CN105548986A - Prewhitening ratio median detector based object detection method in sea cluster background - Google Patents

Abstract

The invention discloses a prewhitening ratio median detector based object detection method in a sea cluster background, mainly for solving the problem of incapability of maintaining constant fault alarm in the prior art. The technical scheme is as follows: an observation vector z of a detection unit and an observation vector z<k>(k=1, 2,..., p) of a reference unit in radar echo data are constructed, and an estimation value M<NSCM><^> of a normalization sample covariance matrix of the P+1 observation vectors is obtained; prewhitening is performed on z and z<k> by use of M<NSCM><^> , an observation vector Z<-> of the detection unit and an observation vector Z<k><-> after the prewhitening are obtained; average power e of Z<-> and average power e<k> of Z<k><-> are calculated; median power e<m> of P reference units by use of e<k> is calculated; a detection statistical amount xi is calculated by use of e and e<m>; and whether an object exists is determined by use of the detection statistical amount xi and an arranged detection threshold T. The object detection method provided by the invention improves the object detection performance and can be applied to object detection in the sea cluster background.

Description

Based on the object detection method of prewhitening median ratio detecting device under sea clutter background

Technical field

The invention belongs to Radar Targets'Detection technical field, be specifically related to the object detection method based on prewhitening median ratio detecting device under a kind of sea clutter background, can be used for the target detection under sea clutter background.

Background technology

Sea clutter is mutually superposed by the back scattering vector of surface scattering body separate in a large number to be formed, and is the key factor affecting sea-surface target detection.During when high-resolution radar is operated in little grazing angle, the amplitude distribution of sea clutter there will be long " hangover " compared with rayleigh distributed, and sea clutter presents stronger non-Gaussian system.Sea clutter K distributed model is the classical statistics model describing non-rayleigh envelope clutter, and the sea clutter under high-resolution radar, little grazing angle, non-homogeneous environment can describe well with sea clutter K distribution.As far back as the forties in last century, CFAR has been just the important research content in radar signal processing field.At present, CFAR etection theory and the method for the whole series of relative maturity is had in the world.Document " Shi Yanling; the performance evaluation of bright .ANMF and CM-CFAR of water roc under K Distribution Clutter. Northwest University's journal; 2012; 42 (3): 359-364. " the middle median ratio detecting device introduced, the method utilizing the sampled value of reference unit to get intermediate value carrys out the power level of estimated background clutter.The use of the method has two important hypotheses: the statistical property of detecting unit clutter is identical with the statistical property of reference unit; Separate between detecting unit and reference unit.But in actual environment, there is stronger correlativity in the echoed signal that radar receives, second hypotheses cannot meet, and causes the performance degradation of median ratio detection method.

Summary of the invention

The object of the invention is to propose the object detection method based on prewhitening median ratio detecting device under a kind of sea clutter background, to improve the detection perform under sea clutter background.

For realizing above-mentioned technical purpose, technical scheme of the present invention comprises as follows:

(1) utilize radar transmitter transponder pulse signal, radar receiver receives the echo data formed through surface scattering, builds the observation vector z of detecting unit and the observation vector z of reference unit in echo data _kfor:

$\left\{\begin{array}{cc}{H}_0:z=c,& z_k=c_k\\ H_1:z=s+c,& z_k=c_k\end{array}\right.,k=1,2,...,P$

Wherein, H ₀representing only has clutter to there is not the hypothesis of target, H ₁indicate clutter and there is the hypothesis of target, c represents the sea clutter vector of detecting unit, c _krepresent the sea clutter vector of reference unit, s represents target echo signal, and P represents reference unit number;

(2) utilize maximum Likelihood, obtain the observation vector z of detecting unit in echo data and the observation vector z of reference unit _kthe estimated value of normalization sample covariance matrix

(3) estimated value of normalization sample covariance matrix is utilized to the observation vector z of detecting unit in echo data and the observation vector z of reference unit _kcarry out albefaction, obtain the observation vector of the detecting unit after albefaction with the observation vector of the reference unit after albefaction

$\tilde{Z}={\hat{M}}_{NSCM}^{-\frac{1}{2}}z,{\tilde{z}}_k={\hat{M}}_{NSCM}^{-\frac{1}{2}}{z}_k;$

(4) suppose to have clutter and there is target, the observation vector of detecting unit after calculating albefaction average power e and albefaction after the observation vector of reference unit average power e _k;

(5) according to the observation vector of reference unit after albefaction average power e _k, calculate the median power e of the observation vector of P reference unit _m:

(5a) by e _kit is made to meet by order arrangement from small to large:

$e_{i_1}\&le;e_{i_2}\&le;...\&le;e_{i_P},1\&le;i_1,i_2,...,i_P\&le;P$

(5b) the median power value e of the observation vector of P reference unit is calculated as follows _m:

(6) according to e and e _m, calculate test statistics ξ:

ξ＝e/e _m；

(7) false-alarm probability P is set _fa, calculate corresponding detection threshold T;

(8) detection statistic ξ and detection threshold T is compared, judge in detecting unit, whether target exists.If ξ >=T, then judge that detecting unit has target, if ξ < is T, then judge that detecting unit does not have target.

The present invention compared with prior art has the following advantages:

1) owing to the present invention is directed to the correlativity of sea clutter, adopt whitening approach to carry out albefaction to sea clutter before detection, improve the detection perform of subsequent detectors.

2) because the present invention utilizes clutter data real-time update sea clutter normalization sample covariance matrix, adaptively can match with the noise performance in testing environment, better detection perform can be obtained.

Accompanying drawing explanation

Fig. 1 is realization flow figure of the present invention;

Fig. 2 is the comparison diagram of target detection probability under different signal to noise ratio of the emulated data obtained with the present invention and existing median ratio detecting device.

Embodiment

Below in conjunction with accompanying drawing, the invention will be further described:

With reference to Fig. 1, performing step of the present invention is as follows:

Step 1, builds the observation vector z of detecting unit and the observation vector z of reference unit in echo data _k.

Radar transmitter transponder pulse signal, radar receiver receives the echo data formed through surface scattering, and this echo data is the three-dimensional data comprising distance dimension, ripple position peacekeeping pulse dimension, each distance peacekeeping ripple position dimension formation resolution element;

In echo data, choose arbitrarily a resolution element as detecting unit, and around detecting unit, choose P resolution element as reference unit;

Build the observation vector z of detecting unit and the observation vector z of reference unit in echo data _k:

When detecting unit only has clutter to there is not target, z=c, z _k=c _k,

When detecting unit has clutter and there is target, z=s+c, z _k=c _k, that is:

$\left\{\begin{array}{cc}{H}_0:z=c,& z_k=c_k\\ H_1:z=s+c,& z_k=c_k\end{array}\right.,k=1,2,...,P---<1>$

Wherein, H ₀representing only has clutter to there is not the hypothesis of target, H ₁indicate clutter and there is the hypothesis of target, c represents the sea clutter vector of detecting unit, c _krepresent the sea clutter vector of reference unit, s represents target echo signal, and P represents reference unit number.

Step 2, utilizes maximum Likelihood, obtains the observation vector z of detecting unit in echo data and the observation vector z of reference unit _kthe estimated value of normalization sample covariance matrix

(2.1) suppose to have clutter and there is target, by the observation vector z of detecting unit in echo data and the observation vector z of reference unit _kcarry out power normalization process respectively, obtain the observation vector of the detecting unit of power normalization with the observation vector of the reference unit of power normalization

$\hat{z}=\frac{z}{\sqrt{\frac{1}{N}{z}^{H}z}},{\hat{z}}_k=\frac{z_k}{\sqrt{\frac{1}{N}{z}_k^H{z}_k}},k=1,2,...,P---<2>$

Wherein, P > 2N, N are accumulation umber of pulse;

(2.2) according to above-mentioned calculate the covariance matrix M of P power normalization reference unit:

$M=E\left\{{\hat{z}}_k{\hat{z}}_k^H\right\}---<3>$

Wherein, () ^hfor getting conjugate transpose, the dimension of M is N × N;

(2.3) according to above-mentioned covariance matrix M, calculate with joint probability density function

$f(\hat{z};{\hat{z}}_k\}=\frac{1}{{\left({\mathrm{\&pi;}}^N\right|\left|M\right|\left|\right)}^{P+1}}\exp (-{\hat{z}}^H{M}^{-1}\hat{z}-\underset{k=1}{\overset{P}{\mathrm{\&Sigma;}}}{\hat{z}}_k^H{M}^{-1}{\hat{z}}_k);---<4>$

(2.4) by joint probability density function get natural logarithm, obtain log-likelihood function

$L(\hat{z};{\hat{z}}_k)=-(P+1)\&lsqb;ln\left({\mathrm{\&pi;}}^N\right)+ln\left(\right|\left|M\right|\left|\right)\&rsqb;-{\hat{z}}^H{M}^{-1}\hat{z}-\underset{k=1}{\overset{P}{\&Sigma;}}{\hat{z}}_k^H{M}^{-1}{\hat{z}}_k;---<5>$

(2.5) by log-likelihood function to M differentiate, and make derivative be zero, obtain the estimated value of sample covariance matrix

${\hat{M}}_{SCM}=\frac{1}{P+1}(\hat{z}{\hat{z}}^H+\underset{k=1}{\overset{P}{\&Sigma;}}{\hat{z}}_k{\hat{z}}_k^H);---<6>$

(2.6) by (2.1) with substitute into the estimated value of above-mentioned sample covariance matrix obtain the estimated value of normalization sample covariance matrix

${\hat{M}}_{NSCM}=\frac{1}{P+1}(\frac{{\mathrm{zz}}^H}{\frac{1}{N}{z}^{H}z}+\underset{k=1}{\overset{P}{\&Sigma;}}\frac{z_k{z}_k^H}{\frac{1}{N}{z}_k^H{z}_k}).---<7>$

Step 3, calculates the observation vector of the detecting unit after albefaction with the observation vector of the reference unit after albefaction

Owing to there is correlativity between detecting unit and reference unit and between reference unit and reference unit in echo data, need to adopt whitening approach to carry out albefaction to detecting unit and reference unit before detection, namely utilize the estimated value of normalization sample covariance matrix to the observation vector z of reference unit in the observation vector z of detecting unit in echo data and echo data _kcarry out albefaction respectively;

Albefaction is carried out to the observation vector z of detecting unit, obtains the observation vector after albefaction

$\tilde{Z}={\hat{M}}_{NSCM}^{-\frac{1}{2}}z---<8>$

To the observation vector z of reference unit _kcarry out albefaction calculating, obtain the observation vector after albefaction

Step 4, calculates test statistics ξ.

(4.1) suppose to have clutter and there is target, calculating the observation vector of the detecting unit after albefaction average power e:

$e=\frac{1}{N}{\tilde{z}}^H\tilde{z}---<10>$

(4.2) suppose to have clutter and there is target, calculating the observation vector of the reference unit after albefaction average power e _k:

$e_k=\frac{1}{N}{\tilde{z}}_k^H{\tilde{z}}_k---<11>$

(4.3) according to the observation vector of reference unit after albefaction average power e _k, calculate the median power e of the observation vector of P reference unit _m;

(4.4) formula e and e is utilized _mcalculate test statistics ξ:

ξ＝e/e _m。<12>

Step 5, arranges false-alarm probability P _fa, calculate corresponding detection threshold T.

Detection threshold T is calculated by Monte Carlo experiment, and its step is as follows:

(5.1) make C be setting be greater than 1 natural number, calculate the detection statistic ξ of the 1st target ₁to the detection statistic ξ of C target _c;

(5.2) by C detection statistic obtaining by descending sort, in C target detection statistic after descending sort, get [CP _fa] individual element value as detection threshold T, [CP _fa] represent be no more than real number CP _famaximum integer.

Step 6, compares detection statistic ξ and detection threshold T, judges in detecting unit, whether target exists.

If ξ >=T, then judge that detecting unit has target, if ξ < is T, then judge that detecting unit does not have target.

Below in conjunction with emulation experiment, effect of the present invention is described further.

1. simulation parameter

The emulated data adopted in experiment is the three-dimensional data comprising distance dimension, ripple position peacekeeping pulse dimension, each distance peacekeeping ripple position dimension formation resolution element, its middle distance dimension 1600, ripple position dimension 600, pulse dimension 5; Signal to noise ratio SCR=[-10dB, 30dB], false-alarm probability P _fa=10 ^-4.

2. emulation experiment content

Adopt the inventive method respectively in emulation experiment, existing median ratio detecting device detects emulated data, and by the Detection results of detection probability com-parison and analysis two kinds of detection methods, detection probability shows that more greatly detecting device detection perform is better.

Emulation experiment step:

First, the transformation range arranging signal to noise ratio is SCR=[-10dB, 30dB];

Then, the resolution element that random selecting 10000 is different in emulated data, and on each resolution element, add the echo signal of corresponding signal to noise ratio;

Finally, under different signal to noise ratio condition, the present invention and existing median ratio detecting device is utilized to carry out target detection to above-mentioned 10000 resolution elements, obtain the change curve of detection probability with signal to noise ratio of the present invention and existing median ratio detecting device, result as shown in Figure 2, in Fig. 2, transverse axis represents signal to noise ratio, and the longitudinal axis represents detection probability.

As can be seen from Figure 2, the detection perform of existing median ratio detecting device is obviously better than under the sea clutter background that proposes of the present invention based on the detection perform of the object detection method of prewhitening median ratio detecting device.

Claims (5)

1. under sea clutter background based on the object detection method of prewhitening median ratio detecting device, it is characterized in that, comprising:

(1) utilize radar transmitter transponder pulse signal, radar receiver receives the echo data formed through surface scattering, builds the observation vector z of detecting unit and the observation vector z of reference unit in echo data _kfor:

$\left\{\begin{array}{cc}{H}_0:z=c,& z_k=c_k\\ H_1:z=s+c,& z_k=c_k\end{array}\right.,k=1,2,...,P$

Wherein, H ₀representing only has clutter to there is not the hypothesis of target, H ₁indicate clutter and there is the hypothesis of target, c represents the sea clutter vector of detecting unit, c _krepresent the sea clutter vector of reference unit, s represents target echo signal, and P represents reference unit number;

(2) utilize maximum Likelihood, obtain the observation vector z of detecting unit in echo data and the observation vector z of reference unit _kthe estimated value of normalization sample covariance matrix

(3) estimated value of normalization sample covariance matrix is utilized to the observation vector z of detecting unit in echo data and the observation vector z of reference unit _kcarry out albefaction, obtain the observation vector of the detecting unit after albefaction with the observation vector of the reference unit after albefaction

$\tilde{z}={\hat{M}}_{NSCM}^{-\frac{1}{2}}z,{\tilde{z}}_k={\hat{M}}_{NSCM}^{-\frac{1}{2}}{z}_k;$

(4) suppose to have clutter and there is target, the observation vector of detecting unit after calculating albefaction average power e and albefaction after the observation vector of reference unit average power e _k;

(5) according to the observation vector of reference unit after albefaction average power e _k, calculate the median power e of the observation vector of P reference unit _m:

(5a) by e _kit is made to meet by order arrangement from small to large:

$e_{i_1}\&le;e_{i_2}\&le;...\&le;e_{i_P},1\&le;i_1,i_2,...,i_P\&le;P$

(5b) the median power value e of the observation vector of P reference unit is calculated as follows _m:

(6) according to e and e _m, calculate test statistics ξ:

ξ＝e/e _m；

(7) false-alarm probability P is set _fa, calculate corresponding detection threshold T;

(8) detection statistic ξ and detection threshold T is compared, judge in detecting unit, whether target exists.If ξ >=T, then judge that detecting unit has target, if ξ < is T, then judge that detecting unit does not have target.

2. under sea clutter background as claimed in claim 1 based on the object detection method of prewhitening median ratio detecting device, it is characterized in that, in described step 2, utilize maximum Likelihood, obtain the observation vector z of detecting unit in echo data and the observation vector z of reference unit _kthe estimated value of normalization sample covariance matrix carry out as follows:

(2.1) suppose to have clutter and there is target, by the observation vector z of detecting unit in echo data and the observation vector z of reference unit _kcarry out power normalization process respectively, obtain the observation vector of the detecting unit of power normalization with the observation vector of the reference unit of power normalization

$\hat{z}=\frac{z}{\sqrt{\frac{1}{N}{z}^{H}z}},$

${\hat{z}}_k=\frac{z_k}{\sqrt{\frac{1}{N}{z}_k^H{z}_k}},k=1,2,...,P$

Wherein, P > 2N, N are accumulation umber of pulse;

(2.2) according to above-mentioned calculate the covariance matrix M of P power normalization reference unit:

$M=E\left\{{\hat{z}}_k{\hat{z}}_k^H\right\}$

Wherein, () ^hfor getting conjugate transpose, the dimension of M is N × N;

(2.3) according to above-mentioned covariance matrix M, calculate with joint probability density function

$f(\hat{z};{\hat{z}}_k)=\frac{1}{{\left({\mathrm{\&pi;}}^N\right|\left|M\right|\left|\right)}^{P+1}}\exp (-{\hat{z}}^H{M}^{-1}\hat{z}-\underset{k=1}{\overset{P}{\&Sigma;}}{\hat{z}}_k^H{M}^{-1}{\hat{z}}_k);$

(2.4) by joint probability density function get natural logarithm, obtain log-likelihood function

$L(\hat{z};{\hat{z}}_k)=-(P+1)\&lsqb;ln\left({\mathrm{\&pi;}}^N\right)+ln\left(\right|\left|M\right|\left|\right)\&rsqb;-{\hat{z}}^H{M}^{-1}\hat{z}-\underset{k=1}{\overset{P}{\&Sigma;}}{\hat{z}}_k^H{M}^{-1}{\hat{z}}_k;$

(2.5) by log-likelihood function to M differentiate, and make derivative be zero, obtain the estimated value of sample covariance matrix

${\hat{M}}_{SCM}=\frac{1}{P+1}(\hat{z}{\hat{z}}^H+\underset{k=1}{\overset{P}{\&Sigma;}}{\hat{z}}_k{\hat{z}}_k^H);$

(2.6) by (2.1) with substitute into the estimated value of above-mentioned sample covariance matrix obtain the estimated value of normalization sample covariance matrix

${\hat{M}}_{NSCM}=\frac{1}{P+1}(\frac{{\mathrm{zz}}^H}{\frac{1}{N}{z}^{H}z}+\underset{k=1}{\overset{P}{\&Sigma;}}\frac{z_k{z}_k^H}{\frac{1}{N}{z}_k^H{z}_k}).$

3. under sea clutter background as claimed in claim 1 based on the object detection method of prewhitening median ratio detecting device, it is characterized in that, in step 4, calculate the observation vector of detecting unit after albefaction average power e, calculated by following formula:

$e=\frac{1}{N}{\tilde{z}}^H\tilde{z}.$

4. under sea clutter background as claimed in claim 1 based on the object detection method of prewhitening median ratio detecting device, it is characterized in that, in step 4, calculate the observation vector of reference unit after albefaction average power e _k, calculated by following formula:

$e_k=\frac{1}{N}{\tilde{z}}_k^H{\tilde{z}}_k.$

5. under sea clutter background as claimed in claim 1 based on the object detection method of prewhitening median ratio detecting device, it is characterized in that, in step 7, false-alarm probability P is set _fa, calculate corresponding detection threshold T, carry out as follows:

(5.1) make C be setting be greater than 1 natural number, calculate the detection statistic ξ of the 1st target ₁to the detection statistic ξ of C target _c;

(5.2) by C detection statistic obtaining by descending sort, in C target detection statistic after descending sort, get [CP _fa] individual element value as detection threshold T, [CP _fa] represent be no more than real number CP _famaximum integer.