CN106772307A - A kind of true and false bullet target identification method based on many radar informations - Google Patents

Abstract

The invention belongs to radar target identification method technical field, it is related to a kind of true and false bullet target identification method based on many radar informations.The difference of the feature that the present invention is obtained according to true and false bullet target under different radar-probing systems, and two classifications are marked on the difference of the intrinsic characteristic variations gone out embodied in flight course；8 kinds of features such as target RCS extracting data target RCS variances, range, the coefficient of variation for being obtained from Narrow-band Radar first；5 kinds of features such as target scattering point number, length are extracted in the one-dimensional picture echo data of target for being obtained from wideband radar again；13 kinds of respective target sample distributions of feature are then counted respectively；Most independent recognition result of the test sample in 13 kinds of features carries out comprehensive distinguishing output at last, realizes the true and false bullet target identification of many radar informations.

Description

A kind of true and false bullet target identification method based on many radar informations

Technical field

The invention belongs to radar target identification method technical field, relate to the use of many radar detection results to shell class target The method being identified, more particularly to true and false bullet classification target recognition methods.

Background technology

In modern battlefield, ballistic missile is the deterrent weapon with extremely strong long-range strike ability, and it is in flight course In can discharge a large amount of bait targets enemy's system of defense is disturbed.Therefore, it is possible to realize target flight stage casing from True bullet target is correctly identified in a large amount of baits, is the precondition that system of defense is implemented effectively to intercept.

At present, Radar Technology is directed to far field objects and is detected and recognized one of maximally efficient technological means.Due to The target information amount that low-resolution radar is obtained is few, and high resolution radar detection range is not enough, therefore, from actual demand, profit With the advantage of two class radars, there is weight specifically designed for many radar signature integrated recognition methods of true and false bullet target identification Study on Problems Want military significance.

The content of the invention

It is an object of the invention to be based on the target various information data that active service Narrow-band Radar and wideband radar are obtained, there is provided A kind of true and false bullet based target integrated identification method based on many radar informations.

The technical scheme is that：

The difference of the feature that the present invention is obtained according to true and false bullet target under different radar-probing systems, and two classifications It is marked on the difference of the intrinsic characteristic variations gone out embodied in flight course；The target RCS data for being obtained from Narrow-band Radar first 8 kinds of features such as middle extraction target RCS variances, range, the coefficient of variation；The one-dimensional picture echo of target for being obtained from wideband radar again 5 kinds of features such as extracting data target scattering point number, length；13 kinds of respective target samples of feature point are then counted respectively Cloth；Most independent recognition result of the test sample in 13 kinds of features carries out comprehensive distinguishing output at last, realizes many radar informations True and false bullet target identification.

As shown in figure 1, a kind of true and false bullet target identification method based on many radar informations, it is characterised in that including with Lower step：

S1, the RCS values by the Narrow-band Radar true bullet of detection acquisition and bait target in flight course, will be per continuous n The value of individual sampling instant constitutes a RCS sequence；The RCS sequences are randomly divided into training dataset and test data set；

S2, the RCS sequences obtained according to S1, each sequence to each target extract the arrowband statistics spy of target respectively Levy, including RCS sequences averageSample variance s, r rank central moment B_r, range Δ d, coefficient of variation C_v, coefficient of skew C_s、 Static coefficient C_eWith sample median R_m, 8 kinds of described arrowband statistical natures are designated as f respectively¹, f²..., f⁸, wherein Represent category label be t target jth section RCS sample sequences in The f for extractingⁱThe value of feature, N_tIt is the RCS sample sequences sum of target t；T=0 as bait target is set, t=1 is true mesh Mark；

S3, one-dimensional range profile data of each target in flight course are obtained by wideband radar, by the one-dimensional range profile Data are randomly divided into training dataset and test data set；Target scattering point spatial feature is extracted to every width one-dimensional range profile, including Target strong scattering dot center number NS, target radial length L, go mesostructure featureTarget range domain structure feature R_first And R_last, 5 kinds of described broadband characteristics are designated as f respectively⁹, f¹⁰..., f¹³, wherein Represent the f that category label goes out as the target of t in jth width one-dimensional range profile extracting dataⁱThe value of feature, N_tIt is target t's One-dimensional range profile sum；

Totally 13 kinds of the feature that S4, the training dataset that will be obtained according to step S2 and step S3 are extracted, for each Feature fⁱ, sample distribution confidence level parameter θ of all target samples of independent statistics on the dimensional featureⁱ, including feature minimum valueProfile maximaStep-length between Statistical AreaCategory distribution vectorCategory distribution confidence level vector

S5, the radar data to certain target to be tested, using extraction target narrowband and broadband characteristics value { y described in S2 and S3ⁱ | i=1,2 ..., 13 }, to every one-dimensional characteristic value yⁱ, the sample distribution confidence level parameter θ of character pair as described in S4ⁱCome to target Classification tentatively judged, obtains discriminant valueWhereinIt is by feature yⁱThe current test target judged Category label,It is the confidence level of the differentiation result；

S6, differentiate result { V using comprehensive decision algorithm is independent to 13 described in S5¹,V²,…,V¹³Merged, calculate Final fusion for classification score value C, the classification number k of sample to be tested is determined according to fusion score value C, if C>0.5 current to be identified Sample is true bullet k=1, and sample to be identified is false target k=0 if C≤0.5.

Further, n=20 described in step S1.

Further, the target narrowband statistical nature of each RCS sequences is extracted described in step S2, specific method is as follows：

Wherein：x_kIt is k-th RCS sampled value in current RCS sequences, k=1,2 ..., n,It is RCS serial means, s is sample This variance, B_rIt is r rank central moments, r=3, Δ d are range, C_vIt is the coefficient of variation, C_sIt is the coefficient of skew, C_eFor static state is Number, sample median R_mBe by n RCS sampled value by from small to large resequence after take numerical value in an intermediate position；Described 8 Plant narrow-band feature and be designated as f respectively¹, f²..., f⁸。

Further, the scattering point spatial feature of target in one-dimensional range profile is extracted described in step S3, specific method is such as Under：

S31, the one-dimensional range profile data according to wideband radar each target of acquisition in flight course, a range is from film size degree Data are designated as { a (1), a (2) ..., a (M) }, wherein, M=256 is the sampled point number of the one-dimensional picture, then find it most significantly Angle valueDetection is slided in one-dimensional picture data using 5 peakvalue's checking windows of point, all values are recorded More than a_max/ 2 wave crest point position { k_j| j=1,2 ..., m；M≤M }, these wave crest points are target scattering center；

S32, extraction target scattering point spatial feature：

NS=count { k_j|x(k_j)>a_max/ 2 }=m

L=Δs d × (k_m-k₁)

R_first=r₁/NS

R_last=r₂/NS

Wherein：Δ d is radar resolution ratio, r₁It is first scattering center and the distance at most strong scattering center, r₂For most The distance at latter scattering center and most strong scattering center, NS is target strong scattering dot center number, L be target radial length,Mesostructure feature, R are not removed_firstAnd R_lastIt is target range domain structure feature.The one-dimensional picture feature in 5 kinds of broadbands is remembered respectively It is f⁹, f¹⁰..., f¹³。

Further, sample described in step S4 is in feature fⁱOn distribution confidence level parameter θⁱStatistical method it is as follows：

Training sample concentrates the feature f that all sample extractions are arrivedⁱIt is designated asWherein N=N₁+N₀, N₀It is non-ballistic head's target training sample sum, N₁It is the training sample sum of true bullet target；

Feature minimum value：

Profile maxima：

Step-length between Statistical Area：

Category distribution vector：

Category distribution confidence level vector：

Wherein, D is the interval number for dividing, D=20； It is in feature fⁱUpper characteristic value is located at d-th non-ballistic head's target sample number of subregion,It is in feature fⁱUpper characteristic value Positioned at d-th sample number of the true bullet target of subregion, the interval range of d-th subregion is

Further, to every one-dimensional characteristic data { y of certain sample to be tested described in step S5ⁱ| i=1,2 ..., 13 } The method for carrying out classification tentatively judgement is as follows：

To i-th dimension characteristic value yⁱ, confidence level parameter θ is distributed as described in S4ⁱDetermine the affiliated distributed area numbering of this feature valueThe classification discriminant value for then obtaining the sample by i-th dimension feature is：

Further, result { V is differentiated using 13 independent characteristics described in step S6¹,V²,…,V¹³Finally divided Class is merged, and value of the confidence level less than 0.5 in 13 classification results will be ignored first, i.e.,：

Then final classification fusion score value is calculated：

Wherein,Represent statistics 13 it is independent differentiate confidence level in results as 0 Characteristic Number.

Beneficial effects of the present invention are, the Radar Objective Characteristics reflected in flight course by using true and false bullet target are not Together, the RCS extracting datas RCS sequence signatures for being obtained from Narrow-band Radar, from wideband radar obtain it is one-dimensional as being carried in data Target scattering point spatial feature is taken, has been devised for the training and recognition methods independently carried out per dimensional feature.Meanwhile, it is fully sharp With Narrow-band Radar and the complementarity of wideband radar, the independent characteristic recognition result of two kinds of radars is carried out into comprehensive distinguishing output, Effectively realize the true and false bullet target identification based on many radar informations.Test, correct recognition rata are identified to emulation data Reach 98.75%.

Brief description of the drawings

Fig. 1 is true and false bullet radar target identification method flow chart of the invention.

Specific embodiment

Summary is described in detail to technical scheme, below by emulation mode, to this The idiographic flow of invention is described in detail：

Using the true bullet target of STK software emulations and the flight path of false bullet target, per class target all by with The mode that machine sets initial transmissions parameter respectively produces 100 flight paths；The RCS of true and false bullet target is calculated using FEKO softwares Value, per every track of classification target and RCS values all according to some model radar specific works parameter add-on system error, The one-dimensional picture data in the frequency simulation calculation of 100Hz its flight course are pressed simultaneously.Then every flight path can be obtained One group of RCS sample sequence and one-dimensional picture sequence data.50 groups are randomly choosed as instruction from 100 groups of emulation data of every classification target Practice data set, remaining 50 groups used as test data set.

RCS values per continuous n=20 sampling instant (in one section of sample window) are constituted into a RCS sequence, each is counted Target narrowband statistical nature in sequence：

Wherein：x_kIt is k-th RCS sampled value in current RCS sequences, k=1,2 ..., n,It is RCS serial means, s is sample This variance, B_rIt is r ranks central moment (r=3), Δ d is range, C_vIt is the coefficient of variation, C_sIt is the coefficient of skew, C_eFor static state is Number, sample median R_mBe by n RCS sampled value by from small to large resequence after take numerical value in an intermediate position.This 8 kinds Narrow-band feature is designated as f respectively¹, f²..., f⁸, Represent target t (t=0 It is false target, t=1 is real goal) f that is extracted in j-th RCS sample sequenceⁱThe value of feature, N_tIt is target t's RCS sample sequences sum.

To the one-dimensional picture data of every breadth band { a (1), a (2) ..., a (256) } of target, find in one-dimensional picture data most Significantly angle valueDetection is slided in one-dimensional picture data using 5 peakvalue's checking windows of point, institute is recorded There is value to be more than a_max/ 2 wave crest point position { k_j| j=1,2 ..., m }, these wave crest points are target scattering center.Extracting should The target scattering point spatial feature of one-dimensional picture：

NS=count { k_j|x(k_j)>a_max/ 2 }=m

L=Δs d × (k_m-k₁)

R_first=r₁/NS

R_last=r₂/NS

Wherein：Δ d is radar resolution ratio, r₁It is first scattering center and the distance at most strong scattering center, r₂For most The distance at latter scattering center and most strong scattering center, NS is target strong scattering dot center number, L be target radial length,Mesostructure feature, R are not removed_firstAnd R_lastIt is target range domain structure feature.This one-dimensional picture feature in 5 kinds of broadbands is remembered respectively It is f⁹, f¹⁰..., f¹³。

Totally 13 dimension arrowband and broadband characteristics collection is obtained after features described above extraction is carried out to training dataset：

For each feature fⁱ(i.e. each column data in FThe all mesh of independent statistics Sample distribution confidence level parameter of the mark training sample on the dimensional featureIncluding：

Feature minimum value：

Profile maxima：

Step-length between Statistical Area：

Category distribution vector：

Category distribution confidence level vector：

Wherein, N is training sample sum, N=N₁+N₀, N₀It is non-ballistic head's target training sample sum, N₁It is true bullet Head's target training sample sum；D is the feature distribution interval number for dividing, D=20；It is in feature fⁱUpper characteristic value is located at D-th non-ballistic head's target sample number of subregion,It is in feature fⁱUpper characteristic value is located at d-th true bullet target of subregion Sample number, thenRepresenting the sample when sample characteristics is located at d-th subregion may be corresponding Category label,Represent the sample class when sample characteristics is located at d-th subregion Marked asConfidence level.The interval range of d-th subregion is defined as

For above-mentioned 13 dimensional feature, can all be counted per one-dimensional characteristic and obtain one group of sample distribution confidence level parameter.

For each sample to be tested in test set, the arrowband and broadband characteristics y={ y of its 13 dimension are equally extracted¹, y²,…,y¹³, for every one-dimensional characteristic yⁱ, i=1,2 ..., 13 independently carry out classification tentatively judges.First by i-th dimension feature Distribution confidence level parameter θⁱDetermine the affiliated distributed area numbering of this feature value Then by affiliated interval numbering pⁱThe classification discriminant value for obtaining the sample in i-th dimension feature is：It is characterized as that foundation judges that the sample said target classification is with i-th dimension And the confidence level of the differentiation result is

13 groups of principium identification result { V can obtain to 13 dimensional feature independent judgments¹,V²,…,V¹³, calculated using comprehensive judgement Method is merged against 13 independent differentiation results.Ignore value of the confidence level less than 0.5 in 13 classification results first, i.e.,：Then final classification fusion score value is calculated：Wherein,Represent statistics 13 it is independent differentiate confidence level in results as 0 Characteristic Number.

Finally, determine that the classification number k of sample to be tested determines according to fusion score value C,K=1 generations The current sample to be identified of table is true bullet, and it is false target that k=0 represents then sample to be identified.

Using the correct recognition rata of emulation data verification true and false bullet radar target identification method of the invention.Emulation is produced The true bullet target of 1 class and the arrowband including 3 class false targets including fragment, weight bait, light weight decoy in flight course And wideband data, emulation data are randomly divided into training dataset and test data set.The true bullet then concentrated to test data Head's mark correct recognition rata is 98.5%, and the correct recognition rata to false target is 99.0%, and average correct recognition rata is 98.75%.

Claims (7)

1. a kind of true and false bullet target identification method based on many radar informations, it is characterised in that comprise the following steps：

S1, the RCS values by the Narrow-band Radar true bullet of detection acquisition and bait target in flight course, individual per continuous n will adopt The value at sample moment constitutes a RCS sequence；The RCS sequences are randomly divided into training dataset and test data set；

S2, the RCS sequences obtained according to S1, each sequence to each target extract the arrowband statistical nature of target respectively, bag Include the average of RCS sequencesSample variance s, r rank central moment B_r, range △ d, coefficient of variation C_v, coefficient of skew C_s, it is static Coefficient C_eWith sample median R_m, 8 kinds of described arrowband statistical natures are designated as f respectively¹, f²..., f⁸, wherein Represent category label be t target jth section RCS sample sequences in The f for extractingⁱThe value of feature, N_tIt is the RCS sample sequences sum of target t；T=0 as bait target is set, t=1 is true mesh Mark；

S3, one-dimensional range profile data of each target in flight course are obtained by wideband radar, by the one-dimensional range profile data It is randomly divided into training dataset and test data set；Target scattering point spatial feature, including mesh are extracted to every width one-dimensional range profile Mark strong scattering dot center number NS, target radial length L, go mesostructure featureTarget range domain structure feature R_firstWith R_last, 5 kinds of described broadband characteristics are designated as f respectively⁹, f¹⁰..., f¹³, wherein Represent the f that category label goes out as the target of t in jth width one-dimensional range profile extracting dataⁱThe value of feature, N_tIt is the one of target t Dimension Range Profile sum；

Totally 13 kinds of the feature that S4, the training dataset that will be obtained according to step S2 and step S3 are extracted, for each feature fⁱ, sample distribution confidence level parameter θ of all target samples of independent statistics on the dimensional featureⁱ, including feature minimum value Profile maximaStep-length between Statistical AreaCategory distribution vectorCategory distribution confidence level vector

S5, the radar data to certain target to be tested, using extraction target narrowband and broadband characteristics value { y described in S2 and S3ⁱ| i= 1,2 ..., 13 }, to every one-dimensional characteristic value yⁱ, the sample distribution confidence level parameter θ of character pair as described in S4ⁱCome to target classification Tentatively judged, obtained discriminant valueWhereinIt is by feature yⁱThe class of the current test target judged Other label,It is the confidence level of the differentiation result；

S6, differentiate result { V using comprehensive decision algorithm is independent to 13 described in S5¹,V²,…,V¹³Merged, calculate final Fusion for classification score value C, the classification number k of sample to be tested is determined according to fusion score value C, if C>0.5 current sample to be identified It is true bullet k=1, sample to be identified is false target k=0 if C≤0.5.

2. a kind of true and false bullet target identification method based on many radar informations according to claim 1, it is characterised in that N=20 described in step S1.

3. a kind of true and false bullet target identification method based on many radar informations according to claim 2, it is characterised in that The target narrowband statistical nature of each RCS sequences is extracted described in step S2, specific method is as follows：

$\stackrel{\‾}{x}=\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{x}_k$

$s=\sqrt{\frac{1}{n-1}\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{(x_k-\stackrel{\‾}{x})}^2}$

$B_r=\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{(x_k-\stackrel{\‾}{x})}^r$

$\mathrm{\Δ}d=\underset{1\≤k\≤n}{max}\left\{x_k\right\}-\underset{1\≤k\≤n}{min}\left\{x_k\right\}$

$C_v=\frac{s}{\stackrel{\‾}{x}}$

$C_s=\frac{n}{(n-1)(n-2)}\frac{\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{(x_k-\stackrel{\‾}{x})}^3}{s^3},n\≤30$

$C_e=\frac{n^2-2n+3}{(n-1)(n-2)(n-3)}\frac{\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{(x_k-\stackrel{\‾}{x})}^4}{s^4}-\frac{3(2n-3)}{n(n-1)(n-2)(n-3)}\frac{{\[\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{(x_k-\stackrel{\‾}{x})}^2\]}^2}{s^4},n\≤30$

$R_m=x_{pos\left(\frac{n}{2}\right)}$

Wherein：x_kIt is k-th RCS sampled value in current RCS sequences, k=1,2 ..., n,It is RCS serial means, s is sample side Difference, B_rIt is r rank central moments, r=3, △ d are range, C_vIt is the coefficient of variation, C_sIt is the coefficient of skew, C_eIt is static coefficient, sample This median R_mBe by n RCS sampled value by from small to large resequence after take numerical value in an intermediate position；Described 8 kinds narrow Band feature is designated as f respectively¹, f²..., f⁸。

4. a kind of true and false bullet target identification method based on many radar informations according to claim 3, it is characterised in that The scattering point spatial feature of target in one-dimensional range profile is extracted described in step S3, specific method is as follows：

S31, the one-dimensional range profile data according to wideband radar each target of acquisition in flight course, a range is from film size degrees of data { a (1), a (2) ..., a (M) } is designated as, wherein, M=256 is the sampled point number of the one-dimensional picture, then find its maximum amplitude valueDetection is slided in one-dimensional picture data using 5 peakvalue's checking windows of point, record all values are more than a_max/ 2 wave crest point position { k_j| j=1,2 ..., m；M≤M }, these wave crest points are target scattering center；

S32, extraction target scattering point spatial feature：

NS=count { k_j|x(k_j)>a_max/ 2 }=m

L=△ d × (k_m-k₁)

$\stackrel{\‾}{A}=\[\underset{k=k_1}{\overset{k_m}{\mathrm{\Σ}}}|x\left(k\right)/(k-k_1+1)|\]/a_{max}$

R_first=r₁/NS

R_last=r₂/NS

Wherein：△ d are radar resolution ratio, r₁It is first scattering center and the distance at most strong scattering center, r₂For last Individual scattering center and the distance at most strong scattering center, NS are target strong scattering dot center number, and L is target radial length, A does not go Mesostructure feature, R_firstAnd R_lastIt is target range domain structure feature.5 kinds of broadbands are one-dimensional as feature is designated as f respectively⁹, f¹⁰..., f¹³。

5. a kind of true and false bullet target identification method based on many radar informations according to claim 4, it is characterised in that Sample is in feature f described in step S4ⁱOn distribution confidence level parameter θⁱStatistical method it is as follows：

Training sample concentrates the feature f that all sample extractions are arrivedⁱIt is designated asWherein N=N₁+N₀, N₀For Non-ballistic head's target training sample sum, N₁It is the training sample sum of true bullet target；

Feature minimum value：

Profile maxima：

Step-length between Statistical Area：

Category distribution vector：

Category distribution confidence level vector：

Wherein, D is the interval number for dividing, D=20； It is in feature fⁱUpper characteristic value is located at d-th non-ballistic head's target sample number of subregion,It is in feature fⁱUpper characteristic value position In d-th sample number of the true bullet target of subregion, the interval range of d-th subregion is

6. a kind of true and false bullet target identification method based on many radar informations according to claim 5, it is characterised in that To every one-dimensional characteristic data { y of certain sample to be tested described in step S5ⁱ| i=1,2 ..., 13 } carry out what classification tentatively judged Method is as follows：

To i-th dimension characteristic value yⁱ, confidence level parameter θ is distributed as described in S4ⁱDetermine the affiliated distributed area numbering of this feature valueThe classification discriminant value for then obtaining the sample by i-th dimension feature is：

7. a kind of true and false bullet target identification method based on many radar informations according to claim 6, it is characterised in that Differentiate result { V using 13 independent characteristics described in step S6¹,V²,…,V¹³Final classification fusion being carried out, will ignore first Value of the confidence level less than 0.5 in 13 classification results, i.e.,：

Then final classification fusion score value is calculated：

Wherein,Represent statistics 13 it is independent differentiate confidence level in results as 0 Characteristic Number.