Range Extended Target Detection Method in Gaussian Clutter Background
By assuming diagonal symmetric covariance matrix and subspace transformation dimensionality reduction method in the Gaussian clutter background, GLRT and Wald test criterion detectors are constructed, which solves the problem of strong dependence on training data and improves the distance-scaling target detection performance.
Patent Information
- Application Number
- CN202410171968.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-02-07
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-02-07
AI Technical Summary
The prior art has strong dependence on training data in the detection of distance extended targets in Gaussian clutter background, especially in changing terrain and dense target environments, and does not consider the covariance matrix. Symmetric structure and dimensionality reduction method.
Using the method based on GLRT and Wald's test criteria, it is assumed that the clutter covariance matrix has oblique symmetry, and the dimensionality reduction method of subspace transformation is used to establish a distance expansion object detector under the background of Gaussian clutter, and the detection statistic ΛRD-P is constructed. -SGLRT and ΛRD-P-SWald to reduce the need for training data.
While ensuring the constant false alarm rate, the detection performance is significantly improved, the training data sample size required for the adaptive process is reduced, and the performance of distance-extended target detection is improved.
Smart Images

Figure CN118033583B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of radar signal processing, and in particular to a method for detecting a range-extended target under a Gaussian clutter background based on GLRT and Wald test criteria. Background Art
[0002] Signal detection in noise is a fundamental problem in the fields of radar, sonar, communication, underground exploration, etc. It is worth noting that technologies such as radar pulse compression ensure the high range resolution of radar, so that the target energy of high range resolution radar is distributed in different range units, showing a range extension feature, forming a range extension target different from a point target. With the widespread application of broadband technology in the field of radar, the problem of adaptive detection of range extension targets is receiving more and more attention and has become one of the hot and difficult issues in the field of radar signal processing in recent years.
[0003] The research on extended range target detection in Gaussian clutter background is quite extensive. In Gaussian clutter background, the optimal detector composed of a cascade of whitening filters and matched filters can be used to solve the signal detection problem with known clutter covariance. However, in practice, the clutter covariance matrix is usually unknown and needs to be estimated using training data collected from distance cells adjacent to the measured distance cell. As a result, target detection methods based on generalized likelihood ratio test (GLRT), Rao, Wald and other test criteria have been formed. However, the adaptive detectors proposed based on several test criteria still have a great dependence on training data. Moreover, in many practical applications, due to changing terrain, dense target environment, etc., it is difficult to obtain such sufficient independent and identically distributed training data, such as spatiotemporal adaptive processing of multi-sensor and multi-time sampling for airborne radars operating in rapidly changing scenes. Therefore, reducing the dependence of detectors on training data, that is, reducing the demand for training data for detectors, is a hot topic in target detection problems.
[0004] One way to reduce the detector's demand for training data is to use the prior information of the clutter covariance matrix, that is, the structure of the covariance matrix can be determined in advance based on the antenna configuration, where the skew-symmetric structure is one of the structures of the covariance matrix estimation. Using the skew-symmetric structure is equivalent to doubling the amount of training data, which can significantly improve the performance of the adaptive algorithm. It is worth noting that if the prior information deviates from the actual information, the method using prior information may suffer a significant performance loss. Another method is dimensionality reduction, that is, first performing a data-independent dimensionality reduction transformation on the test and training data so that the covariance matrix is projected into a low-dimensional subspace, and then performing adaptive processing to reduce the sample size of the required training data. From the public literature, the existing detectors for extended range targets have not yet taken into account the asymmetric structure of the covariance matrix and the dimensionality reduction method. Therefore, the problem of extended range target detection based on the asymmetric structure of the covariance matrix and the dimensionality reduction method urgently needs further attention and research. Summary of the invention
[0005] In order to overcome the problems in the prior art, the present invention proposes a range extended target detection method in a Gaussian clutter background based on GLRT and Wald test criteria.
[0006] The technical solution of the present invention to solve the above technical problems is as follows:
[0007] A method for detecting extended-range targets under Gaussian clutter background comprises the following steps:
[0008] Step 1. Consider a uniform receiving array with N array elements, obtain the data to be detected from K to-be-detected distance units, and obtain L observation data without target signals from the target-free distance units adjacent to the detected units as training data; perform subspace modeling on the target signal, model the Gaussian clutter as a complex Gaussian distribution with a zero-mean clutter covariance matrix that is unknown, and the clutter covariance matrix satisfies a skew-symmetric structure, and establish a preliminary binary hypothesis testing model;
[0009] Step 2. Use the transformation matrix to take advantage of the skew symmetry of the clutter covariance matrix, and then use the dimensionality reduction method based on subspace transformation to establish the final binary hypothesis testing model;
[0010] Step 3. Based on the final binary hypothesis testing model, construct the joint probability density function of the detection data and the training data, and perform maximum likelihood estimation on the unknown parameters in the joint probability density function. Based on the GLRT and Wald test criteria, construct the detection statistic Λ of the range extended target subspace detector under Gaussian clutter background. RD-P-SGLRT and Λ RD-P-SWald ;
[0011] Step 4. Set the detection threshold T according to the preset false alarm probability G, respectively, the detection statistic Λ RD-P-SGLRT ,Λ RD-P-SWald With the detection threshold T G Comparison is performed. If the detection statistic is greater than or equal to the detection threshold, it is determined that there are range-extended targets in the current K range units to be detected; otherwise, if the detection statistic is less than the detection threshold, it is determined that there are no range-extended targets in the current K range units to be detected.
[0012] Furthermore, in step 1, a preliminary binary hypothesis testing model is established, which specifically includes:
[0013]
[0014] Among them, H 0 represents the assumption that there is no target signal, H 1 Indicates the assumption that there is a target signal; X = [x 1 ,x 2 ,…,x k ,…,x K ] represents the data to be detected, X L =[x 1 ,x 2 ,…,x l ,…,x L ] represents L training data; AP represents the target signal, A represents the known N×p-dimensional full-rank complex matrix, P = [p 1 ,p 2 ,…,p k ,…,p K ] represents the target signal coordinate matrix, p k represents the unknown p×1-dimensional signal coordinate vector, p represents the dimension of the target signal subspace, and N represents the number of array elements; C L =[c 1 ,c 2 ,…,c l ,…,c L ] represents the matrix composed of the clutter components of the training data, C = [c 1 ,c 2 ,…,c k ,…,c K ] represents the matrix composed of the clutter components of the detection data, c k represents the clutter component of the detected data, c l represents the clutter component of the training data, k = 1, 2, ..., K, l = 1, 2, ..., L, K represents the number of distance units for collecting detection data, L represents the number of distance units for collecting training data; the clutter component of the detection data c k The clutter covariance matrix M is unknown and has a skew-symmetric structure.
[0015] Furthermore, the final binary hypothesis testing model in step 2 is as follows:
[0016]
[0017] in, D=TA,Y=TX,Y L =TX L , B = TC, Β L =TC L ;
[0018] In the above formula, Z represents the detection data after transformation matrix and dimensionality reduction transformation, W represents the clutter matrix of the detection data after transformation matrix and dimensionality reduction transformation, and Z L represents the training data after transformation matrix and dimensionality reduction transformation, W L represents the clutter matrix of the detected data after transformation matrix and dimensionality reduction transformation; H represents the signal subspace after transformation matrix and dimensionality reduction transformation; D represents the signal subspace after transformation matrix transformation; represents the conjugate transposed matrix of the signal subspace after transformation by the transformation matrix; Y represents the detection data after transformation by the transformation matrix; Y L represents the training data after transformation by the transformation matrix; B represents the clutter matrix of the detection data after transformation by the transformation matrix, L represents the clutter matrix of the training data after transformation by the transformation matrix; T represents the transformation matrix.
[0019] Furthermore, the step 3 specifically includes the following steps:
[0020] Step 3-1: Based on the final binary hypothesis testing model, construct the joint probability density function of the test data and the training data;
[0021] Step 3-2: performing maximum likelihood estimation on the clutter covariance matrix and target coordinates in the joint probability density function;
[0022] Step 3-3: Based on the maximum likelihood estimate of the target coordinates and the maximum likelihood estimate of the clutter covariance matrix Based on GLRT and Wald test criteria, the detection statistic Λ of the range extended target subspace detector under Gaussian clutter background is constructed. RD-P-SGLRT and Λ RD-P-SWald .
[0023] Furthermore, in step 3-1, based on the binary hypothesis testing model, a joint probability density function f of the test data and the training data is constructed. i (Z,Z L ; P, Ri ), specifically:
[0024]
[0025] in, R L represents the clutter covariance matrix of the training data, R i Indicates H 0 and H 1 The clutter covariance matrix under the assumption that i = 0, 1 corresponds to the assumption that there is no target signal H 0 and the hypothesis H with target signal 1 ; tr represents the trace of the matrix; |·| represents the determinant of the matrix, represents the conjugate transpose of the training data after the transformation matrix and dimensionality reduction transformation, Represents the conjugate transpose of a matrix.
[0026] Furthermore, in step 3-2, the maximum likelihood estimation is performed on the clutter covariance matrix and target coordinates in the joint probability density function:
[0027] The joint probability density function is partially derived from the clutter covariance matrix and set to zero to obtain the maximum likelihood estimate of the clutter covariance matrix. The maximum likelihood estimate of the clutter covariance matrix is substituted into the joint probability density function to obtain the joint probability density function of the given signal coordinates. Repeating the above calculation process can obtain the estimated value of the target coordinates.
[0028] Maximum likelihood estimate of the clutter covariance matrix and the maximum likelihood estimate of the target coordinates as follows:
[0029]
[0030]
[0031] in, Indicates H 0 and H 1 The maximum likelihood estimate of the clutter covariance matrix under the assumption; Represents the conjugate transpose of the signal subspace after the transformation matrix and dimensionality reduction transformation.
[0032] Furthermore, in step 3-3, the detection statistic Λ of the range extended target subspace detector under Gaussian clutter background is constructed RD-P-SGLRT and Λ RD-P-SWald :
[0033]
[0034] Λ RD-P-SWald =tr(Λ2 )
[0035] Among them, Λ RD-P-SGLRT represents the detection statistic of the RD-P-SGLRT detector, Λ RD-P-SWald represents the detection statistic of the RD-P-SWald detector; I represents the conjugate transpose of the detection data after the transformation matrix and dimensionality reduction transformation; K Represents the identity matrix, where all elements on the main diagonal are 1 and all other elements are 0.
[0036] Compared with the prior art, the present invention has the following technical effects:
[0037] (1) The present invention assumes that the covariance matrix has a skew-symmetric property and adopts a dimensionality reduction method using subspace transformation, which greatly reduces the amount of training data samples required for the adaptive process.
[0038] (2) Under the GLRT and Wald criteria, a subspace-based range-extended target signal detector is established in the Gaussian clutter background while ensuring the constant false alarm rate characteristic. Compared with the point target detector, the detection performance is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0040] Figure 1 It is a flow chart for realizing the present invention;
[0041] Figure 2 A block diagram of the construction steps of the RD-P-SGLRT detector and the RD-P-SWald detector of the present invention;
[0042] Figure 3 Schematic diagram of the CFAR characteristics of the detector of the present invention; (a) RD-P-SGLRT detector; (b) RD-P-SWald detector;
[0043] Figure 4 Schematic diagram of the relationship between the detection performance of the detector of the present invention and the dimension p of the target signal matrix when N=12, K=15, L=7; (a) RD-P-SGLRT detector; (b) RD-P-SWald detector;
[0044] Figure 5Schematic diagram of the relationship between the detection performance of the detector of the present invention and the target distance unit K when N=12, p=3, L=7; (a) RD-P-SGLRT detector; (b) RD-P-SWald detector;
[0045] Figure 6 Schematic diagram of detector detection performance corresponding to different training data sample sizes when N=12, K=15, p=3; (a) RD-P-SGLRT detector; (b) RD-P-SWald detector;
[0046] Figure 7 Schematic diagram of the comparison of the detection performance of the detector of the present invention with other detectors when N=12, K=15, p=3 under the condition of relatively scarce training data samples; (a) L=4; (b) L=11;
[0047] Figure 8 Schematic diagram of the comparison of detection performance between the detector of the present invention and other detectors when N=12, K=15, p=3 under the condition of sufficient training data sample size; (a) L=12; (b) L=24. DETAILED DESCRIPTION
[0048] In order to further explain the technical means and effects taken by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the technical solutions proposed by the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments. The specific features, structures or characteristics in one or more embodiments may be combined in any suitable form. Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by technicians in the technical field of the present invention.
[0049] Reference Figure 1-Figure 2 The present invention provides a method for detecting a range-extended target in a Gaussian clutter background based on GLRT and Wald test criteria, comprising the following steps:
[0050] Step 1. Consider a uniform receiving array with N array elements, obtain the data to be detected from K to-be-detected distance units, and obtain L observation data without target signals from the target-free distance units adjacent to the detected units as training data; perform subspace modeling on the target signal, model the Gaussian clutter as a complex Gaussian distribution with a zero-mean clutter covariance matrix that is unknown, and the clutter covariance matrix satisfies a skew-symmetric structure, and establish a preliminary binary hypothesis testing model;
[0051] Step 2. Use the transformation matrix to take advantage of the skew symmetry of the clutter covariance matrix, and then use the dimensionality reduction method based on subspace transformation to establish the final binary hypothesis testing model;
[0052] Step 3. Based on the final binary hypothesis testing model, construct the joint probability density function of the detection data and the training data, and perform maximum likelihood estimation on the unknown parameters in the joint probability density function. Based on the GLRT and Wald test criteria, construct the detection statistic Λ of the range extended target subspace detector under Gaussian clutter background. RD-P-SGLRT and Λ RD-P-SWald ;
[0053] Step 4. Set the detection threshold T according to the preset false alarm probability G , respectively, the detection statistic Λ RD-P-SGLRT ,Λ RD-P-SWald With the detection threshold T G Comparison is performed. If the detection statistic is greater than or equal to the detection threshold, it is determined that there are range-extended targets in the current K range units to be detected; otherwise, if the detection statistic is less than the detection threshold, it is determined that there are no range-extended targets in the current K range units to be detected.
[0054] The following is a detailed explanation of each of the above steps:
[0055] Step 1. Obtain the data to be detected from K distance units to be detected, and obtain L observation data without target signals from the target-free distance units adjacent to the detected units as training data; perform subspace modeling on the target signal, and model the Gaussian clutter as a complex Gaussian distribution with an unknown zero-mean clutter covariance matrix, and the clutter covariance matrix satisfies the skew-symmetric structure; based on the above assumptions, establish a preliminary binary hypothesis testing model.
[0056] In some embodiments, this step may include the following sub-steps:
[0057] Step 1-1: Consider a uniform receiving array with N array elements, obtain the data to be detected from K to-be-detected distance units, obtain L observation data from the target-free distance units adjacent to the detected unit as training data, and perform subspace modeling on the target signal to solve the mismatch problem between the actual steering vector and the assumed steering vector in practical applications; Assume that the receiver uses a uniform linear array with N array elements, the target range is extended to K distance units, and the target return of the kth distance unit is represented by an N-dimensional complex vector x k Indicates that k = 1, 2, ..., K, then the data to be detected can be expressed as Select L observation data of adjacent distance units without target signals as training data, and the training data of the lth distance unit is represented by the N-dimensional complex vector x l Indicates that l = 1, 2, ..., L, then the training data can be expressed as The observation vectors of different distance cells are statistically independent of each other. represents the target signal coordinate matrix, and the target signal of the kth distance unit is represented by Apk , the signal subspace A is a known N×p-dimensional full-rank complex matrix, p k is the unknown p×1-dimensional signal coordinate vector, p is the dimension of the target signal subspace; Represents the matrix composed of the clutter components of the detected data, the clutter component c k Independent and identically distributed between different distance units, and obeying a complex Gaussian distribution with a mean of zero and a covariance matrix of M, using a symmetrically spaced linear array for spatial domain processing or a symmetrically spaced pulse train for time domain processing, the clutter covariance matrix M has a skew-symmetric structure, that is, it satisfies M = JM * J, where (·) * represents the conjugate matrix of the corresponding matrix, and J represents the permutation matrix with the secondary diagonal elements being 1 and the other elements being 0. represents the matrix composed of the clutter components of the training data, c l Represents the clutter component of the training data.
[0058] Step 1-2: Based on the above assumptions, the range extended target detection problem under Gaussian clutter background can be expressed as a binary hypothesis test to determine whether the target exists in the data to be detected:
[0059]
[0060] Among them, H 0 represents the assumption that there is no target signal, H 1 Indicates the hypothesis that there is a target signal.
[0061] From the above description, it can be seen that the target coordinate matrix P and the clutter covariance matrix M are unknown.
[0062] Step 2. The transformation matrix is used to exploit the skew symmetry of the clutter covariance matrix, and then a dimensionality reduction method based on subspace transformation is adopted to establish the final binary hypothesis testing model.
[0063] In some embodiments, this step may include the following sub-steps:
[0064] Step 2-1: Use the transformation matrix T to take advantage of the skew symmetry of the clutter covariance matrix M to establish a new binary hypothesis testing model;
[0065] A transformation matrix T is used to transform the clutter covariance matrix M and the signal subspace A, thereby utilizing their skew-symmetric properties to convert the complex matrices M and A into real matrices.
[0066] The transformation matrix T is as follows:
[0067]
[0068] Among them, I N / 2represents the N / 2-dimensional identity matrix, I (N-1) / 2 represents the (N-1) / 2-dimensional identity matrix; J N / 2 represents the N / 2-dimensional permutation matrix, J (N-1) / 2 represents a (N-1) / 2-dimensional permutation matrix, whose off-diagonal elements are 1 and the other elements are 0; i represents the imaginary unit.
[0069] Therefore, formula (1) can be rewritten as a new binary hypothesis testing model as follows:
[0070]
[0071] Among them, Y = TX represents the detection data after transformation by the transformation matrix, B = TC represents the detection data clutter matrix after transformation by the transformation matrix, D = TA represents the signal subspace after transformation by the transformation matrix, Y L =TX L represents the training data after transformation by the transformation matrix, Β L =TC L Represents the clutter matrix of the training data after transformation by the transformation matrix.
[0072] The covariance matrix Σ of the clutter is as follows:
[0073]
[0074] in, Represents the conjugate transposed matrix of the corresponding matrix.
[0075] Step 2-2: Use the dimensionality reduction method based on subspace transformation to establish the final binary hypothesis testing model;
[0076] After the transformation, the signal subspace D and the covariance matrix Σ of the clutter after the transformation matrix become real matrices.
[0077] Next, a dimensionality reduction method based on subspace transformation is used for the above formula (3) to reduce the detection data Y and the training data Y L The conjugate transposed matrix of the signal subspace after premultiplication transformation By projecting the signal subspace onto a low-dimensional subspace, the final binary hypothesis testing model can be obtained as follows:
[0078]
[0079] in, Represents the detection data after transformation matrix and dimensionality reduction transformation, represents the signal subspace after transformation matrix and dimensionality reduction transformation, represents the clutter matrix of the detected data after transformation matrix and dimensionality reduction transformation, represents the training data after transformation matrix and dimensionality reduction transformation, Represents the clutter matrix of the detection data after transformation matrix and dimensionality reduction transformation.
[0080] Because the linear operation maintains the Gaussian nature of the clutter, the clutter is still Gaussian distributed, and C and C L Each column of shares the same covariance matrix, and the clutter covariance matrix R is as follows:
[0081]
[0082] Step 3. Based on the final binary hypothesis testing model, construct the joint probability density function of the detection data and the training data, and perform maximum likelihood estimation on the unknown parameters in the joint probability density function. Based on the GLRT and Wald test criteria, construct the detection statistic Λ of the range extended target subspace detector under Gaussian clutter background. RD-P-SGLRT and Λ RD-P-SWald .
[0083] In some embodiments, this step may include the following sub-steps:
[0084] Step 3-1: Based on the final binary hypothesis testing model, construct the joint probability density function of the test data and the training data;
[0085] Assuming that in K distance units, the signals and clutter between distance units are independent of each other, then in H 0 and H 1 The joint probability density function (PDF) of the test data and training data under the two assumptions is expressed as:
[0086]
[0087] in, R L represents the clutter covariance matrix of the training data, represents the conjugate transpose of the training data after the transformation matrix and dimensionality reduction transformation; R i , i=0,1 means H 0 and H 1 The clutter covariance matrix under the assumption that i = 0, 1 corresponds to the assumption that H 0 and H 1 , |R| represents the determinant of matrix R, R -1 represents the inverse of the matrix R and tr represents the trace of the matrix.
[0088] Step 3-2: performing maximum likelihood estimation on the clutter covariance matrix and target coordinates in the joint probability density function;
[0089] Take the partial derivative of the joint probability density function with respect to the clutter covariance matrix R and set it to zero, thus obtaining H 1 Assume that and H 0 The maximum likelihood estimate of the clutter covariance matrix R under the assumption as follows:
[0090]
[0091] According to the matrix inversion lemma, H 1 Assume that Z 1 = Z-HP, then the maximum likelihood estimate of the clutter covariance matrix is The inverse matrix can be expressed as:
[0092]
[0093] The maximum likelihood estimate of the clutter covariance matrix Substitute H 0 Assume that and H 1 The joint probability density function under the assumption that H 0 The joint probability density function under the assumption and H 1 The joint probability density function of the given signal coordinates P is assumed to be We can get:
[0094]
[0095]
[0096] By performing partial derivative operation on the target signal coordinate P of equation (11) and setting it equal to zero, the maximum likelihood estimate of the target signal coordinate P can be obtained as:
[0097]
[0098] In formula (12), Represents the maximum likelihood estimate of the target signal coordinate P.
[0099] Step 3-3: Based on the maximum likelihood estimate of the target coordinates and the maximum likelihood estimate of the clutter covariance matrix Based on GLRT and Wald test criteria, the detection statistic Λ of the range extended target subspace detector under Gaussian clutter background is constructed. RD-P-SGLRT and Λ RD-P-SWald .
[0100] Assume T G is the detection threshold under a given false alarm probability, so H shown in formula (7) 0 Assumptions and H 1The joint probability density function f under the assumption 0 (Z,Z L ; R) and f 1 (Z,Z L ; P, R), we can get the GLRT detection statistic Λ of the range extended target under Gaussian clutter background RD-P-SGLRT and Λ RD-P-SWald It is expressed as:
[0101]
[0102]
[0103] Where Θ represents the unknown parameter, I(Θ) is the Fisher information matrix of the unknown parameter Θ; r =vec(P) is a relative parameter; Θ s =vec(R) is the interference parameter; Θ 1 Indicates that Θ is in H 1 The maximum likelihood estimate (MLE) under the assumption that is Θ r In H 1 The MLE under the assumption that Represents Θ r In H 0 Under the assumption of MLE, it is clear that is a zero matrix, so equation (14) can be simplified as:
[0104]
[0105] Introducing the logarithmic operation of the joint probability density function, the specific form of the Fisher information matrix I(Θ) is:
[0106]
[0107] in, represents the block matrix form of the Fisher information matrix, They are the Fisher information matrix (Θ r ,Θ r )、(Θ r ,Θ s )、(Θ s ,Θ r )、(Θ s ,Θ s ) position corresponding to the submatrix.
[0108] According to the block matrix inversion lemma, we can get the Fisher information matrix (Θ r ,Θ r ) The inverse matrix corresponding to the position can be expressed as:
[0109]
[0110] Take the logarithm of the joint probability density function of the test data and the training data to obtain the logarithmic joint probability density function, and then calculate P * And P finds the partial derivative, so we have:
[0111]
[0112]
[0113] in,(·) T represents the transposed matrix of the corresponding matrix, P * Represents the conjugate matrix of the target signal coordinates P.
[0114] Substituting equation (18) and equation (19) into equation (16), we can obtain:
[0115]
[0116] Among them, I K Represents the K-dimensional identity matrix.
[0117] because is a zero matrix, so substituting equation (20) into equation (17) yields:
[0118]
[0119] The maximum likelihood estimate of the clutter covariance matrix Substituting the joint probability density function into the expression of the detection statistic in equation (13), the expression of the GLRT detector for a given signal coordinate P can be obtained as follows:
[0120]
[0121] From equations (15) and (21), we can get the Wald detector expression for a given target signal coordinate P and clutter covariance matrix R, as follows:
[0122]
[0123] Substituting the maximum likelihood estimate of the target signal coordinates P into the expression of the detection statistic in equation (22), the expression of the RD-P-SGLRT detector is obtained as follows; substituting the maximum likelihood estimate of the target signal coordinates P and the clutter covariance matrix R into the expression of the detection statistic in equation (23), the expression of the RD-P-SWald detector is obtained as follows:
[0124]
[0125] ΛRD-P-SWald =tr(Λ 2 ) (25)
[0126] in, Λ 1 ,Λ 2 represents the detection statistics of the two detectors; represents the conjugate transpose of the signal subspace after the transformation matrix and dimensionality reduction transformation; Represents the conjugate transpose of the detection data after the transformation matrix and dimensionality reduction transformation.
[0127] Step 4. Set the detection threshold T according to the preset false alarm probability G , respectively, the detection statistic Λ RD-P-SGLRT ,Λ RD-P-SGLRT With the detection threshold T G Comparison is performed. If the detection statistic is greater than or equal to the detection threshold, it is determined that there are range-extended targets in the current K range units to be detected; otherwise, if the detection statistic is less than the detection threshold, it is determined that there are no range-extended targets in the current K range units to be detected.
[0128] The CFAR characteristics of the RD-P-SGLRT detector and the RD-P-SWald detector for the clutter covariance matrix R are proved as follows:
[0129] It is known that the RD-P-SGLRT detector and the RD-P-SWald detector are shown in equations (24) and (25). By analyzing the expression of the detection statistics, it can be seen that if we want to prove that the RD-P-SGLRT detector and the RD-P-SWald detector are 0 Assume that the clutter covariance matrix R has the CFAR property, and need to prove and Under the assumption that H 0 The following is independent of the clutter covariance matrix R.
[0130] (1) Verify Λ 1 In H 0 Assume that R is irrelevant
[0131] make
[0132]
[0133] Then we have:
[0134]
[0135] After whitening treatment, 0 Assume that Subject to L degrees of freedom and covariance matrix I p The real Wishart distribution of Therefore, from formula (27), we can know that Statistically independent of R.
[0136] (2) Verify Λ 2 In H 0 Assume that R is irrelevant
[0137] First, 2 Perform a whitening transformation and then define a unitary matrix satisfy Then 2 It can be written as:
[0138]
[0139] To simplify the expression,
[0140] From formula (28), we can see that under the assumption that H 0 The detection data after whitening transformation and unitary transformation is And the detection data after whitening transformation The covariance matrix after whitening and unitary transformation And the covariance matrix after whitening transformation are statistically equivalent, so In H 0 Assume that it is independent of R.
[0141] Comprehensive analysis, and In H 0 Assume that it is independent of R, that is, Λ RD-P-SGLRT and Λ RD-P-SWald The statistical characteristics of H 0 Assume that it is independent of R, thus proving that the RD-P-SGLRT detector and RD-P-SWald detector are 0 Assume that R has the CFAR property.
[0142] In the present invention, we consider the problem of detecting range-extended targets in a Gaussian clutter background, based on the situation that only a small amount of training data is available in the actual adaptive target detection scenario. To this end, in the design stage, the clutter covariance matrix is first modeled as a skew-symmetric structure, and then a dimensionality reduction method based on subspace transformation is adopted. According to the GLRT and Wald test criteria, a subspace detector for range-extended targets in a Gaussian clutter background is derived. While ensuring the constant false alarm rate characteristics, the amount of training data samples required for the adaptive process is greatly reduced, and the detection performance of the detector is improved, especially when the training data is limited.
[0143] The effect of the present invention can be illustrated by the following simulation experiment:
[0144] Experimental results analysis:
[0145] Figure 3 is a schematic diagram of the CFAR characteristics of the detector; Figure 3 (a) and Figure 3 As can be seen in (b), with the fixed clutter covariance structure parameter ρ, only the clutter power level σ 2 The detection threshold and false alarm probability curves of the detector remain basically unchanged, indicating that the false alarm probability of the RD-P-SGLRT detector and the RD-P-SWald detector is not related to the clutter power. 2 , when only the clutter covariance structure parameter ρ is changed, the detection threshold and false alarm probability curves of the two detectors remain basically unchanged, indicating that the change of the clutter covariance structure parameters does not affect the detection performance of the RD-P-SGLRT detector and the RD-P-SWald detector. The above results show that the RD-P-SGLRT detector and the RD-P-SWald detector have CFAR characteristics for the clutter covariance matrix R, which is consistent with the above theoretical analysis conclusions.
[0146] Figure 4 is a schematic diagram showing the relationship between the detection performance of the detector and the dimension of the target signal matrix; Figure 4 (a) and Figure 4 As can be seen in (b), as the dimension p of the target signal subspace increases, the detection probability of the two detectors at the same signal to clutter ratio (SCR) decreases. This is because the dimension of the clutter covariance matrix R in equation (6) changes with the increase of p, but the number of training data remains unchanged; in other words, when using the same training data, an increase in the value of p will lead to a worse estimate of R, resulting in a decrease in the detection probability of the detector.
[0147] Figure 5 is a schematic diagram showing the relationship between the detection performance of the detector and the target distance unit K; Figure 5 (a) and Figure 5 As can be seen in (b), as the K value increases, that is, the number of target distance units increases, the detection performance of the RD-P-SGLRT detector and the RD-P-SWald detector both improve; but when the number of target distance units increases to a certain value, the performance changes of the two detectors become smaller and smaller. It is worth noting that in practical applications, clutter is limited by its uniformity, and there are not many available K values.
[0148] Figure 6 Schematic diagram of detector detection performance corresponding to different training data sample sizes; Figure 6 (a) and Figure 6As can be seen in (b), at the same SCR, the detection performance of the RD-P-SGLRT detector and the RD-P-SWald detector are improved with the increase of training data, indicating that the increase of training data improves the estimation accuracy of the covariance matrix R, thereby increasing the detection probability and improving the detection performance. This means that the detection performance gain can be obtained by improving the radar resolution. It is worth noting that when L = 1, the RD-P-SGLRT detector and the RD-P-SWald detector can still be used, but there will be a certain performance loss.
[0149] Figure 7 and Figure 8 The detection performance of the detector was compared with that of the RD-P-SGLRT detector and the RD-P-SWald detector of the present invention under different training data sample sizes. Figure 7 As shown in (a), when L = 4 (N>L), that is, when the number of training data samples is small, 1S-PGLRT and 1S-PWald are ineffective. We set L>p, so the two proposed detectors and RD-P-GLRT and STBD can work, among which RD-P-SGLRT has the best detection performance. Figure 7 In (b), it is obvious that both detectors work, but the proposed detector performs better than the comparison detector; when N≤L(L=12,24), that is, the training data sample size is more sufficient, Figure 8 (a) and Figure 8 (b) shows that the detection performance of all detectors has improved, and the proposed detector is still the best. Among them, the detection probability curves of the RD-P-SGLRT detector and the RD-P-SWald detector are the closest. This is because the increase in the number of training data samples will make the estimated value of the clutter covariance matrix closer to the true value, thereby effectively improving the detection performance. Therefore, the comprehensive Figure 7 and Figure 8 It can be seen that under the condition of sufficient training data samples, both the detector proposed in this paper and the existing detector show excellent detection performance. Under the same detection probability (0.8), the performance of the two detectors proposed in this paper improves the signal-to-noise ratio by more than 3dB. When the training data samples are scarce, the detection performance of the existing detector is poor, while the detector proposed in this paper still maintains a high detection probability.
[0150] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.
Claims
1. A method for detecting extended-range targets in Gaussian clutter background. It is characterized in that The following steps are involved: Step 1. Consider a uniform receiving array with N array elements, obtain the data to be detected from K to-be-detected distance units, and obtain L observation data without target signals from the target-free distance units adjacent to the detected units as training data; perform subspace modeling on the target signal, model the Gaussian clutter as a complex Gaussian distribution with a zero-mean clutter covariance matrix that is unknown, and the clutter covariance matrix satisfies a skew-symmetric structure, and establish a preliminary binary hypothesis testing model; Step 2. The transformation matrix is used to exploit the skew symmetry of the clutter covariance matrix, and then a dimensionality reduction method based on subspace transformation is used to establish the final binary hypothesis testing model; Step 3. Based on the final binary hypothesis testing model, construct the joint probability density function of the detection data and the training data, and perform maximum likelihood estimation on the unknown parameters in the joint probability density function. Based on the GLRT and Wald test criteria, construct the detection statistic Λ of the range extended target subspace detector under Gaussian clutter background. RD-P-SGLRT and Λ RD-P-SWald ; Step 4. Set the detection threshold T according to the preset false alarm probability G , respectively, the detection statistic Λ RD-P-SGLRT ,Λ RD-P-SWald With the detection threshold T G Compare, if the detection statistic is greater than or equal to the detection threshold, it is determined that the current K distance units to be detected have a distance extension target; conversely, if the detection statistic is less than the detection threshold, it is determined that the current K distance units to be detected do not have a distance extension target; The step 3 specifically includes the following steps: Step 3-1: Based on the final binary hypothesis testing model, construct the joint probability density function of the test data and the training data; Step 3-2: performing maximum likelihood estimation on the clutter covariance matrix and target coordinates in the joint probability density function; Step 3-3: Based on the maximum likelihood estimate of the target coordinates and the maximum likelihood estimate of the clutter covariance matrix Based on GLRT and Wald test criteria, the detection statistic Λ of the range extended target subspace detector under Gaussian clutter background is constructed. RD-P-SGLRT and Λ RD -P-S Wald; In step 3-1, based on the binary hypothesis testing model, a joint probability density function f of the test data and the training data is constructed. i (Z,Z L ; P, R i ), specifically: in, R L represents the clutter covariance matrix of the training data, R i Indicates H 0 and H 1 The clutter covariance matrix under the assumption that i = 0, 1 corresponds to the assumption that there is no target signal H 0 and the hypothesis H with target signal 1 ; tr represents the trace of the matrix; |·| represents the determinant of the matrix, represents the conjugate transpose of the training data after the transformation matrix and dimensionality reduction transformation, represents the conjugate transpose of the matrix; L represents the number of distance units for collecting training data; K represents the number of distance units for collecting detection data; H represents the signal subspace after the transformation matrix and dimensionality reduction transformation; Z represents the detection data after the transformation matrix and dimensionality reduction transformation; Z L represents the training data after transformation matrix and dimensionality reduction transformation; P = [p 1 ,p 2 ,…,p k ,…,p K ] represents the target signal coordinate matrix, p k represents the unknown p×1-dimensional signal coordinate vector, and p represents the dimension of the target signal subspace; In step 3-2, the maximum likelihood estimation is performed on the clutter covariance matrix and target coordinates in the joint probability density function: The joint probability density function is partially derived from the clutter covariance matrix and set to zero to obtain the maximum likelihood estimate of the clutter covariance matrix. The maximum likelihood estimate of the clutter covariance matrix is substituted into the joint probability density function to obtain the joint probability density function of the given signal coordinates. Repeating the above calculation process can obtain the estimated value of the target coordinates. Maximum likelihood estimate of the clutter covariance matrix and the maximum likelihood estimate of the target coordinates as follows: in, Indicates H 0 and H 1 The maximum likelihood estimate of the clutter covariance matrix under the assumption; represents the conjugate transpose of the signal subspace after the transformation matrix and dimensionality reduction transformation; In step 3-3, the detection statistic Λ of the range extended target subspace detector under Gaussian clutter background is constructed RD-P-SGLRT and Λ RD-P-SWald : L RD-P-SWald =tr(Λ 2 ) Among them, Λ RD-P-SGLRT represents the detection statistic of the RD-P-SGLRT detector, Λ RD-P-SWald represents the detection statistic of the RD-P-SWald detector; I represents the conjugate transpose of the detection data after the transformation matrix and dimensionality reduction transformation; K Represents the identity matrix, where all elements on the main diagonal are 1 and all other elements are 0.
2. According to the method for detecting extended range targets in Gaussian clutter background according to claim 1, It is characterized in that In step 1, a preliminary binary hypothesis testing model is established, which specifically includes: Among them, H 0 represents the assumption that there is no target signal, H 1 Indicates the assumption that there is a target signal; X = [x 1 ,x 2 ,…,x k ,…,x K ] represents the data to be detected, X L =[x 1 ,x 2 ,…,x l ,…,x L ] represents training data; AP represents the target signal, A represents the known N×p-dimensional full-rank complex matrix, P = [p 1 ,p 2 ,…,p k ,…,p K ] represents the target signal coordinate matrix, p k represents the unknown p×1-dimensional signal coordinate vector, p represents the dimension of the target signal subspace, and N is the number of array elements; C L =[c 1 ,c 2 ,…,c l ,…,c L ] represents the matrix composed of the clutter components of the training data, C = [c 1 ,c 2 ,…,c k ,…,c K ] represents the matrix composed of the clutter components of the detection data, c k represents the clutter component of the detected data, c l represents the clutter component of the training data, k = 1, 2, ..., K, l = 1, 2, ..., L, L represents the number of distance units for collecting training data, K represents the number of distance units for collecting detection data; the clutter component of the detection data c k The clutter covariance matrix M is unknown and has a skew-symmetric structure.
3. The method for detecting extended-range targets in a Gaussian clutter background according to claim 2, It is characterized in that The final binary hypothesis test model in step 2 is as follows: Among them, D = TA, Y = TX, Y L = TX L , B = TC, Β L = TC L ; In the above formula, Z represents the detection data after transformation matrix and dimensionality reduction transformation, W represents the clutter matrix of the detection data after transformation matrix and dimensionality reduction transformation, and Z L represents the training data after transformation matrix and dimensionality reduction transformation, W L represents the clutter matrix of the detected data after transformation matrix and dimensionality reduction transformation; H represents the signal subspace after transformation matrix and dimensionality reduction transformation; D represents the signal subspace after transformation matrix transformation; represents the conjugate transposed matrix of the signal subspace after transformation by the transformation matrix; Y represents the detection data after transformation by the transformation matrix; Y L represents the training data after transformation by the transformation matrix; B represents the clutter matrix of the detection data after transformation by the transformation matrix, L represents the clutter matrix of the training data after transformation by the transformation matrix; T represents the transformation matrix.
