CN103400159B - Target classification identification method in fast moving scenes and grader acquisition methods - Google Patents
Target classification identification method in fast moving scenes and grader acquisition methods Download PDFInfo
- Publication number
- CN103400159B CN103400159B CN201310338002.7A CN201310338002A CN103400159B CN 103400159 B CN103400159 B CN 103400159B CN 201310338002 A CN201310338002 A CN 201310338002A CN 103400159 B CN103400159 B CN 103400159B
- Authority
- CN
- China
- Prior art keywords
- signal
- cha
- fast moving
- grader
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Radar Systems Or Details Thereof (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The present invention provides the target classification identification method in a kind of fast moving scenes and grader acquisition methods, this recognition methods includes: calculate the species characteristic of target sample under the conditions of the signal to noise ratio dynamic range of target sample in typical case's fast moving scenes and different signal to noise ratios, determine division progression N and the division limits of signal to noise ratio, signal to noise ratio dynamic range is divided into N number of classification interval;Feature extraction and optimization is carried out, it is thus achieved that the characteristic data set of target sample by ranging N number of different interval N number of target sample signal of sorting out respectively;Characteristic data set inputs set grader respectively be trained, it is thus achieved that with N number of classification interval N classifiers parameter one to one, and then obtain N number of object classifiers;The characteristic of echo signal to be sorted is inputted with self belonging to object classifiers corresponding to signal to noise ratio progression is identified, it is thus achieved that classification results.The present invention be effectively increased target from sensor farther out, signal to noise ratio relatively low in the case of discrimination.
Description
Technical field
The invention belongs to mode identification technology, relate to a kind of target classification identification method, particularly relate to a kind of the most mobile
Target classification identification method in scene and grader acquisition methods.
Background technology
PRS is typically by data acquisition, data prediction, feature extraction and selection, and categorised decision four part forms.
At present the achievement in research in this field focuses mostly at feature extraction and selection and categorised decision two parts.Similarly, Research on Target identification
Classification correlative study is also to be primarily upon the selection of Research on Target feature extracting method, the screening of feature and the choosing of categorised decision algorithm
Take and parameter optimization.In recent years, Research on Target identification field has formed the achievement in research of all kinds of algorithm and method, and these are studied into
Fruit also successfully obtains application in the sensing reconnaissance system of battlefield.Also having a few studies is to pay close attention to the data in Research on Target identification
Preprocess method, it mainly uses enhancing or noise reduction strategies, improves the signal to noise ratio of Research on Target signal, carrys out Lifting scheme with this
Identify the performance of genealogical classification identification.
Typically have due to Research on Target the most mobile, therefore the feature of echo signal occurs with the change of observed range partially
Move.The given mode identification system of research is when for Research on Target Classification and Identification at present, is difficult to adapt to the inclined of echo signal feature
Move, to such an extent as to recognition effect is only effective within certain observed range, the less stable of recognition performance.
Summary of the invention
The shortcoming of prior art in view of the above, it is an object of the invention to provide the target classification in a kind of fast moving scenes
Recognition methods and grader acquisition methods, be used for solving existing target identification technology discrimination in fast moving scenes low, stable
Property difference problem.
For achieving the above object and other relevant purposes, the present invention provides the target classification identification method in a kind of fast moving scenes
And grader acquisition methods.
A kind of target classification identification method in fast moving scenes, including:
Calculate signal to noise ratio dynamic range D of target sample in typical case's fast moving scenesSNRWith target sample under the conditions of different signal to noise ratios
This species characteristic DCHA, determine division progression N and the division limits of signal to noise ratio in typical case's fast moving scenes, signal to noise ratio moved
State scope DSNRIt is divided into N number of classification interval, N >=2;
Feature extraction and optimization is carried out, it is thus achieved that typical case by ranging N number of different interval N number of target sample signal of sorting out respectively
The characteristic data set CHA of target sample in fast moving scenes1~CHAN;
By described characteristic data set CHA1~CHANInput set grader respectively to be trained, it is thus achieved that with N number of classification district
Between N classifiers parameter P one to oneCLA-1~PCLA-N, and then obtain target identification in typical case's fast moving scenes
N number of object classifiers CLA1~CLAN;
Characteristic CHA by echo signal X to be sorted in typical case's fast moving scenesXBelieve belonging to input and echo signal X
Make an uproar and the object classifiers more corresponding than progression is identified, it is thus achieved that the classification results of echo signal to be sorted.
Preferably, described set grader includes artificial nerve network classifier, Bayes classifier, Fisher criterion
Grader, nearest neighbour method grader, Fuzzy Classifier, support vector machine classifier, principal component analysis classifier device, nonnegative matrix are divided
Solve grader or gauss hybrid models grader.
Preferably, by described characteristic data set CHA1~CHANInputting the detailed process that set grader is trained respectively is:
By described characteristic data set CHA1~CHANInput set grader respectively, carry out learning and training according to Training strategy;Institute
State Training strategy include error rate greatest gradient descent method, learning rate changing method, self-adaptative adjustment learning rate method containing factor of momentum,
Resilient BP algorithm, Fletcher-Reeve algorithm, Polar-Ribiere algorithm, Scaled Conjugate Gradient Method, yardstick change gradient algorithm,
Class Newton algorithm, BFGS algorithm, OSS algorithm, Levenberg-Marquardt algorithm, Regularization algorithms or
Gradient descent method with momentum.
Preferably, signal to noise ratio dynamic range D of described target sampleSNRIncluding all types of movements in typical case's fast moving scenes
The signal to noise ratio dynamic range of target.
Preferably, species characteristic D of described target sampleCHAIncluding the spectrum signature of target sample signal, energy feature, small echo
One or more in bag feature and zero passage feature.
Preferably, by characteristic CHA of echo signal X to be sorted in typical case's fast moving scenesXInput and echo signal
The detailed process being identified in the object classifiers that signal to noise ratio progression belonging to X is corresponding includes:
Calculate the signal to noise ratio snr of echo signal X to be sorted in typical case's fast moving scenesX;
According to signal to noise ratio snrXJudge the signal to noise ratio progression N belonging to echo signal XX, NXMore than or equal to 1 less than or equal to N;
Echo signal X is carried out feature extraction and optimization, it is thus achieved that characteristic CHA of echo signal XX;
Choose and signal to noise ratio progression NXCorresponding object classifiers CLAX, CLAXFor CLA1~CLANIn one;
By characteristic CHAXInput object classifiers CLAXIt is identified, obtains the classification results of echo signal X.
A kind of acquisition methods of the grader of target classification identification in fast moving scenes, the acquisition methods bag of described grader
Include:
Calculate signal to noise ratio dynamic range D of target sample in typical case's fast moving scenesSNRWith target sample under the conditions of different signal to noise ratios
This species characteristic DCHA, determine division progression N and the division limits of signal to noise ratio in typical case's fast moving scenes, signal to noise ratio moved
State scope DSNRIt is divided into N number of classification interval, N >=2;
Feature extraction and optimization is carried out, it is thus achieved that typical case by ranging N number of different interval N number of target sample signal of sorting out respectively
The characteristic data set CHA of target sample in fast moving scenes1~CHAN;
By described characteristic data set CHA1~CHANInput set grader respectively to be trained, it is thus achieved that with N number of classification district
Between N classifiers parameter P one to oneCLA-1~PCLA-N;, and then obtain target identification in typical case's fast moving scenes
N number of object classifiers CLA1~CLAN。
Preferably, by described characteristic data set CHA1~CHANInput the detailed process that set grader is trained respectively
For: by described characteristic data set CHA1~CHANInput set grader respectively, carry out learning and training according to Training strategy;
Described set grader includes artificial nerve network classifier, Bayes classifier, Fisher criterion grader, neighbour
Method grader, Fuzzy Classifier, support vector machine classifier, principal component analysis classifier device, Non-negative Matrix Factorization grader or
Gauss hybrid models grader;Described Training strategy include error rate greatest gradient descent method, learning rate changing method, containing factor of momentum
Self-adaptative adjustment learning rate method, resilient BP algorithm, Fletcher-Reeve algorithm, Polar-Ribiere algorithm, quantify conjugation
Gradient method, yardstick become gradient algorithm, class Newton algorithm, BFGS algorithm, OSS algorithm, Levenberg-Marquardt
Algorithm, Regularization algorithms or the gradient descent method of band momentum.
Preferably, signal to noise ratio dynamic range D of described target sampleSNRIncluding all types of movements in typical case's fast moving scenes
The signal to noise ratio dynamic range of target.
Preferably, species characteristic D of described target sampleCHAIncluding the spectrum signature of target sample signal, energy feature, small echo
Bag feature is or/and zero passage feature.
As it has been described above, the target classification identification method in fast moving scenes of the present invention and grader acquisition methods, have
Following beneficial effect:
The present invention, by echo signal is pressed signal to noise ratio matched classifier parameter, is effectively increased Classification and Identification in fast moving scenes
The discrimination of system and the stability of recognition performance, especially target from sensor (observation station) farther out, the relatively low situation of signal to noise ratio
Under discrimination be greatly improved.
Accompanying drawing explanation
Fig. 1 is the schematic flow sheet of the target classification identification method in fast moving scenes of the present invention.
Fig. 2 is the flow process signal of target classification identification in the target classification identification method in fast moving scenes of the present invention
Figure.
Fig. 3 is the flow process signal of the acquisition methods of the grader of target classification identification in fast moving scenes of the present invention
Figure.
Detailed description of the invention
Below by way of specific instantiation, embodiments of the present invention being described, those skilled in the art can be by disclosed by this specification
Content understand other advantages and effect of the present invention easily.The present invention can also be added by the most different detailed description of the invention
To implement or application, the every details in this specification can also be based on different viewpoints and application, in the essence without departing from the present invention
Various modification or change is carried out under god.
Refer to accompanying drawing.It should be noted that the diagram provided in the present embodiment illustrates that the present invention's is basic the most in a schematic way
Conception, the most graphic in component count, shape and size time only display with relevant assembly in the present invention rather than is implemented according to reality
Drawing, during its actual enforcement, the kenel of each assembly, quantity and ratio can be a kind of random change, and its assembly layout kenel is also
It is likely more complexity.
The technical problem to be solved is: under conditions of target quickly moves, and echo signal dynamically becomes with observed range
Changing, signal characteristic also offsets with observed range, uses given categorisation device, system identification performance can be made with the change of observed range
Change and change, the problem of recognition performance poor stability.
The present invention pays close attention to the feature that target quickly moves, and utilizes the characteristic that echo signal feature dynamically changes with observed range, it is provided that
Target classification identification method in a kind of fast moving scenes, the method is that target of based on signal to noise ratio matched classifier parameter is known
Other sorting technique, can be effectively improved the discrimination of target in fast moving scenes and the stability of identification.The present invention can also answer
In battlefield, it is possible to being effectively improved the degree of accuracy of the Classification and Identification of target, i.e. fast moving scenes in battlefield is battlefield, mesh
The mobile operation main body such as the personnel that are designated as in battlefield, vehicle, aircraft.
Below in conjunction with embodiment and accompanying drawing, the present invention is described in detail.
Embodiment one
The present embodiment provides the target classification identification method in a kind of fast moving scenes, as it is shown in figure 1, this fast moving scenes
In target classification identification method include: grader obtain with two parts of target classification identification.
Specifically, grader obtains and comprises the following steps:
Calculate signal to noise ratio dynamic range D of target sample in typical case's fast moving scenesSNRWith target sample under the conditions of different signal to noise ratios
This species characteristic DCHA, determine division progression N and the division limits of signal to noise ratio in typical case's fast moving scenes, signal to noise ratio moved
State scope DSNRIt is divided into N number of classification interval.
Further, signal to noise ratio dynamic range D of described target sampleSNRIncluding all types of movements in typical case's fast moving scenes
The signal to noise ratio dynamic range of target.Species characteristic D of described target sampleCHASpectrum signature, energy including target sample signal
One or more in feature, wavelet packet character and zero passage feature.
Wherein, the computational methods of described signal to noise ratio can be existing any means, and the present embodiment only provides the meter of a kind of signal to noise ratio
Calculate embodiment, but this kind of signal-to-noise ratio computation mode that protection scope of the present invention is not limited to enumerate.
As supposed: burst X1..., XKOn the premise of zero-mean, the energy balane formula of burst is
Assume that signals collecting site environment end energy of making an uproar is En, then the computational methods of signal to noise ratio are
Signal to noise ratio divides the determination principle of progression N and concrete division limits: under the conditions of different signal to noise ratios, in fast moving scenes
Species characteristic D of target sampleCHAMeet characteristic value difference to try one's best big principle, ensure that being divided progression N by signal to noise ratio causes simultaneously
The costs such as the increase of computation complexity and system power dissipation are within the acceptable range.Wherein, N >=2.
Feature extraction and optimization is carried out, it is thus achieved that typical case by ranging N number of different interval N number of target sample signal of sorting out respectively
The characteristic data set CHA of target sample in fast moving scenes1~CHAN。
By described characteristic data set CHA1~CHANInput set grader respectively to be trained, it is thus achieved that with N number of classification district
Between N classifiers parameter P one to oneCLA-1~PCLA-N;, and then obtain target identification in typical case's fast moving scenes
N number of object classifiers CLA1~CLAN。
Further, described set grader includes but not limited to artificial nerve network classifier, Bayes classifier, Fisher
Criterion grader, nearest neighbour method grader, Fuzzy Classifier, support vector machine classifier, principal component analysis classifier device, non-
Negative matrix decomposes grader, gauss hybrid models grader etc..Described support vector machine classifier includes classification based on kernel function
Device.The protection domain of grader of the present invention is not limited to the several types that the present embodiment is enumerated, and every is capable of the present invention
Grader be included in protection scope of the present invention.
Described characteristic data set CHA1~CHAN inputs detailed process that set grader is trained respectively is: by institute
State characteristic data set CHA1~CHAN and input set grader respectively, carry out learning and training according to Training strategy;Described
Training strategy include but not limited to error rate greatest gradient descent method, learning rate changing method, containing factor of momentum self-adaptative adjustment study
Rate method, resilient BP algorithm, Fletcher-Reeve algorithm, Polar-Ribiere algorithm, Scaled Conjugate Gradient Method, yardstick become ladder
Degree algorithm, class Newton algorithm, BFGS algorithm, OSS algorithm, Levenberg-Marquardt algorithm, Bayesian regularization
Algorithm or the gradient descent method etc. of band momentum.The protection domain of Training strategy of the present invention is not limited to what the present embodiment was enumerated
Several types, the Training strategy of every present invention of being capable of is included in protection scope of the present invention.
Target classification identification comprises the following steps:
Characteristic CHA by echo signal X to be sorted in typical case's fast moving scenesXBelieve belonging to input and echo signal X
Make an uproar and the object classifiers more corresponding than progression is identified, it is thus achieved that the classification results of echo signal to be sorted.Implement process
As in figure 2 it is shown, include:
Calculate the signal to noise ratio snr of echo signal X to be sorted in typical case's fast moving scenesX;
According to signal to noise ratio snrXJudge the signal to noise ratio progression N belonging to echo signal XX, NXMore than or equal to 1 less than or equal to N;
Echo signal X is carried out feature extraction and optimization, it is thus achieved that characteristic CHA of echo signal XX;
Choose and signal to noise ratio progression NXCorresponding object classifiers CLAX, CLAXFor CLAA~CLANIn one;
By characteristic CHAXInput object classifiers CLAXIt is identified, obtains the classification results of echo signal X.
The present embodiment also provides for a kind of acquisition methods of the grader of target classification identification in fast moving scenes, such as Fig. 3 institute
Showing, the acquisition methods of described grader includes:
Calculate signal to noise ratio dynamic range D of target sample in typical case's fast moving scenesSNRWith target sample under the conditions of different signal to noise ratios
This species characteristic DCHA, determine division progression N and the division limits of signal to noise ratio in typical case's fast moving scenes, signal to noise ratio moved
State scope DSNRIt is divided into N number of classification interval.
Further, signal to noise ratio dynamic range D of described target sampleSNRIncluding all types of movements in typical case's fast moving scenes
The signal to noise ratio dynamic range of target.Species characteristic D of described target sampleCHAInclude but not limited to that the frequency spectrum of target sample signal is special
Levy, energy feature, wavelet packet character be or/and zero passage feature.The parameter of species characteristic DCHA characterizing described target sample can
A lot of to have, protection scope of the present invention is not limited to several characteristic parameters that the present embodiment is enumerated.
Wherein, the computational methods of described signal to noise ratio can be existing any means, and the present embodiment only provides the meter of a kind of signal to noise ratio
Calculate embodiment, but this kind of signal-to-noise ratio computation mode that protection scope of the present invention is not limited to enumerate.
As supposed: burst X1..., XKOn the premise of zero-mean, the energy balane formula of burst is
Assume that signals collecting site environment end energy of making an uproar is En, then the computational methods of signal to noise ratio are
Signal to noise ratio divides the determination principle of progression N and concrete division limits: under the conditions of different signal to noise ratios, in fast moving scenes
Species characteristic D of target sampleCHAMeet characteristic value difference to try one's best big principle, ensure that being divided progression N by signal to noise ratio causes simultaneously
The costs such as the increase of computation complexity and system power dissipation are within the acceptable range.Wherein, N >=2.
Feature extraction and optimization is carried out, it is thus achieved that typical case by ranging N number of different interval N number of target sample signal of sorting out respectively
The characteristic data set CHA of target sample in fast moving scenes1~CHAN。
By described characteristic data set CHA1~CHANInput set grader respectively to be trained, it is thus achieved that with N number of classification district
Between N classifiers parameter P one to oneCLA-1~PCLA-N;, and then obtain target identification in typical case's fast moving scenes
N number of object classifiers CLA1~CLAN。
Further, described set grader includes but not limited to artificial nerve network classifier, Bayes classifier, Fisher
Criterion grader, nearest neighbour method grader, Fuzzy Classifier, support vector machine classifier, principal component analysis classifier device, non-
Negative matrix decomposes grader or gauss hybrid models grader.The protection domain of grader of the present invention is not limited to this enforcement
The several types that example is enumerated, the grader of every present invention of being capable of is included in protection scope of the present invention.
Described characteristic data set CHA1~CHAN inputs detailed process that set grader is trained respectively is: by institute
State characteristic data set CHA1~CHAN and input set grader respectively, carry out learning and training according to Training strategy;Described
Training strategy include but not limited to error rate greatest gradient descent method, learning rate changing method, containing factor of momentum self-adaptative adjustment study
Rate method, resilient BP algorithm, Fletcher-Reeve algorithm, Polar-Ribiere algorithm, Scaled Conjugate Gradient Method, yardstick become ladder
Degree algorithm, class Newton algorithm, BFGS algorithm, OSS algorithm, Levenberg-Marquardt algorithm, Bayesian regularization
Algorithm or the gradient descent method etc. of band momentum.The protection domain of Training strategy of the present invention is not limited to what the present embodiment was enumerated
Several types, the Training strategy of every present invention of being capable of is included in protection scope of the present invention.
The present invention, by echo signal is pressed signal to noise ratio matched classifier parameter, is effectively increased Classification and Identification in fast moving scenes
The discrimination of system and the stability of recognition performance, especially target from sensor (observation station) farther out, the relatively low situation of signal to noise ratio
Under discrimination be greatly improved.
Embodiment two
The present embodiment provides the specific implementation process of the target classification identification method in a kind of fast moving scenes, wherein, quickly moves
Dynamic scene is battlefield, and the target type in battlefield includes three classes such as wheeled vehicle, endless-track vehicle, low flyer, to be sorted
Research on Target signal is the low frequency audible signal that target travel produces, and uses the 64 dimension FFT features spy as echo signal of signal
Levy data, namely the input feature value as grader.
Assuming that sensor site of deployment environment is made an uproar, end energy is 18000, the sequence length K=1024 of echo signal, burst
X1..., XKFor under zero-mean premise, the energy method computations of burst is defined asSignal-noise ratio computation method is fixed
Justice isAssuming that select the grader that artificial neural network is classified as Research on Target identification.
The specific implementation process of the target classification identification in the battlefield described in the present embodiment includes:
1, grader obtain (i.e. classifier parameters training) to be embodied as step as follows:
Calculate signal to noise ratio dynamic range D of target sample signal in typical case battlefieldSNR, it is assumed that three kinds of typical Research on Target sample signals
The signal to noise ratio dynamic range of (i.e. wheeled vehicle 1, endless-track vehicle 2,3 three kinds of echo signals of low flyer) is respectively DSNR1=[0,
180], DSNR2=[0,200], DSNR3=[0,260];
Species characteristic D of target sample signal in typical battlefield under the conditions of calculating different signal to noise ratioCHA, it is assumed that result of calculation is respectively
Wheeled vehicle at SNR ranges [55,180], endless-track vehicle at SNR ranges [70,200], low flyer in signal to noise ratio
Scope [70,260], target sample species characteristic D under conditions of the SNR ranges interval that self is affiliatedCHAWith it non-
Species characteristic D under conditions of the SNR ranges interval that self is affiliatedCHAThere is between ' significant change, i.e. DCHANotable big
In DCHA', say, that there is obvious characteristic offset in different signal to noise ratio intervals in the species characteristic of target sample signal.
Owing to the coexisting region of the SNR ranges of three kinds of typical Research on Target signals is [0,180], it is thus determined that signal to noise ratio divides
Progression N=3, concrete division limits is A=[0,55], B=[55,70] and C=[70,260].
In the present embodiment, the determination principle of division limits is:
A) all boundaries should comprise the subregion of coexisting region of SNR ranges of typical case's Research on Target signal, i.e. A, B,
C all comprises subregion in [0,180], and it is such should not to occur being similar to [190,210], does not comprises any district in [0,180]
The boundary in territory;
B) limitary value intersection should completely include [0,260] of SNR ranges of typical case's Research on Target signal, i.e. A,
It is region-wide that B, C boundary should cover [0,260] altogether, and should there is not A, B, C boundary and the most still cannot cover the whole district
The situation in territory;
C) border of boundary should be the maximum of SNR ranges of typical case's Research on Target signal or minimum data point, such as [0,260]
0 and 260, or target signature occurs the signal to noise ratio turning point of substantially skew, such as wheeled vehicle target signature to occur substantially
The signal to noise ratio turning point of skew is 55 and 180, and endless-track vehicle target signature occurs the signal to noise ratio turning point of substantially skew to be 70 Hes
200, low flyer target signature occurs the signal to noise ratio turning point of substantially skew to be 70 and 260, but be because 180,200,
260, all beyond the coexisting region [0,180] of SNR ranges of typical case's Research on Target signal, the most only take 55 and 70 two turnovers
Point.
To sum up, determine in A, B, C boundary and be: A=[0,55], B=[55,70] and C=[70,260].
Calculate the signal to noise ratio of test sample signal, according to the signal to noise ratio of test sample signal, this test sample sorted out, it is judged that
Which kind of in A, B, C be this test sample belong to.
To be respectively belonging to A, B, C 33 test sample signals of class carry out FFT feature extraction and characteristic optimization, obtain
The characteristic data set CHA of three quasi-representative Research on Target samplesA, CHAB, CHAC;
By characteristic data set CHAA, CHAB, CHACInput set artificial nerve network classifier respectively, according to error rate
Greatest gradient descent method Training strategy carries out artificial neural network study and training, obtains dividing 3 corresponding component classes with signal to noise ratio
Device parameter PCLA-A, PCLA-B, PCLA-C, and then form 3 the Research on Target grader CLAs corresponding with 3 SNR rangesA,
CLAB, CLAC。
2, Research on Target Classification and Identification to be embodied as step as follows:
Calculate signal to noise ratio X of Research on Target signal X to be sortedSNR, it is assumed that for XSNR=115;
Judge the signal to noise ratio progression N belonging to Research on Target signal XX=C;
Battlefield echo signal X is carried out FFT feature extraction and optimization, obtains characteristic CHA of Research on Target signal XX;
Choose with Research on Target signal X belonging to corresponding for signal to noise ratio progression C grader CLAC, by the spy of Research on Target signal X
Levy data CHAXInput grader CLACCarry out Classification and Identification, obtain the classification results T of Research on Target signalX。
Traditional classification recognition methods and classifying identification method of the present invention are compared by the present embodiment, to pan soil road, sandstone
At road, meadow three, in scene, collection low flyer, wheeled vehicle, 3 kinds of Research on Target signals of endless-track vehicle emulate, system
Counting 3 kinds of landform every classification target correct recognition ratas of two kinds of methods, recognition result is as shown in Table 1 and Table 2.
Table 1: the simulation result statistical form of traditional classification recognition methods
Table 2: the simulation result statistical form of classifying identification method of the present invention
By the contrast of Tables 1 and 2 it is seen that, the present invention by typical case's Research on Target signal by signal to noise ratio matched classifier
The method of parameter, can be effectively improved the discrimination of battlefield Classification and Identification and the stability of recognition performance, especially Research on Target from
Sensor farther out, signal to noise ratio relatively low in the case of discrimination.
The present invention can improve the discrimination of target classification identification in fast moving scenes, the especially signal to noise ratio of target sample signal
Dynamic range is relatively big, the species characteristic of target sample changes the discrimination of target classification identification in obvious scene with signal to noise ratio.
In sum, the present invention effectively overcomes various shortcoming of the prior art and has high industrial utilization.
The principle of above-described embodiment only illustrative present invention and effect thereof, not for limiting the present invention.Any it is familiar with this skill
Above-described embodiment all can be modified under the spirit and the scope of the present invention or change by the personage of art.Therefore, such as
All that in art, tool usually intellectual is completed under without departing from disclosed spirit and technological thought etc.
Effect is modified or changes, and must be contained by the claim of the present invention.
Claims (10)
1. the target classification identification method in a fast moving scenes, it is characterised in that the target classification in described fast moving scenes
Recognition methods includes:
Calculate signal to noise ratio dynamic range D of target sample in typical case's fast moving scenesSNRWith target under the conditions of different signal to noise ratios
Species characteristic D of sampleCHA, determine division progression N and the division limits of signal to noise ratio in typical case's fast moving scenes, will letter
Make an uproar ratio dynamic range DSNRIt is divided into N number of classification interval, N >=2;
Feature extraction and optimization is carried out by ranging N number of different interval N number of target sample signal of sorting out respectively, it is thus achieved that
The characteristic data set CHA of target sample in typical case's fast moving scenes1~CHAN;
By described characteristic data set CHA1~CHANInput set grader respectively to be trained, it is thus achieved that with N number of classification
Interval N classifiers parameter P one to oneCLA-1~PCLA-N, and then obtain target in typical case's fast moving scenes
The N number of object classifiers CLA identified1~CLAN;
Characteristic CHA by echo signal X to be sorted in typical case's fast moving scenesXInput and echo signal X institute
Belong in the object classifiers that signal to noise ratio progression is corresponding and being identified, it is thus achieved that the classification results of echo signal to be sorted.
Target classification identification method in fast moving scenes the most according to claim 1, it is characterised in that: described set dividing
Class device include artificial nerve network classifier, Bayes classifier, Fisher criterion grader, nearest neighbour method grader,
Fuzzy Classifier, support vector machine classifier, principal component analysis classifier device, Non-negative Matrix Factorization grader or Gaussian Mixture mould
Type grader.
Target classification identification method in fast moving scenes the most according to claim 2, it is characterised in that by described characteristic
According to collection CHA1~CHANInputting the detailed process that set grader is trained respectively is:
By described characteristic data set CHA1~CHANInput set grader respectively, carry out artificial god according to Training strategy
Through e-learning and training;Described Training strategy include error rate greatest gradient descent method, learning rate changing method, containing factor of momentum
Self-adaptative adjustment learning rate method, resilient BP algorithm, Fletcher-Reeve algorithm, Polar-Ribiere algorithm, quantify altogether
Yoke gradient method, yardstick become gradient algorithm, class Newton algorithm, BFGS algorithm, OSS algorithm, Levenberg-Marquardt
Algorithm, Regularization algorithms or the gradient descent method of band momentum.
Target classification identification method in fast moving scenes the most according to claim 1, it is characterised in that: described target sample
Signal to noise ratio dynamic range DSNRIncluding the dynamic model of signal to noise ratio of all types of mobile targets in typical case's fast moving scenes
Enclose.
Target classification identification method in fast moving scenes the most according to claim 1, it is characterised in that: described target sample
Species characteristic DCHAIncluding the spectrum signature of target sample signal, energy feature, wavelet packet character or/and zero passage feature.
Target classification identification method in fast moving scenes the most according to claim 1, it is characterised in that typical case is quickly moved
Characteristic CHA of echo signal X to be sorted in dynamic sceneXInput with echo signal X belonging to signal to noise ratio progression corresponding
Object classifiers in the detailed process that is identified include:
Calculate the signal to noise ratio snr of echo signal X to be sorted in typical case's fast moving scenesX;
According to signal to noise ratio snrXJudge the signal to noise ratio progression N belonging to echo signal XX, NXMore than or equal to 1 less than or equal to N;
Echo signal X is carried out feature extraction and optimization, it is thus achieved that characteristic CHA of echo signal XX;
Choose and signal to noise ratio progression NXCorresponding object classifiers CLAX, CLAXFor CLA1~CLANIn one;
By characteristic CHAXInput object classifiers CLAXIt is identified, obtains the classification results of echo signal X.
7. the acquisition methods of the grader of target classification identification in fast moving scenes, it is characterised in that described grader
Acquisition methods includes:
Calculate signal to noise ratio dynamic range D of target sample in typical case's fast moving scenesSNRWith target under the conditions of different signal to noise ratios
Species characteristic D of sampleCHA, determine division progression N and the division limits of signal to noise ratio in typical case's fast moving scenes, will letter
Make an uproar ratio dynamic range DSNRIt is divided into N number of classification interval, N >=2;
Feature extraction and optimization is carried out by ranging N number of different interval N number of target sample signal of sorting out respectively, it is thus achieved that
The characteristic data set CHA of target sample in typical case's fast moving scenes1~CHAN;
By described characteristic data set CHA1~CHANInput set grader respectively to be trained, it is thus achieved that with N number of classification
Interval N classifiers parameter P one to oneCLA-1~PCLA-N, and then obtain target in typical case's fast moving scenes
The N number of object classifiers CLA identifiedA~CLAN。
The acquisition methods of the grader of target classification identification in fast moving scenes the most according to claim 7, its feature exists
In, by described characteristic data set CHA1~CHANInputting the detailed process that set grader is trained respectively is:
By described characteristic data set CHA1~CHANInput set grader respectively, according to Training strategy carry out study with
Training;
Described set grader include artificial nerve network classifier, Bayes classifier, Fisher criterion grader,
Nearest neighbour method grader, Fuzzy Classifier, support vector machine classifier, principal component analysis classifier device, Non-negative Matrix Factorization are classified
Device or gauss hybrid models grader;
Described Training strategy includes error rate greatest gradient descent method, learning rate changing method, self-adaptative adjustment containing factor of momentum
Habit rate method, resilient BP algorithm, Fletcher-Reeve algorithm, Polar-Ribiere algorithm, Scaled Conjugate Gradient Method, yardstick
Become gradient algorithm, class Newton algorithm, BFGS algorithm, OSS algorithm, Levenberg-Marquardt algorithm, pattra leaves
This regularization algorithm or the gradient descent method of band momentum.
The acquisition methods of the grader of target classification identification in fast moving scenes the most according to claim 7, its feature exists
In, signal to noise ratio dynamic range D of described target sampleSNRIncluding all types of mobile targets in typical case's fast moving scenes
Signal to noise ratio dynamic range.
The acquisition methods of the grader of target classification identification in fast moving scenes the most according to claim 7, it is special
Levy and be, species characteristic D of described target sampleCHAIncluding the spectrum signature of target sample signal, energy feature, small echo
Bag feature is or/and zero passage feature.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310338002.7A CN103400159B (en) | 2013-08-05 | 2013-08-05 | Target classification identification method in fast moving scenes and grader acquisition methods |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310338002.7A CN103400159B (en) | 2013-08-05 | 2013-08-05 | Target classification identification method in fast moving scenes and grader acquisition methods |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103400159A CN103400159A (en) | 2013-11-20 |
CN103400159B true CN103400159B (en) | 2016-09-07 |
Family
ID=49563777
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310338002.7A Active CN103400159B (en) | 2013-08-05 | 2013-08-05 | Target classification identification method in fast moving scenes and grader acquisition methods |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103400159B (en) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107015228A (en) * | 2016-11-15 | 2017-08-04 | 财团法人车辆研究测试中心 | The Environment identification system of automobile-used millimetre-wave radar |
CN110321951B (en) * | 2019-07-01 | 2021-03-16 | 青岛海科虚拟现实研究院 | VR simulated aircraft training evaluation method |
WO2022021083A1 (en) * | 2020-07-28 | 2022-02-03 | 深圳市大疆创新科技有限公司 | Image processing method, image processing device, and computer readable storage medium |
CN115168345B (en) * | 2022-06-27 | 2023-04-18 | 天翼爱音乐文化科技有限公司 | Database classification method, system, device and storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101147160A (en) * | 2005-04-01 | 2008-03-19 | 英国电讯有限公司 | Adaptive classifier, and method of creation of classification parameters therefor |
CN102799862A (en) * | 2012-06-29 | 2012-11-28 | 陕西省交通规划设计研究院 | System and method for pedestrian rapid positioning and event detection based on high definition video monitor image |
CN102902951A (en) * | 2012-06-29 | 2013-01-30 | 陕西省交通规划设计研究院 | System and method for vehicle target location and event detection on basis of high-definition video monitoring images |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2012054830A1 (en) * | 2010-10-21 | 2012-04-26 | SET Corporation | Method and system of video object tracking |
-
2013
- 2013-08-05 CN CN201310338002.7A patent/CN103400159B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101147160A (en) * | 2005-04-01 | 2008-03-19 | 英国电讯有限公司 | Adaptive classifier, and method of creation of classification parameters therefor |
CN102799862A (en) * | 2012-06-29 | 2012-11-28 | 陕西省交通规划设计研究院 | System and method for pedestrian rapid positioning and event detection based on high definition video monitor image |
CN102902951A (en) * | 2012-06-29 | 2013-01-30 | 陕西省交通规划设计研究院 | System and method for vehicle target location and event detection on basis of high-definition video monitoring images |
Non-Patent Citations (1)
Title |
---|
一种基于支持向量机的无源雷达目标识别方法;曾卫民等;《现代防御技术》;20100228;第38卷(第1期);89-93,108 * |
Also Published As
Publication number | Publication date |
---|---|
CN103400159A (en) | 2013-11-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103810503B (en) | Depth study based method for detecting salient regions in natural image | |
CN104408481B (en) | Classification of Polarimetric SAR Image method based on depth wavelet neural network | |
CN103400159B (en) | Target classification identification method in fast moving scenes and grader acquisition methods | |
CN103258213B (en) | A kind of for the dynamic vehicle model recognizing method in intelligent transportation system | |
CN108062764A (en) | A kind of object tracking methods of view-based access control model | |
CN105320966A (en) | Vehicle driving state recognition method and apparatus | |
CN105389480B (en) | Multiclass imbalance genomics data iteration Ensemble feature selection method and system | |
CN108459311A (en) | Micro-doppler objective classification method under convolutional neural networks based on Hough transform | |
CN103473776B (en) | A kind of method, system and autofocus control method of movement images definition | |
CN103824088B (en) | SAR target variant recognition method based on multi-information joint dynamic sparse representation | |
CN102147866A (en) | Target identification method based on training Adaboost and support vector machine | |
CN103218831A (en) | Video moving target classification and identification method based on outline constraint | |
CN105930794A (en) | Indoor scene identification method based on cloud computing | |
CN106056101A (en) | Non-maximum suppression method for face detection | |
CN103985119A (en) | Method for partitioning cytoplasm and cell nucleuses of white blood cells in color blood cell image | |
CN105913090B (en) | SAR image objective classification method based on SDAE-SVM | |
CN103853922B (en) | A kind of visualization method for quantitatively evaluating of oil gas field similitude | |
CN104182985A (en) | Remote sensing image change detection method | |
CN109492596A (en) | A kind of pedestrian detection method and system based on K-means cluster and region recommendation network | |
CN105913083A (en) | Dense SAR-SIFT and sparse coding-based SAR classification method | |
CN105976376A (en) | High resolution SAR image target detection method based on part model | |
CN103994820B (en) | A kind of moving target recognition methods based on micropore diameter microphone array | |
CN109800713A (en) | The remote sensing images cloud detection method of optic increased based on region | |
CN101251896A (en) | Object detecting system and method based on multiple classifiers | |
CN103700118B (en) | Based on the moving target detection method of pulse coupled neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |