CN104977585B - A kind of motion sonar target detection method of robust - Google Patents

Abstract

The invention relates to a robust moving sonar target detection method. In one embodiment, the method includes: obtaining a set of echo data by the sonar array and selecting main data and auxiliary data having the same reverberation covariance matrix; Maximum likelihood estimation of covariance matrix of statistical properties; maximum likelihood estimation of target reflection intensity is computed from primary data, then supplemented with target nominal steering vector, using Durbin test criterion robust to steering vector mismatch Calculate the detection statistics; judge whether the target exists according to the comparison between the detection statistics and the threshold determined by the false alarm probability. The invention makes more full use of the echo data by introducing the Durbin test criterion, thereby greatly improving the robustness of the space-time adaptive detector in the case of the mismatch of the target steering vector, and has the ability to deal with the unknown reverberation background Constant false alarm feature.

Description

A Robust Detection Method for Moving Sonar Targets

technical field

The invention relates to a target detection method, in particular to a robust moving sonar target detection method.

Background technique

Ocean reverberation is the main interference of shallow sea active sonar, especially when the sonar carrier has a certain speed, the reverberation in different azimuths has different Doppler frequency shifts, so that the sonar element level reverberation Spreading phenomenon appears on the frequency spectrum. In this way, the low-speed moving target signal will be covered by reverberation, and Doppler cannot be used for reverberation suppression, and even if conventional beamforming is used, it is difficult to effectively eliminate the reverberation entering the receiver from side lobes. Due to the space-time coupling characteristics of motion sonar reverberation, its target detection problem has brought up a new topic for sonar workers.

The generation mechanism and some characteristics of moving sonar reverberation such as underwater vehicles are very similar to the ground clutter of airborne radar. In 1973, Brennan proposed the concept of space-time adaptive processing (STAP) for the first time. Their research shows that STAP can combine the advantages of airspace and time domain processing, effectively compensate the radar platform motion effect, and obtain ideal Clutter suppression performance. Then Reed proposed the sample covariance matrix inversion (SMI) detector, thus theoretically developing STAP into an organic combination of filtering and detection method, called space-time adaptive detection (STAD) method. In recent years, research on STAD in the field of motion sonar is very active. The research results show that the STAD method can make full use of the space-time distribution characteristics of reverberation and improve the detection performance of motion sonar.

For the STAD of point targets in the background of Gaussian distribution reverberation, the optimal consistent maximum potential test does not exist, so people have proposed many suboptimal solutions based on the generalized likelihood ratio (GLRT), Rao and other test criteria. It is worth noting that these methods are based on two assumptions. One is to assume that sufficient uniform auxiliary data can be obtained to estimate the reverberation covariance matrix of the unit under test (primary data), thereby constructing an adaptive detection Statistics. In order to ensure uniformity, auxiliary data is generally obtained from distance units adjacent to the main data. The second is to assume that the direction of the target is known, that is, the steering vector of the target is known, which is called the nominal steering vector.

In practical applications, the robustness of existing STAD methods needs to be improved. The actual steering vector of the sonar target and the nominal steering vector often have a mismatch, the reasons include the deviation of the beam pointing, the calibration error of the sonar array, underwater multi-path propagation and so on. When this mismatch occurs, existing methods suffer from significant detection loss.

Contents of the invention

The purpose of the present invention is to realize more fully utilization of echo data, greatly improve the robustness of the space-time adaptive detector (STAD) under the mismatch situation of the target steering vector, and has the ability to deal with the unknown reverberation background. Constant false alarm feature.

In order to achieve the above purpose, an embodiment of the present invention provides a robust moving sonar target detection method. The methods include:

Obtain a set of echo data through echoes received by the sonar array, select one echo data in the set of echo data as main data, and select one set of echo data except the main data in the set of echo data Multiple echo data are used as auxiliary data, and the main data and the auxiliary data have the same reverberation covariance matrix;

The main parameter in the Durbin detection is formed by the real part and the imaginary part of the parameter representing the reflection intensity of the target, and the auxiliary parameter in the Durbin detection is formed by the reverberation covariance matrix, and the main parameter is calculated by using the main data and the auxiliary data. parameter and the maximum likelihood estimation of said auxiliary parameter;

Using the estimated value of the main parameter and the estimated value of the auxiliary parameter, the value of the derivative of the likelihood function to the main parameter under the target assumption is obtained;

Calculate the detection statistic according to the maximum likelihood estimation of the main parameter under the target assumption and the true value sum under the non-target assumption, the derivative and the nominal steering vector obtained by calculating the target direction;

The detection statistic is compared with a threshold obtained by a given false alarm probability, and whether the target exists is judged according to the comparison result.

Preferably, the sampling covariance matrix is calculated according to the auxiliary data.

Preferably, the derivative A(θ) can be expressed as:

Among them, z is the main data, Z is the NxN dimensional auxiliary data, θ _A is the main parameter, K represents the number of uniform auxiliary data constituting the auxiliary data, and H ₁ represents the target situation.

Preferably, the threshold is obtained by setting false alarm probability and using Monte Carlo simulation.

Preferably, comparing the detection statistic with a threshold obtained by a given false alarm probability can be determined by the following formula:

where η is the detection threshold, H ₀ represents the case of no target, H ₁ represents the case of a target, ^H represents the conjugate transpose operation, ^-1 represents the matrix inverse operation, v is the nominal steering vector of the target, and S is the sampling covariance matrix , z means master data.

The present invention adopts the Monte-Carlo simulation method in performance detection, and compares it with traditional GLRT and Rao detectors.

The present invention adopts the Monte-Carlo simulation method in performance detection, and compares it with traditional GLRT and Rao detectors. It is shown by comparison that the method of the present invention realizes more full utilization of observation data, not only greatly improves the robustness of STAD in the case of target-steering vector mismatch, but also maintains very good stability when the target-steering vector matches. Good detection performance.

Description of drawings

Fig. 1 is when the present invention is at SRR=20dB, the relationship curve of detection probability P _d and cos ² φ of traditional GLRT, traditional Rao and the present invention Durbin three kinds of method detectors;

Fig. 2 is the relationship curves of detectors P _d and SRR of the traditional GLRT, traditional Rao and the present invention's Durbin three methods in the case of target-oriented vector matching in the present invention.

detailed description

The invention provides a new space-time self-adaptive detection method for a point target, which uses binary hypothesis data combined with the Durbin test principle to obtain a detection principle formula based on the Durbin test, thereby detecting whether the target exists.

The specific implementation steps are as follows:

1) Based on the echo data received by a linear sonar array composed of N array elements, the point target detection can be attributed to the following binary assumptions:

Among them, H ₀ and H ₁ respectively represent the hypothesis of no target and the hypothesis of target existence; n and n _t , t=1,...,K are independent, zero-mean N-dimensional composite Gaussian reverberation vectors, and their covariance matrix is E[nn ^H ]=E[ _nt n _t ^H ]=M, ^H represents the conjugate transpose operation; z represents an echo data, also known as the main data; z _t , t=1,...,K represents Uniform auxiliary data of length K, and has the same reverberation covariance matrix as the main data z; α is the target reflection intensity parameter, its real part and imaginary part constitute the main parameter in the Durbin detection; v is the target direction Computed nominal steering vector.

In order to facilitate the design of the detector, two simplified formulas are defined, that is, the 2-dimensional signal main parameter vector θ _A = [α _R ,α _I ] ^T , where α _R and α _I are the real part and imaginary part of α respectively; N ² +2 dimensional vector Among them, θ _B is an N ² -dimensional redundant parameter column vector, also known as an auxiliary parameter, which is composed of elements of the covariance matrix M. Thus, the joint probability density function of the main data z and auxiliary data Z=[=z ₁ ,z ₂ ,...,z _k ] in the case of H ₁ is

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)

Where S is the sampling covariance matrix, that is, S=ZZ ^H , det( ) and tr( ) represent the determinant and matrix trace of the matrix respectively, and the auxiliary data Z is NxN dimensional auxiliary data, and has the same Reverb covariance matrix.

In order to realize more fully utilization of observed data, the present invention adopts Durbin test criterion

The detection statistics formed according to the main parameter θ _A , auxiliary parameter θ _B , and derivative A(θ) are compared with the fixed threshold η. The threshold controls the false alarm rate. Using the Durbin test, it can be expressed as:

Wherein, adopt Monte Carlo simulation to obtain the threshold value, θ _A,0 is the true value of θ _A under the H ₀ situation, namely θ _A,0 =[00] ^T ; is the maximum likelihood estimate of θ _B in the case of H ₀ ; is the maximum likelihood estimate of θ _A in the case of H ₁ , so the derivative A(θ) can be expressed as:

noticed

Among them, Re[·] and Im[·] represent the real part and imaginary part of the data in [·] respectively. Substituting formula (4) into formula (3), we get

where I ₂ is the 2nd order identity matrix, is H _i , the maximum likelihood estimation of M in the case of i=0,1.

The present invention makes full use of the main data z and auxiliary data Z to calculate the maximum likelihood estimation of M through the Durbin test Expressed as:

Will Substituting into formula (5), the expression of A(θ) including main data z, auxiliary data Z and nominal steering vector v of the target is obtained:

In the case that H ₁ has a target, the maximum likelihood estimate of the target reflection intensity parameter α is

Substituting formula (7) and formula (8) into formula (2), the principle formula based on Durbin test is obtained:

where η is an appropriate modification of the detection threshold in equation (2).

From the principle formula of the Durbin test, it can be seen that the observation data of the Durbin detector can be expressed as the relationship between the traditional generalized likelihood probability detection (GLRT) and adaptive matched filter detection (AMF) two detection statistics:

t _Durbin = t _AMF (1-t _GLRT ) (9)

in,

Equation (9) has constant persistence. Since (t _AMF , t _GLRT ) is a set of maximum invariant statistics, any detection statistic composed of this set of invariants has constant false alarm performance. Therefore, the Durbin detector proposed by the present invention has The constant false alarm performance of the variance matrix is convenient for the application of actually detecting the target signal.

2) The performance detection of the present invention adopts the Monte-Carlo simulation method, and compares it with traditional GLRT and Rao detectors. The target nominal steering vector v=[1,…,1] ^T /N, and the signal-to-mix ratio is defined as SRR=v ^H M _- ¹ v. The actual steering vector of the target is denoted by v _m , and the degree of mismatch between it and v is measured by cos ² φ, specifically defined as

cos ² φ=1 represents a matching situation, that is, v _m =v. cos ² φ<1 represents a mismatch situation, and the smaller the cos ² φ value, the greater the mismatch between v _m and v.

The specific parameters in the simulation are set to N=8, K=32, false alarm probability P _fa =10 ^-3 , the reverberation is a common exponentially correlated composite Gaussian vector, and the covariance matrix M=0.9 ^|ij| , where (i, j) is the coordinate of the matrix element.

Fig. 1 is when SRR=20dB, the detection probability P _d of the present invention, traditional GLRT and traditional Rao three kinds of method detectors and the relationship curve of steering vector mismatch degree cos ² phi, as can be seen, when target steering vector appears relatively When there is a large mismatch, the method of the present invention can still maintain a high detection probability. For example, when cos ² φ=0.4, the P _d of the inventive method=1.0, while the P _d of the traditional GLRT ≈0.77, and the P _d of the traditional Rao<0.05; and when cos ² φ=0.1, the P _d of the inventive method = 0.67, _Pd <0.06 for conventional _GLRT , Pd <0.01 for conventional Rao.

Fig. 2 is under the target-oriented vector matching situation, the relationship curve of the detection probability P _d and the signal-to-mixing ratio SRR of the present invention, traditional GLRT and traditional Rao three kinds of method detectors, as can be seen, under the matching situation, the present invention's method compares with two The detection loss of the traditional method is very small, and the detection loss is within 0.3dB when the signal-to-mix ratio is low, and the method of the present invention has detection performance equivalent to the traditional GLRT method when the signal-to-mix ratio is high, and is obviously better than the traditional Rao method.

The comparison results of Fig. 1 and Fig. 2 show that the method of the present invention realizes more fully utilization of observation data, not only greatly improves the robustness of STAD in the case of target-steering vector mismatch, but also in the case of target-steering vector matching, The method of the present invention maintains very good detection performance.

The specific embodiments described above have further described the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above descriptions are only specific embodiments of the present invention and are not intended to limit the scope of the present invention. Protection scope, within the spirit and principles of the present invention, any modification, equivalent replacement, improvement, etc., shall be included in the protection scope of the present invention.

Claims (5)

1. A robust motion sonar target detection method comprising: Obtain a set of echo data through echoes received by the sonar array, select one echo data in the set of echo data as main data, and select one set of echo data except the main data in the set of echo data Multiple echo data are used as auxiliary data, and the main data and the auxiliary data have the same reverberation covariance matrix; The main parameter in the Durbin detection is formed by the real part and the imaginary part of the parameter representing the reflection intensity of the target, and the auxiliary parameter in the Durbin detection is formed by the reverberation covariance matrix, and the main parameter is calculated by using the main data and the auxiliary data. parameter and the maximum likelihood estimation of said auxiliary parameter; Using the estimated value of the main parameter and the estimated value of the auxiliary parameter, the value of the derivative of the likelihood function to the main parameter under the target assumption is obtained; calculating a detection statistic according to the main parameter, the auxiliary parameter, the derivative, and the nominal steering vector calculated through the target direction; The detection statistic is compared with a threshold obtained by a given false alarm probability, and whether the target exists is judged according to the comparison result. 2. The detection method according to claim 1, wherein the sampling covariance matrix is calculated according to the auxiliary data. 3. detection method according to claim 1, is characterized in that, derivative A (θ) is expressed as: Among them, z is the main data, Z is N*N dimensional auxiliary data, θ _A is the main parameter, K represents the number of uniform auxiliary data constituting the auxiliary data, and H ₁ represents the target situation. 4. The detection method according to claim 1, wherein the threshold is obtained by setting the false alarm probability and adopting Monte Carlo simulation. 5. detection method according to claim 1, is characterized in that, described detection statistic is compared with the threshold value that obtains by given false alarm probability, is determined by following formula: ${\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; b</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>\frac{{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; v</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; S</span>}}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; v</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; S</span>}}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; v</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; zz</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; \&rsqb;</span>\underset{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{\overset{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{\begin{array}{c}<span\; class="notranslate">\; ></span>\\ <span\; class="notranslate">\; <</span>\end{array}}}\mathrm{<span\; class="notranslate">\; \&eta;</span>}$ where η is the detection threshold, H ₀ means no target, H ₁ means there is a target, H represents the conjugate transpose operation, -1 represents the matrix inverse operation, v is the nominal steering vector of the target, and S is the sampling covariance matrix , z means master data.