CN104977585A - Robust motion sonar target detection method - Google Patents

Abstract

The invention relates to a robust motion sonar target detection method. In one embodiment, the method comprises the steps that a set of echo data are acquired via a sonar array and main data and auxiliary data with the same reverberation covariance matrix are selected from the echo data; the maximum likehood estimation of the covariance matrix representing the reverberation statistical characteristics is calculated according to the main data and the auxiliary data; the maximum likehood estimation of target reflection intensity is calculated according to the main data, and then detection statistical amount is calculated by utilizing Durbin inspection guidelines with mismatching of robustness to a guide vector through complementation of a target nominal guide vector; and existence of the target is judged according to comparison of the detection statistical amount and a threshold value confirmed by false alarm probability. Utilization of the echo data is enabled to be more fully by introduction of the Durbin inspection guidelines so that robustness of a space-time self-adaptive detector under the situation of mismatching of the target guide vector is greatly enhanced, and the method has the constant false alarm characteristic for the unknown reverberation background.

Description

A kind of motion sonar target detection method of robust

Technical field

The present invention relates to a kind of object detection method, particularly relate to a kind of motion sonar target detection method of robust.

Background technology

Marine reverberation is the main interference of shallow sea active sonar, and especially when sonar carrier has certain movement velocity, the reverberation of different azimuth has different Doppler shifts, thus makes the reverberation of sonar array element level present expansion on frequency spectrum.The echo signal of such low-speed motion will cover by reverberation, Doppler cannot be utilized to carry out Reverberation Rejection, even and if adopt conventional beamformer to be also difficult to effectively eliminate the reverberation being entered receiver by secondary lobe.Due to coupled characteristic during motion sonar reverberation empty, its target detection problems is that sonar worker proposes new problem.

Mechanism of production and the part of properties of the motion sonar reverberation such as submarine navigation device are closely similar with the ground clutter of airborne radar.1973, Brennan proposes the concept of space-time adaptive process (STAP) first, and their research shows, STAP can be good in conjunction with spatial domain and Time Domain Processing advantage separately, the Platform movement effect of effective compensation radar, thus obtain desirable clutter recognition performance.Reed proposes sample covariance matrix and to invert (SMI) detecting device subsequently, thus in theory STAP is developed into a kind of filtering and detect the method organically combined, and is called that space-time adaptive detects (STAD) method.In recent years, STAD is very active in the research in motion sonar field, and distribution character when result of study shows that STAD method can make full use of reverberation empty, improves the detection perform of motion sonar.

For the STAD of point target under Gaussian distribution Reverberation, optimum consistent maximal potential inspection is non-existent, and therefore people propose the solution of many suboptimums based on test criterions such as Generalized Likelihood Ratio (GLRT), Rao.It should be noted that these methods are based upon on the basis of two assumed conditions, one is that hypothesis can obtain sufficient even auxiliary data, in order to estimate the reverberation covariance matrix of to-be-measured cell (master data), thus structure self-adapting detecting statistic.For ensureing homogeneity, the range unit that auxiliary data is generally closed on from master data obtains.Two is that the direction of hypothetical target is known, and namely the guiding vector of target is known, is referred to as nominal guiding vector.

In actual applications, the robustness of existing STAD method has much room for improvement.Actual guiding vector and the nominal guiding vector of sonar target often there will be mismatch, and its reason comprises deviation, the calibration error of sonar transducer array, under water Multi-path propagation etc. of beam position.When this mismatch condition occurs, existing method can suffer no small Detectability loss.

Summary of the invention

The object of the invention is to realize utilizing more fully echo data, significantly improve the robustness of space-time adaptive detecting device (STAD) detecting device under goal orientation vector mismatch condition, and there is the CFAR characteristic to unknown Reverberation.

For achieving the above object, a kind of motion sonar target detection method of robust is embodiments provided.Described method comprises:

One group of echo data is obtained by the echo received by sonar battle array, choose an echo data in described one group of echo data as master data, choose multiple echo datas in described one group of echo data except described master data as auxiliary data, described master data and described auxiliary data have identical reverberation covariance matrix;

The main parameter in guest Du detection is formed by the real part and imaginary part that characterize target strength parameter, form the auxiliary parameter in guest Du detection by reverberation covariance matrix, utilize described master data, described auxiliary data to calculate the maximal possibility estimation of described main parameter and described auxiliary parameter;

Utilize the estimated value of the estimated value of described main parameter and described auxiliary parameter, under obtaining goal hypothesis, likelihood function is to the value of the difference quotient of main parameter;

According to the true value of described main parameter under the maximal possibility estimation had under goal hypothesis and driftlessness hypothesis and, described difference quotient and the nominal guiding vector that calculated by target direction, calculate detection statistic;

Described detection statistic and the threshold value obtained by given false-alarm probability are compared, and judges whether described target exists according to comparative result.

Preferably, described sample covariance matrix calculates according to described auxiliary data.

Preferably, described difference quotient A (θ) can be expressed as:

wherein z is master data, and Z is that NxN ties up auxiliary data, θ _abe main parameter, K represents the number of the even auxiliary data forming auxiliary data, H ₁indicate target conditions.

Preferably, described threshold value adopts Monte-Carlo Simulation to obtain by setting false-alarm probability.

Preferably, described described detection statistic and the threshold value to be obtained by given false-alarm probability to be compared, can be determined by following formula:

Wherein η is detection threshold, H ₀represent driftlessness situation, H ₁indicate target conditions, ^hrepresent conjugate transposition operation, ^-1represent matrix inversion operation, v is the nominal guiding vector of target, and S is sample covariance matrix, and z represents master data.

The present invention adopts Meng Te-Caro emulation mode in Performance Detection, compares with traditional GLRT and Rao detecting device.

The present invention adopts Meng Te-Caro emulation mode in Performance Detection, compares with traditional GLRT and Rao detecting device.Show that the inventive method achieves by comparing observation data is utilized more fully, not only significantly improve the robustness of STAD under goal orientation vector mismatch condition, and in goal orientation Vectors matching situation, the inventive method maintains again extraordinary detection perform.

Accompanying drawing explanation

Fig. 1 be the present invention when SRR=20dB, the detection probability P of traditional GLRT, traditional Rao and Durbin of the present invention tri-kinds of method detecting devices _dand cos ²the relation curve of φ;

Fig. 2 be the present invention in goal orientation Vectors matching situation, traditional GLRT, traditional Rao and Durbin of the present invention tri-kinds of method detecting device P _dwith the relation curve of SRR.

Embodiment

The invention provides a kind of new point target space-time adaptive detection method, utilize dualism hypothesis data in conjunction with Durbin inspection principle, draw the Cleaning Principle formula based on Durbin inspection, detect target thus and whether exist.

Specific implementation step is as follows:

1) receive echo data based on the linear sonar battle array be made up of N number of array element, Point Target Detection can be summed up as following dualism hypothesis thus:

$\left\{\begin{array}{c}{H}_0:\left\{\begin{array}{c}z=n\\ z_t=n_t,t=1,...,K\end{array}\right.\\ H_1:\left\{\begin{array}{c}z=\mathrm{\αv}+n,\\ z_t=n_t,t=1,...,K\end{array}\right.\end{array}\right.---\left(1\right)$

Wherein H ₀and H ₁represent driftlessness hypothesis respectively and have target to there is hypothesis; N and n _t, t=1 ..., K be independently, zero-mean N ties up complex Gaussian reverberation vector, its covariance matrix is E [nn ^h]=E [n _tn _t ^h]=M, ^hrepresent conjugate transposition operation; Z represents an echo data, also known as master data; z _t, t=1 ..., K represents that length is the even auxiliary data of K, and has identical reverberation covariance matrix with master data z; α is target strength parameter, and its real part and imaginary part form the main parameter in guest Du detection; V is the nominal guiding vector calculated by target direction.

For the ease of the design of detecting device, define two simplified styles, i.e. the signal of 2 dimensions main parameter vector θ _a=[α _r, α _i] ^t, wherein α _rand α _ireal part and the imaginary part of α respectively; N ²+ 2 dimensional vectors wherein θ _bfor N ²the nuisance parameter column vector of dimension, also known as auxiliary parameter, is made up of the element of covariance matrix M.Thus, H ₁described master data z and auxiliary data Z=[=z in situation ₁, z ₂..., z _k] joint probability density function be

f(z,Z|θ,H ₁)＝π ^-N(K+1)det(M) ^-(K+1)exp{-tr[M ^-1((z-αv)(z-αv) ^H+S)]} (2)

Wherein S is sample covariance matrix, i.e. S=ZZ ^h, det () and tr () represents determinant of a matrix and trace of a matrix respectively, and auxiliary data Z is that NxN ties up auxiliary data, and has identical reverberation covariance matrix with master data z.

In order to realize utilizing more fully observation data, the present invention adopts Durbin test criterion

According to main parameter θ _a, auxiliary parameter θ _b, the detection statistic that formed of difference quotient A (θ) compares with fixing threshold value η, threshold value controls false alarm rate, and utilizing Durbin to check can be expressed as:

Wherein, Monte-Carlo Simulation is adopted to obtain threshold value, θ _{a, 0}h ₀θ in situation _atrue value, i.e. θ _{a, 0}=[00] ^t; h ₀θ in situation _bmaximal possibility estimation; h ₁θ in situation _amaximal possibility estimation, being therefore expressed as of difference quotient A (θ):

$\begin{array}{c}A\left(\mathrm{\θ}\right)=-\frac{1}{K+1}\frac{\∂\ln f(z,Z|\mathrm{\θ},H_1)}{{\∂\mathrm{\θ}}_A{\mathrm{\θ}}_A^T}\\ =-\frac{1}{K+1}\left[\begin{array}{cc}\frac{{\∂}^2\ln f(z,Z|\mathrm{\θ},H_1)}{{\∂\mathrm{\α}}_R{\∂\mathrm{\α}}_R}& \frac{{\∂}^2\ln f(z,Z|\mathrm{\θ},H_1)}{{\∂\mathrm{\α}}_R{\∂\mathrm{\α}}_I}\\ \frac{{\∂}^2\ln f(z,Z|\mathrm{\θ},H_1)}{{\∂\mathrm{\α}}_I{\∂\mathrm{\α}}_R}& \frac{{\∂}^2\ln f(z,Z|\mathrm{\θ},H_1)}{{\∂\mathrm{\α}}_I{\∂\mathrm{\α}}_I}\end{array}\right]\end{array}---\left(3\right)$

Notice

$\frac{\∂\ln f(z,Z|\mathrm{\θ},H_1)}{{\∂\mathrm{\θ}}_A}=\left[\begin{array}{c}{v}^H{M}^{-1}(z-\mathrm{\αv})+{(z-\mathrm{\αv})}^H{M}^{-1}v\\ {-\mathrm{jv}}^H{M}^{-1}(z-\mathrm{\αv})+j{(z-\mathrm{\αv})}^H{M}^{-1}v\end{array}\right]=2\left[\begin{array}{c}\mathrm{Re}\left[v^H{M}^{-1}(z-\mathrm{\αv})\right]\\ \mathrm{Im}\left[v^H{M}^{-1}(z-\mathrm{\αv})\right]\end{array}\right]---\left(4\right)$

Wherein Re [] and Im [] represents the real part and imaginary part of getting [] interior data respectively.Formula (4) is substituted into formula (3), draws

${A\left(\mathrm{\θ}\right)}_{\mathrm{\θ}={\widehat{\mathrm{\θ}}}_i}=-\frac{1}{K+1}{v}^H{\hat{M}}_i^{-1}{\mathrm{vI}}_2---\left(5\right)$

Wherein I ₂be 2 rank unit matrix, h _i, i=0, the maximal possibility estimation of M in 1 situation.

The present invention is checked by Durbin, takes full advantage of master data z and auxiliary data Z to calculate the maximal possibility estimation of M be expressed as:

${\hat{M}}_0=(S+{\mathrm{zz}}^H)/(K+1)---\left(6\right)$

Will substitution formula (5), obtains the expression formula of the A (θ) of the nominal guiding vector v containing master data z, auxiliary data Z and target:

${A\left(\mathrm{\θ}\right)|}_{\mathrm{\θ}={\widehat{\mathrm{\θ}}}_0}=v^H{(S+{\mathrm{zz}}^H)}^{-1}{\mathrm{vI}}_2---\left(7\right)$

At H ₁under having target conditions, the maximal possibility estimation of target strength parameter alpha is

$\widehat{\mathrm{\α}}=\frac{v^H{S}^{-1}z}{v^H{S}^{-1}v}---\left(8\right)$

Formula (7) and formula (8) are substituted into formula (2), obtain the principle type based on Durbin inspection:

Wherein η is a suitably amendment of detection threshold in formula (2).

The principle type checked by Durbin, can find out that the observation data of Durbin detecting device can be expressed as the relational expression of traditional Generalized Likelihood Probability Detection (GLRT) and adaptive matched filter detection (AMF) two kinds of detection statistic:

t _Durbin＝t _AMF(1-t _GLRT) (9)

Wherein, $t_{\mathrm{AMF}}=\frac{{\left|v^H{S}^{-1}z\right|}^2}{v^H{S}^{-1}v},t_{\mathrm{GLRT}}=\frac{{\left|v^H{S}^{-1}z\right|}^2}{(1+z^H{S}^{-1}z)v^H{S}^{-1}v},$

Formula (9) has permanent persistence.Due to (t _aMF, t _gLRT) be one group of maximal invariant statistic, any detection statistic formed with this group invariant CFAR performance all, therefore, the Durbin detecting device that the present invention proposes has unknown reverberation covariance matrix CFAR performance, is convenient to the application of actual detection echo signal.

2) Performance Detection of the present invention adopts Meng Te-Caro emulation mode, and compares with traditional GLRT and Rao detecting device.Target nominal steering vector v=[1 ..., 1] ^t/ N, letter is mixed than being defined as SRR=v ^hm _- ¹v.The actual guiding vector v of target _mrepresent, the mismatch cos between it and v ²φ weighs, and is specifically defined as

${\cos }^2\mathrm{\φ}=\frac{{\left|v_m^H{M}^{-1}v\right|}^2}{\left(v_m^H{M}^{-1}{v}_m\right)\left(v^H{M}^{-1}v\right)}---\left(8\right)$

Cos ²φ=1 represents match condition, i.e. v _m=v.Cos ²φ <1 represents mismatch condition, and cos ²φ value is less, v _mwith mismatch between v is larger.

Design parameter in emulation is set to N=8, K=32, false-alarm probability P _fa=10 ^-3, reverberation is common correlation of indices complex Gaussian vector, covariance matrix M=0.9 ^{| i-j|}, wherein (i, j) coordinate that is matrix element.

Fig. 1 is when SRR=20dB, the detection probability P of the present invention, traditional GLRT and traditional Rao tri-kinds of method detecting devices _dwith guiding vector mismatch cos ²the relation curve of φ, can find out, when larger mismatch condition appears in goal orientation vector, the inventive method still can keep higher detection probability.Such as work as cos ²during φ=0.4, the P of the inventive method _d=1.0, and the P of traditional GLRT _d≈ 0.77, the P of traditional Rao _d<0.05; And work as cos ²during φ=0.1, the P of the inventive method _d=0.67, the P of traditional GLRT _d<0.06, the P of traditional Rao _d<0.01.

Fig. 2 is in goal orientation Vectors matching situation, the detection probability P of the present invention, traditional GLRT and traditional Rao tri-kinds of method detecting devices _drelation curve than SRR mixed with letter, can find out, under match condition, the inventive method is very little relative to the Detectability loss of two kinds of classic methods, low confidence mix than time Detectability loss within 0.3dB, high letter mixed than time the inventive method there is the detection perform suitable with traditional GLRT method, be obviously better than traditional Rao method.

Show that the inventive method achieves by the comparative result of Fig. 1 and Fig. 2 to utilize more fully observation data, not only significantly improve the robustness of STAD under goal orientation vector mismatch condition, and in goal orientation Vectors matching situation, the inventive method maintains again extraordinary detection perform.

Above-described embodiment; object of the present invention, technical scheme and beneficial effect are further described; be understood that; the foregoing is only the specific embodiment of the present invention; the protection domain be not intended to limit the present invention; within the spirit and principles in the present invention all, any amendment made, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (5)

1. a motion sonar target detection method for robust, comprising:

One group of echo data is obtained by the echo received by sonar battle array, choose an echo data in described one group of echo data as master data, choose multiple echo datas in described one group of echo data except described master data as auxiliary data, described master data and described auxiliary data have identical reverberation covariance matrix;

The main parameter in guest Du detection is formed by the real part and imaginary part that characterize target strength parameter, form the auxiliary parameter in guest Du detection by reverberation covariance matrix, utilize described master data, described auxiliary data to calculate the maximal possibility estimation of described main parameter and described auxiliary parameter;

Utilize the estimated value of the estimated value of described main parameter and described auxiliary parameter, under obtaining goal hypothesis, likelihood function is to the value of the difference quotient of main parameter;

According to described main parameters, described auxiliary parameter, described difference quotient and the nominal guiding vector that calculated by target direction, calculate detection statistic;

Described detection statistic and the threshold value obtained by given false-alarm probability are compared, and judges whether described target exists according to comparative result.

2. detection method according to claim 1, is characterized in that, described sample covariance matrix calculates according to described auxiliary data.

3. detection method according to claim 1, is characterized in that, described difference quotient A (θ) can be expressed as:

$A\left(\mathrm{\&theta;}\right)=-\frac{1}{K+1}\frac{\&PartialD;\ln f(z,Z|\mathrm{\&theta;},H_1)}{{\&PartialD;\mathrm{\&theta;}}_A{\mathrm{\&theta;}}_A^T},$ Wherein z is master data, and Z is that NxN ties up auxiliary data, θ _abe main parameter, K represents the number of the even auxiliary data forming auxiliary data, H ₁indicate target conditions.

4. detection method according to claim 1, is characterized in that, described threshold value adopts Monte-Carlo Simulation to obtain by setting false-alarm probability.

5. detection method according to claim 1, is characterized in that, describedly described detection statistic and the threshold value to be obtained by given false-alarm probability is compared, and can be determined by following formula:

Wherein η is detection threshold, H ₀represent driftlessness situation, H ₁indicate target conditions, ^hrepresent conjugate transposition operation, ^-1represent matrix inversion operation, v is the nominal guiding vector of target, and S is sample covariance matrix, and z represents master data.