CN106468770A - Closely optimum radar target detection method under K Distribution Clutter plus noise - Google Patents

Abstract

The invention discloses the closely optimum radar target detection method under a kind of K Distribution Clutter plus noise, mainly solve the problems, such as that prior art is not exclusively useful in K Distribution Clutter and adds and under white Gaussian noise background, carry out target detection.Implementation step is：1) obtain echo data matrix, and piecemeal is carried out to it；2) choose range cell z to be detected of b-th echo data block_k, calculate its covariance matrix 3) and utilize covariance matrix to calculate miscellaneous noise ratio CNR of echo data；4) utilize this CNR and statistic of test ξ of calculating range cell to be detected_k；5) detection threshold T is calculated according to false-alarm probability_ξ；6) by comparing ξ_kAnd T_ξSize judge that target whether there is.The present invention improves target detection performance, can be used for the radar motion detection under sea clutter background.

Description

Closely optimum radar target detection method under K Distribution Clutter plus noise

Technical field

The invention belongs to Radar Targets'Detection technical field is and in particular to a kind of nearly optimum under K Distribution Clutter plus noise Radar target detection method, can be used for the target detection under sea clutter background.

Background technology

Target detection under sea clutter background is an important applied field of radar.

Matched filter MF and adaptive matched filter AMF is the optimum detector under Gaussian Clutter background.With thunder Reach the raising of resolution, Gauss model is no longer applicable, this two detectors are no longer applicable because of model mismatch.

Complex Gaussian model is the sea clutter model that current scholar is widely recognized as, and it is that the slow positive stochastic variable texture becoming divides Amount and the product of the fast multiple Gauss random vector speckle component becoming.When the texture component of sea clutter obeys gamma distribution, it Amplitude obeys K distribution.Optimum K distribution detector OKD, because there is Bessel function in expression formula, is difficult at present.Document P- L.Shui,M.Liu,and S-W.Xu,"Shape-parameter-dependent coherent radar target detection in K-distributed clutter,"IEEE Trans.Aerospace Electron.Systems 52 (1):It is optimum and calculate the attainable form parameter that depends on that 451-465,2016. discusses K Distribution Clutter background lower aprons Detector α-MF and adaptive detector α-the AMF depending on form parameter.However, when system noise and external noise are present in In radar and when can not ignore, this two detectors are no longer applicable because of model mismatch.

There are some researches show, K Distribution Clutter adds Gaussian noise model and is more suitable for describing the actual ghosts that radar receives Data.Document .F.Gini, " Suboptimal coherent radar detection in a mixture of K- distributed and Gaussian clutter,"IEE Proc.-Radar,Sonar,Navig.,144(1):39-47, 1997. with document F.Gini, M.V.Greco, A.Farina, P.Lombardo, " Optimum and mismatched detection against K-distributed clutter plus Gaussian clutter,"IEEE Trans.Aerospace Electron.Systems 34(3):Discuss optimum under this Model Background in 860-876,1998. This white noise detector OKGD is increased in detector optimum K distribution, but because detector expression formula is excessively complicated, computational efficiency is extremely low not Can apply to reality.Add under white Gaussian noise background in K Distribution Clutter, at present the matched filter MF of use and Adaptive matching The detection performance of wave filter AMF quickly reduces with the rising of miscellaneous noise ratio, depends on detector α-MF and the dependence of form parameter Detect poor-performing in the adaptive detector α-AMF of form parameter when miscellaneous noise ratio is relatively low.

Content of the invention

It is an object of the invention to proposing the closely optimum radar target detection method under a kind of K Distribution Clutter plus noise, with Realize compromise between detection performance optimization and computational efficiency.

For realizing above-mentioned technical purpose, technical scheme includes as follows：

(1) continuous pulse signal launched by radar transmitter, and radar receiver receives the echo data matrix X of Q × M dimension, Wherein, Q represents the accumulation umber of pulse of echo data, and M represents the range cell number of echo data；

(2) echo data matrix X is divided into B echo data block along pulse dimension：X₁,X₂…,X_b,…,X_B, wherein, X_bRepresent b-th echo data block, each echo data block is the matrix of N × M dimension, N represents b-th echo data block X_bArteries and veins Rush number；

(3) choose b-th echo data block X_bK-th range cell be X_bRange cell z to be detected_k, using sample Covariance matrix estimation method calculates range cell z to be detected_kCovariance matrix

(4) utilize range cell z to be detected_kCovariance matrixCalculate miscellaneous noise ratio CNR of echo data；

(5) miscellaneous noise ratio CNR of echo data and range cell z to be detected are utilized_kCovariance matrixCalculating is treated Detecting distance unit z_kStatistic of test ξ_k:

Wherein, p represents Doppler's steering vector of target, and subscript H represents and takes conjugate transpose, subscript -1 represent take inverse, | | represent Modulus of access, ν represents the form parameter of texture component, and subscript γ represents exponential factor, and γ is any real number more than 0, takes It is worth for γ=2；

(6) false-alarm probability f being given according to system, calculates detection threshold T by Monte Carlo experiment_ξ；

(7) according to statistic of test ξ_kWith detection threshold T_ξJudge that target whether there is：If ξ_k≥T_ξ, then show b-th Echo data block X_bRange cell to be detected have target, if ξ_k<T_ξ, then show b-th echo data block X_bTo be detected away from There is no target from unit.

The present invention compared with prior art has advantages below：

1) because K Distribution Clutter of the present invention adds the radar return number in Gaussian noise model and actual environment More mate according to characteristic, compare the data model that existing detector is adopted, improve the detection performance of radar target.

2) due to present invention utilizes miscellaneous noise ratio information, dividing in conjunction with the optimality of matched filter MF and K under Gaussian Background Depend on the near-optimization of the detector α-MF of form parameter under cloth clutter background, improve K distribution plus the white Gaussian noise back of the body The detection performance of target under scape.

3) due to form simple expression formula calculating unit checks statistic to be detected in the present invention, make computational efficiency far high In optimum K distribution increase this white noise detector OKGD it is achieved that K distribution plus white Gaussian noise background under optimal detection performance with Computational efficiency compromise.

Brief description

Fig. 1 is the flowchart of the present invention；

Fig. 2 is to be added in white K Distribution Clutter with the present invention and existing method to carry out target detection under white Gaussian noise background Result schematic diagram；

Fig. 3 is to be added in coloured K Distribution Clutter with the present invention and existing method to carry out target detection under white Gaussian noise background Result schematic diagram.

Specific embodiment

The invention will be further described below in conjunction with the accompanying drawings：

With reference to Fig. 1, the present invention to realize step as follows：

Step 1, obtains echo data matrix X.

Continuous pulse signal launched by radar transmitter, and pulse signal is irradiated to body surface and produces echo, and radar receives Machine receives echo data matrix X, and echo data matrix X is the matrix that size is Q × M dimension, and wherein, Q represents the long-pending of echo data Tired umber of pulse, M represents the range cell number of echo data.

Step 2, is processed to the echo data partitioning of matrix.

Echo data matrix X is divided into the echo data block of B N × M dimension along pulse dimension, wherein, N represents each time The umber of pulse of ripple data block, B echo data block is expressed as X₁,X₂…,X_b,…,X_B, X_bRepresent b-th echo data block, B=1, the value of 2 ..., B, B is natural number more than 1 and meets B × N≤Q, takes B=10 in present example⁴.

Step 3, chooses range cell to be detected according to echo data block, calculates the covariance matrix of range cell to be detected Estimate

3.1) choose b-th echo data block X_bK-th range cell as this echo data block distance to be detected Unit z_k, k=1,2 ..., M；

3.2) remove range cell z to be detected_kAnd its two adjacent range cells, with echo data block X_bRemaining common L Individual range cell is as range cell z to be detected_kReference distance unit, wherein L is the natural number more than 1；

3.3) calculate range cell z to be detected_kCovariance matrix

Calculate range cell z to be detected_kCovariance matrixMainly there are three kinds of methods：

The first is sample covariance matrix method of estimation, and its computing formula is:

Wherein, z_qRepresent q-th reference distance unit, subscript H represents and take conjugate transpose, L represents reference distance unit number；

Second is normalization sample covariance matrix method of estimation, and its computing formula is：

The third is power intermediate value normalized covariance matrix estimation method, and its computing formula is:

Wherein, meadia { } expression takes intermediate value.

This example uses but is not limited to calculate range cell z to be detected using first method_kCovariance matrix

Step 4, using range cell z to be detected_kCovariance matrixCalculate miscellaneous noise ratio CNR of echo data.

4.1) Doppler frequency f according to given target_s, calculate normalization Doppler frequency f_d：

Wherein, t represents radar pulse transmit cycle；

4.2) utilize normalization Doppler frequency f_d, calculate the Doppler steering vector p of target：

Wherein, subscript T represents and takes transposition；

4.3) the Doppler steering vector p of target and range cell z to be detected are utilized_kCovariance matrixMeter Calculate miscellaneous noise ratio CNR of echo data：

Wherein, μ represents the scale parameter of texture component, and ν represents the form parameter of texture component, σ²Represent white Gaussian noise Power,Represent the power ratio of K Distribution Clutter and white Gaussian noise in echo data.

Step 5, using miscellaneous noise ratio CNR and range cell z to be detected of echo data_kCovariance matrixMeter Calculate range cell z to be detected_kStatistic of test ξ_k:

Wherein, subscript -1 represent take inverse, | | represent Modulus of access, subscript γ represents exponential factor, γ is any more than 0 Real number, this example value is γ=2.

Step 6, false-alarm probability f being given according to system, detection threshold T is calculated by Monte Carlo experiment_ξ.

6.1) C is made to be the natural number more than 100/f setting, value is C=10⁶, the C distance list without target for the emulation Unit, calculates the statistic of test of each range cell：

Wherein, z_wRepresent w-th range cell, ξ_wRepresent the statistic of test of w-th range cell；

6.2) C obtained statistic of test is arranged in descending order, take [Cf] the individual statistic of test conduct after arrangement Detection threshold T_ξ, the wherein maximum integer less than real number Cf for [Cf] expression.

Step 7, according to range cell z to be detected_kStatistic of test ξ_kWith detection threshold T_ξJudge that target whether there is.

By range cell z to be detected_kStatistic of test ξ_kWith detection threshold T_ξIt is compared：If ξ_k≥T_ξ, then show Range cell to be detected has target, if ξ_k<T_ξ, then show that range cell to be detected does not have target.

With reference to emulation experiment, the effect of the present invention is described further.

1. simulation parameter

The echo data adopting in emulation experiment is that the K Distribution Clutter being produced by Matlab software adds white Gaussian noise number According to.

Parameter 1, produces white K Distribution Clutter using Matlab software emulation and adds white Gaussian noise data, emulation data Parameter is set to：Umber of pulse N=16, form parameter ν=0.5 of texture component, white Gaussian noise power σ²=1, normalization is many General Le frequency f_dIt is the random number between 0 to 0.5, false-alarm probability f=10^-4, change the scale parameter μ of texture component, make letter Miscellaneous noise ratio SCNR=9dB, miscellaneous noise ratio CNR is from -20dB to 20dB.

Parameter 2, produces coloured K Distribution Clutter using Matlab software emulation and adds white Gaussian noise data, emulation data Parameter is set to：Umber of pulse N=8, form parameter ν=1 of texture component, scale parameter μ=2 of texture component, Gauss white noise Acoustical power σ²=1, speckle covariance matrix R=[m_ij]_1≤i,j≤N,m_ij=ρ^|i-j|,0<ρ<1, wherein | | represent Modulus of access, m_ij Represent the element of speckle covariance matrix R the i-th row jth row, ρ represents correlation coefficient, ρ=0.5, believe miscellaneous noise ratio SCNR=3dB, ginseng Examine unit number L=32, false-alarm probability f=10^-4, normalization Doppler frequency value is from 0 to 0.5.

2. emulation experiment content

Emulation experiment passes through to compare the detection probability analysis detection performance of distinct methods under same background, and detection probability is bigger Show that detector detection performance is better.

Emulation experiment 1

Given normalization Doppler frequency, speckle covariance matrix R=I, when miscellaneous noise ratio CNR changes from -20dB to 20dB When, using matched filter MF, depend on form parameter detector α-MF, this white noise detector OKGD is increased in optimum K distribution With the present invention, carry out target detection under parameter 1, testing result is as shown in Fig. 2 the transverse axis in Fig. 2 represents that miscellaneous noise ratio CNR becomes Change, the longitudinal axis represents detection probability, and the solid line in Fig. 2 represents the detection probability curve of the detector α-MF depending on form parameter, Dotted line represents the detection probability curve of matched filter MF, and frame setting-out represents the detection probability curve of the present invention, and dotted line represents The detection probability curve of this white noise detector OKGD is increased in optimum K distribution.

From Figure 2 it can be seen that under K Distribution Clutter plus noise background, method performance proposed by the present invention adds close to optimum K distribution White Gaussian noise detector OKGD, it is better than the matched filter MF and detector α-MF depending on form parameter, the i.e. present invention Performance loss can be reduced to a very low level.

Emulation experiment 2

When normalization Doppler frequency value is from 0 to 0.5 change, the miscellaneous noise ratio depending on Doppler frequency is with normalization Doppler frequency value changes, using adaptive matched filter AMF, depend on the adaptive detector α of form parameter- AMF and the present invention, carry out target detection under parameter 2, and testing result is as shown in figure 3, the transverse axis in Fig. 3 represents that how general normalization is Strangle frequency change, the longitudinal axis represents detection probability, and the solid line in Fig. 3 represents the detection of the detector α-AMF depending on form parameter Probability curve, dotted line represents the detection probability curve of adaptive matched filter AMF, and frame setting-out represents the detection probability of the present invention Curve.

Fig. 3 shows, when the less i.e. target of normalization Doppler frequency value is dominant area in clutter, the present invention and near-optimization Depend on form parameter adaptive detector α-AMF performance identical, and performance be better than adaptive matched filter AMF.When When the larger i.e. target of normalization Doppler frequency value is outside clutter is dominant area, performance of the present invention is close to even slightly better than near-optimization Adaptive matched filter AMF, and performance is more much better than the adaptive detector α-AMF depending on form parameter.Total comes See, the detection performance of the present invention is better than existing method.

Claims (4)

1. the closely optimum radar target detection method under a kind of K Distribution Clutter plus noise, including：

(1) continuous pulse signal launched by radar transmitter, and radar receiver receives the echo data matrix X of Q × M dimension, wherein, Q represents the accumulation umber of pulse of echo data, and M represents the range cell number of echo data；

(2) echo data matrix X is divided into B echo data block along pulse dimension：X₁,X₂…,X_b,…,X_B, wherein, X_bTable Show b-th echo data block, each echo data block is the matrix of N × M dimension, N represents b-th echo data block X_bPulse Number；

(3) choose b-th echo data block X_bK-th range cell be X_bRange cell z to be detected_k, using sample association side Difference matrix estimation method calculates range cell z to be detected_kCovariance matrix

(4) utilize range cell z to be detected_kCovariance matrixCalculate miscellaneous noise ratio CNR of echo data；

(5) miscellaneous noise ratio CNR of echo data and range cell z to be detected are utilized_kCovariance matrixCalculate to be detected Range cell z_kStatistic of test ξ_k:

${\mathrm{\&xi;}}_k=\frac{|p^H{{\hat{R}}_k}^{-1}{z}_k{|}^2}{\left(p^H{{\hat{R}}_k}^{-1}p\right)(1+{\mathrm{CNR}}^{\mathrm{\&gamma;}}{\left({z_k}^H{{\hat{R}}_k}^{-1}{z}_k\right)}^{\frac{N}{N+v}})}$

Wherein, p represents Doppler's steering vector of target, and subscript H represents and takes conjugate transpose, subscript -1 represent take inverse, | | table Show Modulus of access, ν represents the form parameter of texture component, subscript γ represents exponential factor, γ is any real number more than 0, value For γ=2；

(6) false-alarm probability f being given according to system, calculates detection threshold T by Monte Carlo experiment_ξ；

(7) according to statistic of test ξ_kWith detection threshold T_ξJudge that target whether there is：If ξ_k≥T_ξ, then show b-th echo Data block X_bRange cell to be detected have target, if ξ_k<T_ξ, then show b-th echo data block X_bDistance to be detected single Unit does not have target.

2. the method for claim 1 is it is characterised in that utilize sample covariance matrix method of estimation meter in step (3) Calculate range cell z to be detected_kCovariance matrixCarry out as follows：

(3a) remove echo data block X_bRange cell z to be detected_kAnd its two adjacent range cells, with remaining L away from From unit as range cell z to be detected_kReference distance unit；

(3b) sample covariance matrix method of estimation is utilized to calculate range cell z to be detected_kCovariance matrix

${\hat{R}}_k=\frac{1}{L}\underset{q=1}{\overset{L}{\&Sigma;}}{z}_q{z}_q^H$

Wherein, z_qRepresent q-th reference distance unit.

3. the method for claim 1 is it is characterised in that calculate miscellaneous noise ratio CNR of echo data, by such as in step (4) Lower step is carried out：

(4a) Doppler frequency f according to given target_s, calculate normalization Doppler frequency f_d：

$f_d=\frac{f_{s}t}{2},$

Wherein, t represents radar pulse transmit cycle；

(4b) utilize normalization Doppler frequency f_d, calculate the Doppler steering vector p of target：

$p={\&lsqb;1,e^{j4{\mathrm{\&pi;f}}_d},e^{j8{\mathrm{\&pi;f}}_d},...,e^{j4\mathrm{\&pi;}(N-1)f_d}\&rsqb;}^T$

Wherein, subscript T represents and takes transposition；

(4c) the Doppler steering vector p of target and the covariance matrix of range cell to be detected are utilizedCalculate echo Miscellaneous noise ratio CNR of data：

$CNR=\frac{\mathrm{\&mu;}v}{{\mathrm{\&sigma;}}^2}\&times;\frac{p^H{\hat{R}}_{k}p}{p^{H}p}$

Wherein, μ represents the scale parameter of texture component, and ν represents the form parameter of texture component, σ²Represent white Gaussian noise power,Represent the power ratio of K Distribution Clutter and white Gaussian noise in echo data, subscript H represents and takes conjugate transpose.

4. the method for claim 1 is it is characterised in that according to false-alarm probability f that system is given in step (6), pass through Monte Carlo experiment calculates detection threshold T_ξ, carry out as follows：

(6a) C is made to be the natural number more than 100/f setting, value is C=10⁶, emulate the C range cell without target and enter C Monte Carlo experiment of row, calculates the statistic of test of each range cell：

${\mathrm{\&xi;}}_w=\frac{|p^H{{\hat{R}}_k}^{-1}{z}_w{|}^2}{\left(p^H{{\hat{R}}_k}^{-1}p\right)(1+{\mathrm{CNR}}^{\mathrm{\&gamma;}}{\left({z_w}^H{{\hat{R}}_k}^{-1}{z}_w\right)}^{\frac{N}{N+v}})}$

Wherein, z_wRepresent w-th range cell, ξ_wRepresent the statistic of test of w-th range cell；

(6b) C obtained statistic of test is arranged in descending order, take [Cf] the individual statistic of test after arrangement as detection Thresholding T_ξ, the wherein maximum integer less than real number Cf for [Cf] expression.