CN103941244A - Radar target one-dimensional range profile local optimal sub-space recognition method - Google Patents

Abstract

The present invention provides a kind of radar target-range image local optimum subspace identification methods, the effective recognition performance improved to radar target, this method calculates nearest inter- object distance scatter matrix and nearest between class distance scatter matrix first with training data, then, a local optimum subspace is established by optimum ratio criterion, target signature is extracted, is classified using minimum distance classifier, it is final to determine classification belonging to input target; Specially vector is determined using radar target-range image trained vector xij And vector Determine vector With It determines matrix D W and DB, determines m vector a1, a2.., am of local optimum subspace, local optimum subspace is determined according to λ i and vector ai (i=1,2.., m) Template library is determined using projection of the trained vector in the A of subspace, determine the local optimum sub- picture of input target one-dimensional range profile xt, it determines the distance between local optimum picture and library template vector, determines classification belonging to input target one-dimensional range profile using minimum distance classifier.

Description

A kind of radar target-range image local optimum subspace recognition methods

Technical field

The invention belongs to Technology of Radar Target Identification field, relate to a kind of radar target-range image local optimum subspace recognition methods.

Background technology

Subspace method is classical mode identification method, is widely used in image recognition, recognition of face, also has a lot of application in radar target recognition.Common proper subspace method and Canonical sub-space method have all obtained good recognition effect in radar target recognition.Wherein, proper subspace can keep in low dimensional feature space the energy of raw data, but is not optimum from classification performance.For proper subspace, by making inter-object distance minimum, between class distance maximum is extracted target signature to Canonical sub-space, has improved to a certain extent target identification performance.

But, Canonical sub-space method adopts average and the interior average distribution between class structure in macroscopic view is described class of class between class, but may not be optimum from part, Canonical sub-space dimension be determined by target classification number simultaneously, and target classification a few hours can be caused the loss of classified information.Therefore, the recognition performance of Canonical sub-space method has further room for improvement.

Summary of the invention

The object of the present invention is to provide a kind of radar target-range image local optimum subspace recognition methods, effectively improved the recognition performance to radar target, its technical scheme is:

A kind of radar target-range image local optimum subspace recognition methods, it is characterized in that, first utilize training data to calculate nearest inter-object distance scatter matrix and nearest between class distance scatter matrix, then, set up a local optimum subspace by optimum ratio criterion, extract target signature, adopt minimum distance classifier to classify, classification under final definite input target, comprises the following steps:

1) utilize radar target-range image trained vector x _ijdetermine vector and vector

2) according to x _ij, with determine vector with

3) according to vector with determine matrix D _wand D _b;

4) utilize matrix D _wand D _bdetermine m vector a of local optimum subspace ₁, a ₂..., a _m;

5) according to λ _iwith vector a _i(i=1,2 ..., m) determine local optimum subspace

6) utilize the projection of trained vector in the A of subspace to determine template base;

7) determine input target one-dimensional range profile x _tlocal optimum picture;

8) determine the distance between local optimum picture and library template vector;

9) utilize minimum distance classifier to determine the affiliated classification of input target one-dimensional range profile.

Further preferably, utilize the local optimum picture of every classification target one-dimensional range profile training sample to set up template base, determine the distance between local optimum picture and library template, judge target classification by minimum distance criterion.

Further preferably, judge that the concrete way of target is: to matrix carry out eigen decomposition, determine local optimum subspace by a front m dominant eigenvalue and corresponding latent vector training objective one-dimensional range profile is by formula y=A ^tx calculates the sub-picture of local optimum, by the template base { y of the local optimum picture composition respective objects of every classification target training one-dimensional range profile _ij, wherein y _ijbe j training one-dimensional range profile vector of i classification target; One-dimensional range profile to input target is pressed x _t, y=A ^tx calculates local optimum as y _t, and calculate following distance:

s _ij＝||y _t-y _ij||i＝1,2,…g；j＝1,2,…,N _i

Determine the minor increment in i class target:

s _i＝min{si _j}

If

$k=\arg \underset{\left\{i\right\}}{\min }\left\{s_i\right\},$ Sentencing input target is k class.

Beneficial effect of the present invention:

The present invention utilizes training sample data to set up local optimum subspace and extracts target signature, make the nearest inter-object distance of one-dimensional range profile character pair reach minimum and recently between class distance reach maximum, improve target classification performance; Meanwhile, the size of local optimum subspace and number of targets are irrelevant, can obtain the subspace of suitable size, more intactly keep the classified information of raw data, effectively improve the recognition performance of radar to multi-class targets.

Brief description of the drawings

Fig. 1 is the process flow diagram of radar target-range image local optimum subspace recognition methods of the present invention.

Embodiment

Below in conjunction with embodiment and accompanying drawing, the present invention is described in further details.

Local optimum varitron space:

If x _ij(n ties up column vector) is i ^thclassification target j ^thindividual training one-dimensional range profile, i=1,2 ..., g; J=1,2 ..., N _i, N ₁+ N ₂+ ... + N _g=N, wherein N _ibe i ^thclassification target training one-dimensional range profile sample number, N is training one-dimensional range profile total sample number.Definition x _ijnearest class in one-dimensional range profile sample and recently the outer one-dimensional range profile sample of class be

$x_{\mathrm{ij}}^W=\underset{\left\{x_{\mathrm{ir}}\right\}}{\arg \min }{\left|\right|x_{\mathrm{ij}}-x_{\mathrm{ir}}\left|\right|}_p,r=\mathrm{1,2},...,N_i---\left(1\right)$

$x_{\mathrm{ij}}^B=\underset{\left\{x_{\mathrm{kr}}\right\}}{\arg \min }{\left|\right|x_{\mathrm{ij}}-x_{\mathrm{kr}}\left|\right|}_p,k=\mathrm{1,2},...,g,k\≠i,r=\mathrm{1,2},...,N_k---\left(2\right)$

Wherein || || _pfor p norm, for x _ijdecent of one dimension in corresponding nearest class, for x _ijdecent of the outer one dimension of corresponding nearest class.Sample x _ijthe corresponding nearest between class distance vector of nearest inter-object distance vector is

$d_{\mathrm{ij}}^W=|x_{\mathrm{ij}}-x_{\mathrm{ij}}^W|---\left(3\right)$

$d_{\mathrm{ij}}^B=|x_{\mathrm{ij}}-x_{\mathrm{ij}}^B|---\left(4\right)$

Wherein || represent that vector element takes absolute value, with be respectively x _ijthe nearest between class distance vector of nearest inter-object distance vector.

Make a for the dimension of n arbitrarily column vector, calculate following ratio

$F=\frac{\underset{i=1}{\overset{g}{\mathrm{\Σ}}}\underset{j=1}{\overset{N_i}{\mathrm{\Σ}}}{a}^T{d}_{\mathrm{ij}}^W{\left(d_{\mathrm{ij}}^W\right)}^{T}a}{\underset{i=1}{\overset{g}{\mathrm{\Σ}}}\underset{j=1}{\overset{N_i}{\mathrm{\Σ}}}{a}^T{d}_{\mathrm{ij}}^B{\left(d_{\mathrm{ij}}^B\right)}^{T}a}=\frac{a^T{D}_{W}a}{a^T{D}_{B}a}---\left(5\right)$

Wherein D _wfor the nearest inter-object distance scatter matrix of each training sample, D _bfor the nearest between class distance scatter matrix of each training sample, its calculation expression is as follows

$D_W=\underset{i=1}{\overset{g}{\mathrm{\Σ}}}\underset{j=1}{\overset{N_i}{\mathrm{\Σ}}}{d}_{\mathrm{ij}}^W{\left(d_{\mathrm{ij}}^W\right)}^T---\left(6\right)$

$D_B=\underset{i=1}{\overset{g}{\mathrm{\Σ}}}\underset{j=1}{\overset{N_i}{\mathrm{\Σ}}}{d}_{\mathrm{ij}}^B{\left(d_{\mathrm{ij}}^B\right)}^T---\left(7\right)$

Formula (5) shows, ratio F is less, and the nearest inter-object distance of each sample is less, and the nearest class distance of each sample is larger, thereby can obtain distribution between class structure in the premium class of each sample, is conducive to strengthen classification performance.Formula (5) both sides are to vector a differentiate, and make result wait until in zero, and abbreviation can obtain

$D_B^{-1}{D}_{W}a=\mathrm{\λa}---\left(8\right)$

Wherein λ is eigenvalue, and a is the latent vector that λ is corresponding.If putting in order as λ of the non-zero eigenvalue in formula (8) ₁> λ ₂λ _m> 0, corresponding latent vector is a ₁, a ₂..., a _m, get front m maximum non-zero eigenvalue and corresponding latent vector matrix composed as follows

$A=[\sqrt{{\mathrm{\λ}}_1}{a}_1\sqrt{{\mathrm{\λ}}_2}{a}_2...\sqrt{{\mathrm{\λ}}_m}{a}_m]---\left(9\right)$

Claim that matrix A is optimal partial subspace.Sample x _ijcarry out projection to subspace A

y _ij＝A ^Tx _ij (10)

Claim yi _jfor xi _jcorresponding projection vector.Then, utilize projection vector to identify as target signature.

Radar target recognition based on local optimum subspace:

Suppose to have g class target, can utilize every classification target one-dimensional range profile training sample set to set up local optimum subspace by formula (8) and formula (9), the one-dimensional range profile of training objective obtains son picture by formula (10) to local optimum subspace projection, by the library template vector of such target of local optimum picture composition of every classification target one-dimensional range profile training sample, i classification target template base is

$\{y_{i1}{y}_{i2}...y_{{\mathrm{iN}}_i}\}---\left(11\right)$ To the one-dimensional range profile x of input target _t, can calculate optimum Nonlinear Orthogonal projection vector by formula (10) is y _t, and calculate with

Lower distance

s _ij＝||y _t-y _ij||i＝1,2,…g；j＝1,2,…,N _i (12)

$s_i=\underset{\left\{j\right\}}{\min }\left\{s_{\mathrm{ij}}\right\}---\left(13\right)$

If

$k=\arg \underset{\left\{i\right\}}{\min }\left\{s_i\right\}---\left(14\right)$

Sentencing input target is k class.

Accompanying drawing illustrates the process flow diagram of radar target-range image local optimum subspace recognition methods of the present invention.Process flow diagram starts from step 201.

In step 202, utilize g classification target training one-dimensional range profile vector x _ijdetermine following vector:

$x_{\mathrm{ij}}^W=\underset{\left\{x_{\mathrm{ir}}\right\}}{\arg \min }{\left|\right|x_{\mathrm{ij}}-x_{\mathrm{ir}}\left|\right|}_p$

$x_{\mathrm{ij}}^B=\underset{\left\{x_{\mathrm{kr}}\right\}}{\arg \min }{\left|\right|x_{\mathrm{ij}}-x_{\mathrm{kr}}\left|\right|}_p$

Wherein r=1,2 ..., N _i; K=1,2 ..., g, k ≠ i, r=1,2 ..., N _k; G is target classification number, N _iit is i classification target number of training;

In step 2031, determine vector:

$d_{\mathrm{ij}}^W=|x_{\mathrm{ij}}-x_{\mathrm{ij}}^W|$

$d_{\mathrm{ij}}^B=|x_{\mathrm{ij}}-x_{\mathrm{ij}}^B|$

In step 2032, determine matrix:

$D_W=\underset{i=1}{\overset{g}{\mathrm{\Σ}}}\underset{j=1}{\overset{N_i}{\mathrm{\Σ}}}{d}_{\mathrm{ij}}^W{\left(d_{\mathrm{ij}}^W\right)}^T$

$D_B=\underset{i=1}{\overset{g}{\mathrm{\Σ}}}\underset{j=1}{\overset{N_i}{\mathrm{\Σ}}}{d}_{\mathrm{ij}}^B{\left(d_{\mathrm{ij}}^B\right)}^T$

In step 2033, determine the vector in local optimum subspace: a _ifor matrix m the eigenvector that dominant eigenvalue is corresponding;

In step 2034, determine local optimum subspace: wherein λ _ifor vector a _icorresponding eigenvalue;

In step 204, determine every class target template base:

In step 2051, determine input target one-dimensional range profile x _tlocal optimum picture: y _t=A ^tx _t;

In step 2052, determine the distance between local optimum picture and To Template vector:

s _ij＝||y _t-y _ij||i＝1,2,…g；j＝1,2,…,N _i

$s_i=\underset{\left\{j\right\}}{\min }\left\{s_{\mathrm{ij}}\right\}$

In step 2053, determine the affiliated classification of input target one-dimensional range profile:

End at step 206 according to the flow process of radar target-range image local optimum subspace recognition methods of the present invention.

The simulation experiment result: the validity of extracting method in order to verify, carry out following emulation experiment.

Design four kinds of point targets: " | " font, " V " font, " doing " font and " little " font target.The bandwidth of radar transmitted pulse is 150MHZ (range resolution is 1m, and radar radially sample interval is 0.5m), and target is set to even scattering point target, " | " scattering point of target is 5, the scattering of its excess-three target is counted and is 9.Object attitude angle be within the scope of 0 °～50 ° in the one-dimensional range profile of 1 °, get object attitude angle and be 0 °, 2 °, 4 °, 6 ° ..., the one-dimensional range profile of 50 ° trains, the one-dimensional range profile of all the other attitude angle is as test data, and every classification indicates 25 test sample books.

To four kinds of targets (" | " font target, " V " font target, " doing " font target and " little " font target), within the scope of attitude angle 0o～50o, utilize radar target-range image local optimum subspace recognition methods herein and Canonical sub-space method to carry out identification experiment, result as shown in Table 1.

From table one, to target " | ", the discrimination of Canonical sub-space method is 88%, and the discrimination of radar target-range image local optimum subspace recognition methods is herein 96%; To target " V ", the discrimination based on Canonical sub-space method is 86%, and the discrimination of radar target-range image local optimum subspace recognition methods is herein 90%; Target " is done ", and the discrimination based on Canonical sub-space method is 78%, and the discrimination of radar target-range image local optimum subspace recognition methods is herein 85%; To target " little ", the discrimination of Canonical sub-space method is 85%, and the discrimination of radar target-range image local optimum subspace recognition methods is herein 90%.On average, to four class targets, the correct recognition rata of radar target-range image local optimum subspace recognition methods herein, higher than Canonical sub-space method, illustrates that radar target-range image local optimum subspace recognition methods herein can improve the recognition performance of multi-class targets really.

The recognition result of table one or two kind of method

Claims (3)

1. a radar target-range image local optimum subspace recognition methods, it is characterized in that, first utilize training data to calculate nearest inter-object distance scatter matrix and nearest between class distance scatter matrix, then, set up a local optimum subspace by optimum ratio criterion, extract target signature, adopt minimum distance classifier to classify, classification under final definite input target, comprises the following steps:

1) utilize radar target-range image trained vector x _ijdetermine vector and vector

2) according to x _ij, with determine vector with

3) according to vector with determine matrix D _wand D _b;

4) utilize matrix D _wand D _bdetermine m vector a of local optimum subspace ₁, a ₂..., a _m;

5) according to λ _iwith vector a _i(i=1,2 ..., m) determine local optimum subspace

6) utilize the projection of trained vector in the A of subspace to determine template base;

7) determine input target one-dimensional range profile x _tlocal optimum picture;

8) determine the distance between local optimum picture and library template vector;

9) utilize minimum distance classifier to determine the affiliated classification of input target one-dimensional range profile.

2. by a kind of radar target-range image local optimum subspace recognition methods described in claim 1, it is characterized in that, utilize the local optimum picture of every classification target one-dimensional range profile training sample to set up template base, determine the distance between local optimum picture and library template, judge target classification by minimum distance criterion.

3. by a kind of radar target-range image local optimum subspace recognition methods described in claim 1, it is characterized in that, judge that the concrete way of target is: to matrix carry out eigen decomposition, determine local optimum subspace by a front m dominant eigenvalue and corresponding latent vector training objective one-dimensional range profile is by formula y=A ^tx calculates the sub-picture of local optimum, by the template base { y of the local optimum picture composition respective objects of every classification target training one-dimensional range profile _ij, wherein y _ijbe j training one-dimensional range profile vector of i classification target; One-dimensional range profile to input target is pressed x _t, y=A ^tx calculates local optimum as y _t, and calculate following distance:

s _ij＝||y _t-yi _j||i＝1,2,…g；j＝1,2,…,N _i

Determine the minor increment in i class target:

s _i＝min{si _j}

If

$k=\arg \underset{\left\{i\right\}}{\min }\left\{s_i\right\},$ Sentencing input target is k class.