal of the hyperplane to the preset quantity, It is embodied as:
s.t.yi(wTxi+b)≥1-ξi, i=1,2 ..., N
ξi>=0, i=1,2 ..., N
S.t. constraint condition is indicated, w indicates the normal vector of the hyperplane, | | | | expression takes norm, and D indicates punishment parameter, N Indicate the quantity of the sample subsignal, ξiIndicate the corresponding slack variable of i-th of sample subsignal, yi(wTxi+ b) indicate i-th Spacing distance of a sample supporting vector to divisional plane, yiIndicate the corresponding mark value of i-th of sample subsignal, wTIndicate turning for w It sets, xiIndicate i-th of sample supporting vector, b indicates intercept；
Based on Lagrangian and function dualization method, the Solve problems of the division mode are converted, it is described after conversion Division mode indicates are as follows:
0≤αi≤ D, i=1,2 ..., N
S.t. constraint condition, α are indicatediIndicate the value of the corresponding Lagrange multiplier of i-th of sample subsignal, αjIndicate j-th of sample The value of the corresponding Lagrange multiplier of this subsignal, yiIndicate the corresponding mark value of i-th of sample subsignal, yjIndicate i-th of sample The corresponding mark value of this subsignal,The transposition of the corresponding sample supporting vector of i-th of sample subsignal, xjIndicate j-th of sample The corresponding sample supporting vector of this subsignal,<>indicate that inner product operation, D indicate punishment parameter, and N indicates the number of sample subsignal Amount；
Based on SMO algorithm, the value of Lagrange multiplier is determined；
Using the property of sample supporting vector, the value of amount of bias is determined, wherein the value b of the amount of bias*It indicates are as follows:
N indicates the quantity of sample subsignal, ynIndicate the corresponding mark value of n-th of sample supporting vector, yiIndicate i-th of sample The corresponding mark value of supporting vector,Indicate the allelomorph of the corresponding Lagrange multiplier of i-th of sample supporting vector,Table Show the transposition of i-th of sample supporting vector, xnThe corresponding sample supporting vector of n-th of sample subsignal is indicated, in<>expression Product operation；
The value of value and the amount of bias based on the Lagrange multiplier, redefines categorised decision function.
9. a kind of unmanned plane signal detecting device characterized by comprising
Signal acquisition module to be detected, for obtaining the wireless signal in frequency range to be detected, as signal to be detected；
Signal acquisition module to be observed, for dividing the signal to be detected, obtaining in multiple frequency ranges according to predeterminated frequency interval Wireless subsignal, as signal to be observed；
Power feature extraction module is based on preset energy characteristic formula, extracting should be to for being directed to each signal to be observed The energy feature of observation signal；
Cumulant characteristic extracting module is based on preset cumulative measure feature formula, extracts for being directed to each signal to be observed The accumulation measure feature of the signal to be observed；
Unmanned plane signal determining module, it is for being directed to each signal to be observed, the energy of the signal to be observed is special Sign is with the accumulation measure feature as input data, input SVM classifier trained in advance, using in the SVM classifier Categorised decision function determines in the signal to be detected whether include unmanned plane signal, wherein the SVM classifier is base It is obtained in the characteristic training of sample signal, the characteristic of the sample signal includes by each in the sample signal The sample supporting vector and the sample subsignal that the energy feature and accumulation measure feature of sample subsignal collectively form are corresponding Mark value, the mark value indicates to whether there is unmanned plane signal in the sample subsignal.
10. device according to claim 9, which is characterized in that further include:
Sample signal obtains module, for obtaining the wireless signal of preset quantity in frequency range to be detected according to preset condition, makees For sample signal；
Sample subsignal obtains module, for dividing the sample signal, obtaining in multiple frequency ranges according to predeterminated frequency interval Wireless subsignal, as sample subsignal；
Characteristic obtain module, for be directed to each sample subsignal, extract the sample subsignal energy feature and Accumulate measure feature；
Sample signal categorization module will be by sample in the sample signal for the sample signal based on the preset quantity The sample supporting vector and the corresponding mark value of the sample subsignal that the energy feature and accumulation measure feature of signal are constituted are made For input data, input utilizes current class decision letter using current class device parameter and the SVM classifier of preset structure Number, classifies to the sample signal of the preset quantity, completes a wheel training, wherein when training for the first time, described current point Class device parameter is default preliminary classification device parameter, and the current class decision function is according to the default preliminary classification device parameter Determining；
Penalty values determining module, for being trained for each round, based on the sample signal of the preset quantity, according to default loss Function determines the penalty values of the SVM classifier；
Grader generation module, for completing instruction when determining that the SVM classifier reaches preset standard based on the penalty values Practice, determines the corresponding categorised decision function of the SVM classifier；
Classifier parameters adjust module, for when determining that the SVM classifier is not up to preset standard based on the penalty values, According to default adjustment mode, the classifier parameters are adjusted, obtain new categorised decision function, and use new categorised decision letter Several and the preset quantity sample signal completes new round training.
CN201810885975.5A 2018-08-06 2018-08-06 A kind of unmanned plane signal detection method and device Pending CN109034088A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810885975.5A CN109034088A (en) 2018-08-06 2018-08-06 A kind of unmanned plane signal detection method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810885975.5A CN109034088A (en) 2018-08-06 2018-08-06 A kind of unmanned plane signal detection method and device

Publications (1)

Publication Number Publication Date
CN109034088A true CN109034088A (en) 2018-12-18

Family

ID=64649750

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810885975.5A Pending CN109034088A (en) 2018-08-06 2018-08-06 A kind of unmanned plane signal detection method and device

Country Status (1)

Country Link
CN (1) CN109034088A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110139300A (en) * 2019-05-16 2019-08-16 西安电子科技大学 Sensorcraft method based on the identification of wireless device configuration information
CN110764152A (en) * 2019-10-30 2020-02-07 桂林电子科技大学 Device and method for rapid detection and identification of unmanned aerial vehicle
CN112163569A (en) * 2020-10-29 2021-01-01 上海特金无线技术有限公司 Signal detection method and device, electronic equipment and storage medium

Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101083649A (en) * 2007-07-13 2007-12-05 西安电子科技大学 Method for identifying OFDM modulation system of multi-path Rayleigh fast fading channel
CN102661866A (en) * 2012-05-11 2012-09-12 天津工业大学 Engine fault identification method based on time-domain energy and support vector machine
CN103067325A (en) * 2013-01-31 2013-04-24 南京邮电大学 Cooperative modulation identification method based on multi-class characteristic parameters and evidence theory
CN103259759A (en) * 2013-04-12 2013-08-21 西安电子科技大学 Single channel time-frequency overlap signal modulation identification method
CN104243391A (en) * 2012-07-27 2014-12-24 南京邮电大学 Evidence-discount-based cooperative modulation identification method
CN105469049A (en) * 2015-11-24 2016-04-06 昆明理工大学 Leakage sound emission signal identification method based on multi-scale morphological decomposition energy spectrum entropy and support vector machine
CN105721368A (en) * 2016-01-25 2016-06-29 北京邮电大学 Method and device for identifying overlapped digital signals
CN105718856A (en) * 2016-01-13 2016-06-29 杭州电子科技大学 EEMD-based vehicle micro-tremor signal extraction and classification method
US20170094527A1 (en) * 2015-09-28 2017-03-30 Department 13, LLC Unmanned Aerial Vehicle Intrusion Detection and Countermeasures
CN107682109A (en) * 2017-10-11 2018-02-09 北京航空航天大学 A kind of interference signal classifying identification method suitable for UAV Communication system
CN107748862A (en) * 2017-09-21 2018-03-02 清华大学 A kind of unmanned plane sorting technique and device based on dual-frequency radar signal time-frequency distributions
CN107993648A (en) * 2017-11-27 2018-05-04 北京邮电大学 A kind of unmanned plane recognition methods, device and electronic equipment
CN108090919A (en) * 2018-01-02 2018-05-29 华南理工大学 Improved kernel correlation filtering tracking method based on super-pixel optical flow and adaptive learning factor
CN108182279A (en) * 2018-01-26 2018-06-19 有米科技股份有限公司 Object classification method, device and computer equipment based on text feature

Patent Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101083649A (en) * 2007-07-13 2007-12-05 西安电子科技大学 Method for identifying OFDM modulation system of multi-path Rayleigh fast fading channel
CN102661866A (en) * 2012-05-11 2012-09-12 天津工业大学 Engine fault identification method based on time-domain energy and support vector machine
CN104243391A (en) * 2012-07-27 2014-12-24 南京邮电大学 Evidence-discount-based cooperative modulation identification method
CN103067325A (en) * 2013-01-31 2013-04-24 南京邮电大学 Cooperative modulation identification method based on multi-class characteristic parameters and evidence theory
CN103259759A (en) * 2013-04-12 2013-08-21 西安电子科技大学 Single channel time-frequency overlap signal modulation identification method
US20170094527A1 (en) * 2015-09-28 2017-03-30 Department 13, LLC Unmanned Aerial Vehicle Intrusion Detection and Countermeasures
CN105469049A (en) * 2015-11-24 2016-04-06 昆明理工大学 Leakage sound emission signal identification method based on multi-scale morphological decomposition energy spectrum entropy and support vector machine
CN105718856A (en) * 2016-01-13 2016-06-29 杭州电子科技大学 EEMD-based vehicle micro-tremor signal extraction and classification method
CN105721368A (en) * 2016-01-25 2016-06-29 北京邮电大学 Method and device for identifying overlapped digital signals
CN107748862A (en) * 2017-09-21 2018-03-02 清华大学 A kind of unmanned plane sorting technique and device based on dual-frequency radar signal time-frequency distributions
CN107682109A (en) * 2017-10-11 2018-02-09 北京航空航天大学 A kind of interference signal classifying identification method suitable for UAV Communication system
CN107993648A (en) * 2017-11-27 2018-05-04 北京邮电大学 A kind of unmanned plane recognition methods, device and electronic equipment
CN108090919A (en) * 2018-01-02 2018-05-29 华南理工大学 Improved kernel correlation filtering tracking method based on super-pixel optical flow and adaptive learning factor
CN108182279A (en) * 2018-01-26 2018-06-19 有米科技股份有限公司 Object classification method, device and computer equipment based on text feature

Non-Patent Citations (10)

* Cited by examiner, † Cited by third party
Title
MINGYANG LIU ET AL: "Human Emotion Recognition Based on Galvanic Skin Response Signal Feature Selection and SVM", 《2016 INTERNATIONAL CONFERENCE ON SMART CITY AND SYSTEMS ENGINEERING (ICSCSE)》 *
万建伟，王玲编著: "《信号处理仿真技术》", 31 March 2008 *
冯小平等: "《通信对抗原理》", 31 August 2009 *
吴艇帆等: "机器学习在肺内恶性磨玻璃密度结节的应用研究", 《广州大学学报( 自然科学版)》 *
张传超主编: "制导与控制", 《2015中国无人机系统峰会论文集》 *
曾向阳: "《智能水中目标识别》", 31 March 2016 *
杨东海: "基于声音识别的四旋翼无人机探测技术研究与实现", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 *
白光清: "技术综述", 《光电领域专利审查研究》 *
韩一娜: "《局部多核学习与图形分类》", 28 February 2014 *
黄赛: "复杂电磁环境下高性能频谱感知与分析方法研究", 《中国博士学位论文全文数据库 信息科技辑 》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110139300A (en) * 2019-05-16 2019-08-16 西安电子科技大学 Sensorcraft method based on the identification of wireless device configuration information
CN110139300B (en) * 2019-05-16 2021-05-14 西安电子科技大学 Unmanned aerial vehicle detection method based on wireless device configuration information identification
CN110764152A (en) * 2019-10-30 2020-02-07 桂林电子科技大学 Device and method for rapid detection and identification of unmanned aerial vehicle
CN110764152B (en) * 2019-10-30 2021-06-11 桂林电子科技大学 Device and method for rapid detection and identification of unmanned aerial vehicle
CN112163569A (en) * 2020-10-29 2021-01-01 上海特金无线技术有限公司 Signal detection method and device, electronic equipment and storage medium
CN112163569B (en) * 2020-10-29 2022-11-29 上海特金无线技术有限公司 Signal detection method and device, electronic equipment and storage medium

Similar Documents

Publication Publication Date Title
CN104135327B (en) frequency spectrum sensing method based on support vector machine
CN105678273B (en) The starting point detection algorithm of radio-frequency fingerprint identification technology transient signal
CN103812577B (en) The automatic recognition system and its method of improper radio signal
CN110334591A (en) A kind of detection of unmanned plane Frequency Hopping Signal and recognition methods based on clustering
CN109034088A (en) A kind of unmanned plane signal detection method and device
CN101334434B (en) Electromagnetic environment test system for extracting electromagnetic leakage signal by utilizing wavelet transformation
CN106211084B (en) Environment perception method based on GSM signal
CN106128452A (en) Acoustical signal detection keyboard is utilized to tap the system and method for content
CN109614930A (en) A kind of unmanned plane spectrum detection method based on deep learning
CN108648763B (en) Personal computer usage behavior monitoring method and system based on acoustic channels
CN106936517A (en) A kind of automatic recognition system and its method of abnormal radio signal
CN110048978A (en) A kind of signal modulate method
CN113971437B (en) Cross-domain gesture recognition method based on commercial Wi-Fi equipment
CN109884591A (en) A kind of multi-rotor unmanned aerial vehicle acoustical signal Enhancement Method based on microphone array
CN106991773A (en) A kind of safety-protection system vibration signal recognition based on S-transformation feature extraction
Behura et al. WiST ID—Deep learning-based large scale wireless standard technology identification
CN109512390A (en) Sleep stage method and wearable device based on EEG time domain various dimensions feature and M-WSVM
Tremblay et al. Multiscale community mining in networks using spectral graph wavelets
CN108809463A (en) A kind of multiple reference clock weighting synthetic method
CN109874104A (en) User location localization method, device, equipment and medium
CN112580486A (en) Human behavior classification method based on radar micro-Doppler signal separation
CN102346948B (en) Circumference invasion detection method and system
CN108919241A (en) A kind of underwater signal time-frequency endpoint parameter estimation method based on CFAR detection
CN114757224A (en) Specific radiation source identification method based on continuous learning and combined feature extraction
CN111093163B (en) Passive target classification method based on channel state information

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: 20181218