Target detection method and device based on inverse diagonal joint robust detector
By using inverse diagonal combined robust detector in multi-channel phased array radar, the target-guided vector model and binary detection model are constructed, and the detection formula is derived by using inverse diagonal matrix and unitary matrix to solve the reduction of training data and the loss of guide vectors. The problem of low detection sensitivity caused by matching achieves efficient target detection in two-dimensional array radar.
Patent Information
- Application Number
- CN202411210179.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-30
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-08-30
AI Technical Summary
In the case of reduced training data and mismatch of guide vectors, the target detection sensitivity of multi-channel phased array radar is low, especially in two-dimensional array radars. The existing generalized likelihood ratio detectors are not applicable, resulting in detection performance decline.
Using the method based on the inverse diagonal joint robust detector, the target-oriented vector model and binary detection model are constructed, and the detection formula is derived using the inverse diagonal matrix and unitary matrix, and the robust detector is constructed by the rotation coordinate method and the Monte Carlo method. , reduce the dependence on the number of training data and improve detection sensitivity.
In the case of training data reduction and guidance vector mismatch, the target detection sensitivity of two-dimensional array radar is improved, the sharp decline in detection performance is slowed down, and efficient target detection under limited data conditions is achieved.
Smart Images

Figure CN119044919A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of radar detection technology, and in particular to a target detection method and device based on an inverse diagonal joint robust detector. Background Art
[0002] Multi-channel phased array radar systems have been widely used on many platforms due to their wide detection range and strong anti-interference ability. Among them, adaptive detection technology can greatly enhance the target detection and early warning functions of multi-channel radar systems. Adaptive detection mainly uses training data to estimate the interference and noise covariance matrix in the detection scene. Adaptive detection is to detect target signals from noise when there is sufficient training data and accurate guidance vectors. However, in actual scenarios, due to factors such as distance, spatial changes in geographical environment, increased complexity of electromagnetic environment, special configuration of array, and improved radar resolution, the background conditions for target detection become very harsh and complex. In complex detection scenarios, the training samples are contaminated by unnecessary components, resulting in limited available training data, especially for two-dimensional array radars with large-scale arrays, it is often difficult to meet the situation of sufficient training samples. In addition, in actual situations, the target is rarely consistent with the nominal point direction, which leads to a mismatch between the nominal and actual target guidance vectors, and when the radar beam is turned to the direction close to the target, a target component is also generated. Therefore, studying robust detectors in the case of insufficient training samples and mismatched guidance vectors is crucial to improving the detection capability of two-dimensional array radars.
[0003] At present, a generalized likelihood ratio test (GLRT) is usually used for target detection of multi-channel phased array radar. However, the GLRT is only suitable for application under the conditions of sufficient radar training data and accurate steering vectors. The GLRT is not applicable when the training data is reduced or the steering vector is mismatched, which makes the target detection sensitivity low under the conditions of reduced training data and steering vector mismatch. Summary of the invention
[0004] The purpose of the embodiments of the present invention is to provide a method and device for target detection based on an inverse diagonal joint robust detector, so as to solve the problem of low sensitivity of target detection when the training data is reduced and the steering vector is mismatched.
[0005] To solve the above technical problems, the embodiments of the present invention provide the following technical solutions:
[0006] A first aspect of the present invention provides a target detection method based on an inverse diagonal joint robust detector, the target detection method comprising:
[0007] When receiving the radar echo, a target guidance vector model is constructed in full increment form according to the three-dimensional spatial position of the target and the target azimuth position estimated by the radar;
[0008] A binary detection model is constructed according to the target-oriented vector model, the noise component in the test data, the noise component in the training data, and the signal amplitude, and a covariance matrix of the target is determined according to the noise component in the test data and the noise component in the training data, and the binary detection model is used to determine the test data and the training data;
[0009] Determine a joint probability density function of a test data matrix and a training data matrix according to the inverse diagonal matrix, the covariance matrix of the target, the estimated covariance matrix, the signal amplitude matrix, the first unitary matrix and the second unitary matrix, wherein the test data matrix is determined by the test data and the training data matrix is determined by the training data;
[0010] According to the joint probability density function and the generalized likelihood ratio test criterion, a first preset detection formula and a second preset detection formula of the inverse diagonal parameter are determined, and the first preset detection formula and the second preset detection formula are both fractionally optimized to obtain the corresponding first final detection formula and the second final detection formula;
[0011] Solving the first final detection formula and the second final detection formula respectively according to the rotation coordinate method to construct a first robust detector and a second robust detector;
[0012] According to the Monte Carlo method, the first robust detector and the second robust detector, it is detected whether there is a target.
[0013] A second aspect of the present invention provides a target detection device based on an inverse diagonal joint robust detector, the target detection device comprising:
[0014] The first construction module is used to construct a target guidance vector model in a full increment form according to the three-dimensional spatial position of the target and the target azimuth position estimated by the radar when the radar echo is received;
[0015] A second construction module is used to construct a binary detection model according to the target-oriented vector model, the noise component in the test data, the noise component in the training data, and the signal amplitude, and determine the covariance matrix of the target according to the noise component in the test data and the noise component in the training data, and the binary detection model is used to determine the test data and the training data;
[0016] A determination module is used to determine a joint probability density function of a test data matrix and a training data matrix according to an inverse diagonal matrix, a covariance matrix of a target, an estimated covariance matrix, a signal amplitude matrix, a first unitary matrix and a second unitary matrix, wherein the test data matrix is determined by the test data and the training data matrix is determined by the training data;
[0017] A score optimization module, used for determining a first preset detection formula and a second preset detection formula of the inverse diagonal parameter according to a joint probability density function and a generalized likelihood ratio test criterion, and performing score optimization on both the first preset detection formula and the second preset detection formula to obtain a corresponding first final detection formula and a second final detection formula;
[0018] A third construction module is used to solve the first final detection formula and the second final detection formula respectively according to the rotation coordinate method to construct a first robust detector and a second robust detector;
[0019] The detection module is used to detect whether there is a target according to the Monte Carlo method, the first robust detector and the second robust detector.
[0020] Compared with the prior art, the target detection method and device based on the inverse diagonal joint robust detector provided by the present invention, when receiving the radar echo, constructs a target-oriented vector model in full increment form according to the three-dimensional spatial position of the target and the target azimuth position estimated by the radar; constructs a binary detection model according to the target-oriented vector model, the noise component in the test data, the noise component in the training data and the signal amplitude, and determines the covariance matrix of the target according to the noise component in the test data and the noise component in the training data, and the binary detection model is used to determine the test data and the training data; according to the inverse diagonal matrix, the covariance matrix of the target, the estimated covariance matrix, The signal amplitude matrix, the first unitary matrix and the second unitary matrix determine the joint probability density function of the test data matrix and the training data matrix; according to the joint probability density function and the generalized likelihood ratio test criterion, the first preset detection formula and the second preset detection formula of the inverse diagonal parameters are determined, and the first preset detection formula and the second preset detection formula are both fractionally optimized to obtain the corresponding first final detection formula and the second final detection formula; the first final detection formula and the second final detection formula are respectively solved according to the rotating coordinate method to construct the first robust detector and the second robust detector; according to the Monte Carlo method, the first robust detector and the second robust detector, detect whether there is a target. In this way, a binary detection model of joint target detection and azimuth estimation based on the target-guided vector is constructed by using the target-guided vector model constructed in full incremental form. When the target is detected, the first final detection formula and the second final detection formula are solved by using the rotation coordinate method, and the target cosine offset can be adaptively output, thereby slowing down the sharp decline in the detection performance of the two-dimensional array radar when the target-guided vector is mismatched; the first preset detection formula and the second preset detection formula are derived by using the inverse diagonal matrix, the first unitary matrix and the second unitary matrix, and the inverse diagonal structure is used so that half of the parameters are needed to characterize the covariance matrix of the unknown target, which greatly reduces the dependence on the number of training data, so that the sensitivity of target detection is higher when the training data is reduced and the guidance vector is mismatched. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] By reading the detailed description below with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:
[0022] Figure 1 The flowchart of the object detection method based on the inverse diagonal joint robust detector is schematically shown;
[0023] Figure 2 A three-dimensional image of cosine similarity between a target guidance vector and an actual guidance vector is schematically shown;
[0024] Figure 3 Schematically showing the probability curves of each detector detection when the training data is sufficient and insufficient;
[0025] Figure 4 The mean square error (MSE) curve of the detector of the present invention is schematically shown;
[0026] Figure 5 The structure of a target detection device based on an inverse diagonal joint robust detector is schematically shown. DETAILED DESCRIPTION
[0027] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.
[0028] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in the present invention should have the common meanings understood by those skilled in the art to which the present invention belongs.
[0029] The method in the embodiment of the present invention is described in detail below.
[0030] Figure 1 The flowchart of the target detection method based on the anti-diagonal joint robust detector in the embodiment of the present invention is schematically shown. Figure 1 As shown, the target detection method may include:
[0031] S101. When a radar echo is received, a target guidance vector model is constructed in a full increment form according to the three-dimensional spatial position of the target and the target azimuth position estimated by the radar.
[0032] The target bearing position estimated by the radar includes a first estimated target bearing position and a second estimated target bearing position.
[0033] Specifically, according to the three-dimensional spatial position of the target and the target azimuth position estimated by the radar, a target guidance vector model is constructed using the full increment form, including:
[0034] Step A1: When receiving the radar echo, the three-dimensional spatial position of the target is converted into the cosine space to obtain the true target direction position.
[0035] The real target direction position includes a first real target direction position and a second real target direction position.
[0036] Assume that a two-dimensional array radar with N×M uniformly rectangular array elements receives a radar echo reflected from a target at a distance of R, an azimuth of θ, and an elevation of φ′. First, the three-dimensional spatial position of the target is converted to cosine space, that is, cosine uv space is:
[0037] u=sin(θ)cos(φ′), v=sin(θ)sin(φ′);
[0038] Among them, u is the first real target direction position, that is, a component of the target direction in the radar array plane, which is related to the direction in the horizontal plane of the array. v is the second real target direction position, that is, another component of the target direction in the radar array plane, which is related to the vertical direction of the array.
[0039] Cosine space is the space in which cosine offsets are calculated.
[0040] Step A2: forming a preset steering vector of the target according to the first real target direction position, the second real target direction position, the number of row array elements and the number of column array elements in the uniform rectangular distribution array.
[0041] The expression of the preset steering vector of the target is:
[0042]
[0043] Among them, g(u,v) is the preset steering vector of the target, g u (u) is the weighted response of each antenna element to the signal from direction u, g v (v) is the weighted response of each antenna element to the signal coming from direction v, is a complex matrix space with dimension Z×1, is the Hadamard product, Z is the product of the number of row array elements and the number of column array elements, Z=NM, N is the number of row array elements, M is the number of column array elements, u is the first real target direction position, and v is the second real target direction position.
[0044] g u (u) and g v The expressions of (v) are:
[0045]
[0046] Where λ0 is the wavelength of the radar transmission wave, j is the imaginary unit, (·) T is the transposed symbol, u is the first real target direction position, v is the second real target direction position, assuming that the coordinate center is located at the center of the two-dimensional array, x0, x1,…, x N-1 and y0,y1,…,y M-1 are the coordinate positions of the row and column elements in the two-dimensional array, x N-1 is the coordinate position of the N-1th row element in the two-dimensional array, y M-1 is the coordinate position of the M-1th row element in the two-dimensional array, is a complex space with dimension N×1, is a complex space of dimension M×1.
[0047] Step A3: Determine a first direction cosine offset according to the first real target direction position and the first estimated target azimuth position, and determine a second direction cosine offset according to the second real target direction position and the second estimated target azimuth position.
[0048] In actual scenarios, the target arrival direction is usually not consistent with the array guidance vector, so the direction cosine offset (Δu, Δv) is needed to replace the target's true angular position in (u, v), that is:
[0049]
[0050] Where Δu is the first direction cosine offset, is the first estimated target azimuth position, Δv is the second direction cosine offset, is the second estimated target position, It is the target position estimated by radar.
[0051] Step A4: constructing a target steering vector model according to the first estimated target azimuth position, the second estimated target azimuth position, the preset steering vector of the target, the first direction cosine offset and the second direction cosine offset.
[0052] The preset steering vector of the target includes a first real target direction position and a second real target direction position.
[0053] The following steps A41 and A42 are specific operations for linearizing the array around the direction of the steering vector.
[0054] Specifically, a target steering vector model is constructed according to the first estimated target orientation position, the second estimated target orientation position, a preset steering vector of the target, a first direction cosine offset, and a second direction cosine offset, including:
[0055] Step A41: Replace the first real target direction position in the preset steering vector of the target with the first estimated target orientation position, and replace the second real target direction position with the second estimated target orientation position, so as to obtain the estimated steering vector of the target.
[0056] Replace u in the expression of the target's preset steering vector in step A2 with Replace v with The estimated steering vector of the target can be obtained
[0057] Step A42: Using the first direction cosine offset and the second direction cosine offset, Taylor expansion is performed on the first estimated target orientation position and the second estimated target orientation position in the estimated steering vector of the target to obtain a target steering vector model.
[0058] Specifically, using the first direction cosine offset and the second direction cosine offset, Taylor expansion is performed on the first estimated target orientation position and the second estimated target orientation position in the estimated steering vector of the target to obtain a target steering vector model, including:
[0059] Using the first direction cosine offset and the second direction cosine offset (Δu, Δv), the first estimated target orientation position and the second estimated target orientation position in the estimated steering vector of the target are Taylor expansion is performed into full incremental form, and the goal-oriented vector model is obtained using the following formula:
[0060]
[0061]
[0062] Among them, g a (Δu,Δv) is the target-oriented vector model, is the estimated steering vector of the target, Δu is the first direction cosine offset, Δv is the second direction cosine offset, is the first estimated target position, is the second estimated target position, is a complex matrix space of dimension Z×1, where Z is the product of the number of row array elements and the number of column array elements, Z=NM, where N is the number of row array elements and M is the number of column array elements. Calculate the partial derivative of the target-oriented vector model for the first true target direction position u, The partial derivative of the target-oriented vector model for the second real target direction position v is obtained, j is the imaginary unit, λ0 is the wavelength of the radar transmission wave, For each antenna element pair from The weighted response of the signal in the direction, x N-1 is the coordinate position of the N-1th row element in the two-dimensional array, For each antenna element pair from The weighted response of the signal in the direction, ⊙ is the Kronecker product, y M-1 is the coordinate position of the M-1th row element in the two-dimensional array, It is the Hadamard product.
[0063] After step A42, to simplify the notation, It is represented by g, and Respectively expressed as and Then the target-oriented vector model can be expressed as:
[0064]
[0065] in, g a (Δu, Δv) is the target-oriented vector model, Δu is the first direction cosine offset, Δv is the second direction cosine offset, is the estimated steering vector of the target, is the first estimated target position, is the second estimated target position, Calculate the partial derivative of the target-oriented vector model for the first true target direction position u, The target-oriented vector model is used to obtain the partial derivative of the second real target direction position v, G is the concatenated partial steering matrix, and Δθ is the target cosine deviation angle information.
[0066] S102, constructing a binary detection model according to the target-oriented vector model, the noise component in the test data, the noise component in the training data, and the signal amplitude, and determining the covariance matrix of the target according to the noise component in the test data and the noise component in the training data.
[0067] Among them, the binary detection model is used to determine the test data and training data.
[0068] The binary detection model is a detection model that combines target detection and orientation estimation.
[0069] Based on the target-oriented vector model in the two-dimensional array obtained in step S102, assuming that the useful radar target echo signal is submerged in Gaussian interference with unknown spectral characteristics, and assuming that there are K≥N uniform training samples (i.e., data vectors that do not contain useful target echo signals), the detection problem of detecting a target located at (u, v) in the cosine uv space can be transformed into a binary hypothesis problem. According to the target-oriented vector model, the noise component in the test data, the noise component in the training data, and the signal amplitude, a binary detection model is constructed:
[0070]
[0071] in, is the condition without target echo signal, is the condition with target echo signal, K is the number of training unit data, k is any one of the numbers of training unit data, l is the test data, i is the noise component in the test data, l k is the training data, i k is the noise component in the training data, a is the signal amplitude, which is unknown, and g a (Δu, Δv) is the target-oriented vector model, Δu is the first direction cosine offset, and Δv is the second direction cosine offset.
[0072] In addition, the noise component i in the training data k The test data l are modeled as independent complex circular Gaussian random vectors, and the covariance matrix of the target of noise and interference information in the target space is:
[0073]
[0074] Among them, S1 is the covariance matrix of the target, which is an unknown covariance matrix. For the matrix expectations, is the conjugate transpose of the test data i, is the noise component i in the training data k The conjugate transpose of .
[0075] S103, determining a joint probability density function of the test data matrix and the training data matrix according to the inverse diagonal matrix, the target covariance matrix, the estimated covariance matrix, the signal amplitude matrix, the first unitary matrix and the second unitary matrix.
[0076] Among them, the test data matrix is determined by the test data, and the training data matrix is determined by the training data.
[0077] Specifically, according to the inverse diagonal matrix, the covariance matrix of the target, the estimated covariance matrix, the signal amplitude matrix, the first unitary matrix and the second unitary matrix, a joint probability density function of the test data matrix and the training data matrix is determined, including:
[0078] Step B1: Determine the target true covariance matrix based on the conjugate matrix corresponding to the inverse diagonal matrix and the target covariance matrix.
[0079] Before step B1, it is necessary to construct an inverse diagonal matrix. The inverse diagonal structure indicates that the main diagonal and the secondary diagonal of the covariance matrix are symmetric, that is:
[0080]
[0081] Among them, J is the inverse diagonal matrix, N is the number of row array elements, and M is the number of column array elements.
[0082] When the two-dimensional uniform linear array is symmetric in the spatial domain and has the same interval, the target covariance matrix and the target-oriented vector model both have an inverse diagonal structure. The target true covariance matrix is determined based on the conjugate matrix corresponding to the inverse diagonal matrix and the target covariance matrix:
[0083] S=JS1 * J;
[0084] Among them, S is the target true covariance matrix, S1 * is the conjugate of the covariance matrix of the target, and J is the inverse diagonal matrix.
[0085] Correspondingly, g a (Δu, Δv) = Jg a (Δu,Δv) * , where g a (Δu, Δv) is the target-oriented vector model, g a (Δu,Δv) * is the conjugate of the target-oriented vector model.
[0086] Step B2: Determine a test data matrix based on the test data, and determine a training data matrix based on the training data.
[0087] According to the test data, the test data matrix is determined as According to the training data, the training data matrix is determined as Among them, R is the test data matrix, R s is the training data matrix, l is the test data, l k is the training data, K is the number of training data, is a complex space of dimension Z×K.
[0088] Step B3: Determine a preset joint probability density function of the test data matrix and the training data matrix according to the target true covariance matrix, the estimated covariance matrix, the adjacency matrix corresponding to the test data matrix, and the adjacency matrix corresponding to the signal amplitude matrix.
[0089] According to the target true covariance matrix, the estimated covariance matrix, the adjacency matrix corresponding to the test data matrix and the adjacency matrix corresponding to the signal amplitude matrix, the test data matrix R and the training data matrix R under the two-dimensional array are constructed. s The preset joint probability density function is:
[0090]
[0091] In order to simplify the notation, g a =g a (Δu, Δv), For R and R s The preset joint probability density function, R is the test data matrix, R s is the training data matrix, is the condition without target echo signal, is the condition with target echo signal, K is the number of training data, N is the number of row array elements, S is the target true covariance matrix, H i is the estimated information matrix for the test data, |·| is the determinant of the matrix, tr(·) is the rank of the matrix, exp[·] is the exponential with base e, (·) -1 is the inverse of the matrix.
[0092] Test data estimate information matrix H i It is expressed as:
[0093]
[0094] in, To estimate the covariance matrix, R p is the adjacency matrix for the real and imaginary parts of the test data matrix, a p is the adjacency matrix for the real and imaginary parts of the signal amplitude, g a is the target-oriented vector model, is the conjugate transpose of the adjacency matrix for the real and imaginary parts of the signal amplitude, and R p 、a p Can be expressed as:
[0095]
[0096] Among them, R sis the training data matrix, is the conjugate transpose of the training data matrix, J is the inverse diagonal matrix, is a matrix space of dimension Z×Z, R e is the real part of the test data matrix R, R o is the imaginary part of the test data matrix R, is a matrix space with dimension Z×2, a e is the real part of the signal amplitude a, a o is the imaginary part of the signal amplitude a, is a matrix space with dimension 2×1, It represents the conjugate transpose of the matrix. The conjugate transpose of a real matrix is its transposed matrix, and the conjugate transpose of a complex matrix is the conjugate transpose of all its elements.
[0097] R e , R o 、a e 、a o It can be expressed in detail as:
[0098]
[0099]
[0100] Among them, R is the test data matrix, R * is the conjugate of the test data matrix, a is the signal amplitude, is a matrix space with dimension Z×1.
[0101] Step B4: Use the first unitary matrix and the second unitary matrix to perform unitary transformation on the adjacency matrix corresponding to the signal amplitude matrix, the adjacency matrix corresponding to the test data matrix, the estimated covariance matrix and the target covariance matrix, respectively, to obtain the real amplitude matrix after unitary transformation, the real test unit data matrix after unitary transformation, the real estimated covariance matrix after unitary transformation and the real steering vector after unitary transformation.
[0102] Define two unitary matrices D P and V P , respectively:
[0103]
[0104]
[0105] Among them, D p is the first unitary matrix, representing a unitary matrix of dimension Z×Z, V p is the second unitary matrix, representing a unitary matrix of dimension 2, I N is an N-row, N-column identity matrix, and j is the imaginary unit.
[0106] Using the first unitary matrix D p and the second unitary matrix V p , respectively, for the adjacency matrix corresponding to the signal amplitude matrix and the adjacency matrix R corresponding to the test data matrix p , estimated covariance matrix and the covariance matrix g of the target a Performing unitary transformation, we can obtain:
[0107]
[0108] Among them, a r is the real amplitude matrix after unitary transformation, R r is the real test unit data matrix after unitary transformation, is the real estimated covariance matrix after unitary transformation, g r is the real steering vector after unitary transformation, is the unitary transformation symbol, is the conjugate transpose of the second unitary matrix, is a real matrix space with dimension Z×2, To estimate the covariance matrix, is the conjugate transpose of the first unitary matrix, is a real matrix space of dimension Z×Z, is a real matrix space with dimension Z×1, For the practical department, To find the imaginary part.
[0109] Step B5: Substitute the unitary transformed real amplitude matrix, the unitary transformed real test unit data matrix, the unitary transformed real estimated covariance matrix and the unitary transformed real steering vector into the preset joint probability density function to obtain the joint probability density function.
[0110] Specifically, the unitary transformed real amplitude matrix, the unitary transformed real test unit data matrix, the unitary transformed real estimated covariance matrix and the unitary transformed real steering vector are substituted into the preset joint probability density function to obtain the joint probability density function, including:
[0111] Substitute the unitary transformed real amplitude matrix, the unitary transformed real test unit data matrix, the unitary transformed real estimated covariance matrix and the unitary transformed real steering vector into the preset joint probability density function, and use the following formula to obtain the joint probability density function:
[0112]
[0113]
[0114]
[0115] in, is the joint probability density function, i=0 means after transformation The joint probability density function under the assumption that i = 1 means after transformation The joint probability density function under the assumption that R is the test data matrix, R s is the training data matrix, is the condition without target echo signal, is the condition with target echo signal, S r is a known real covariance matrix, is the known actual test data information matrix, Z is the product of the number of row array elements and the number of column array elements, Z = NM, N is the number of row array elements, M is the number of column array elements, K is the number of training unit data, D p is the first unitary matrix, S is the target true covariance matrix, is the matrix conjugate transpose, To find the real part, J is the inverse diagonal matrix, To find the imaginary part, is the real estimated covariance matrix after unitary transformation, R r is the real test unit data matrix after unitary transformation, g r is the real steering vector after unitary transformation, is the transpose of the real magnitude matrix after the unitary transformation.
[0116] S104. Determine a first preset detection formula and a second preset detection formula for the inverse diagonal parameter according to a joint probability density function and a generalized likelihood ratio test criterion, and perform score optimization on both the first preset detection formula and the second preset detection formula to obtain corresponding first final detection formula and second final detection formula.
[0117] Step S104 is to transform the inverse diagonal parameter detection under the one-step method and the two-step method into a fractional optimization problem based on GLRT. The one-step method refers to jointly solving all unknown parameters through the GLRT criterion, and the two-step method refers to first solving the unknown parameters in the known real covariance matrix S r The GLRT decision statistic is derived under the assumption that ; then, the real estimated covariance matrix after unitary transformation is Substitute into the decision statistic obtained in the first step, replacing the original known real covariance matrix.
[0118] Specifically, according to the joint probability density function, the maximum likelihood criterion detection decision formula based on the one-step method is obtained as follows:
[0119]
[0120] in, for The joint probability density function under the assumption, for The joint probability density function under the assumption, is the condition without target echo signal, is the condition with target echo signal, ar is the real amplitude matrix after unitary transformation, Sr is the known real covariance matrix, Δu is the first direction cosine offset, Δv is the second direction cosine offset, R is the test data matrix, R s is the training data matrix, ξ is the detection threshold set to ensure constant false alarm, α is the first constraint parameter, β is the second constraint parameter, α and β are constraint parameters to ensure that the target steering vector, i.e. the approximate steering vector, is consistent with the actual steering vector, and are generally set within the 3dB beam width of the radar.
[0121] Since the real amplitude matrix a after unitary transformation r , the cosine offset (Δu, Δv) and the known real covariance matrix S r is unknown, so we need to use GLRT to find the value with the largest likelihood ratio. At the same time, find S for the numerator and denominator of the above test formula. r The minimum value of and further simplification can be obtained:
[0122]
[0123] Among them, R r is the real test unit data matrix after unitary transformation, is the real estimated covariance matrix after unitary transformation, ξ is the set detection threshold used to ensure constant false alarm, g r is the real steering vector after unitary transformation, is the transpose of the real amplitude matrix after unitary transformation, and |·| is the determinant of the matrix.
[0124] Minimizing the denominator of the above formula yields:
[0125]
[0126] Among them, a r is the real amplitude matrix after unitary transformation, is the transpose of the real test unit data matrix after unitary transformation.
[0127] The desired α r Substituting into the test formula, we can get the following parameter test formula:
[0128]
[0129] in, φ is a variable, is the transpose of the real steering vector after the unitary transformation. To facilitate the solution of the optimization problem, some parameters of the above equation are whitened, that is, The final score detection optimization problem under the GLRT one-step method, that is, the first final detection formula is expressed as:
[0130]
[0131] in, is the actual steering vector of the target whitening, is the whitening partial steering matrix, Δθ is the target cosine deviation angle information, is the first whitened covariance matrix, is the transpose of the first whitened covariance matrix,
[0132]
[0133] D p is the first unitary matrix, J is the inverse diagonal matrix, g * is the conjugate of the estimated steering vector of the target, is the real estimated covariance matrix after unitary transformation, G * is the conjugate of the concatenated partial steering matrix, R r is the actual test unit data matrix after unitary transformation, φ is the variable, I is the unit matrix, is the transpose of the actual test unit data matrix after unitary transformation, α is the first constraint parameter, β is the second constraint parameter, Δu is the first direction cosine offset, and Δv is the second direction cosine offset.
[0134] In the two-step GLRT detector, it is assumed that the real covariance matrix S is known r is known, then the two-step GLRT test formula is:
[0135]
[0136] in, is the joint probability density function under the no-target assumption, is the joint probability density function under the target hypothesis, R is the test data matrix, S r is a known real covariance matrix, a r is the real amplitude matrix after unitary transformation, Δθ is the target cosine deviation angle information, ξ is the set detection threshold for ensuring constant false alarm, α is the first constraint parameter, β is the second constraint parameter, is the condition without target echo signal, is the condition with target echo signal, Δu is the first direction cosine offset, Δv is the second direction cosine offset;
[0137] Among them, the real covariance matrix S is known r for:
[0138]
[0139] Among them, R s is the training data matrix, is the transpose of the training data matrix, and J is the inverse diagonal matrix.
[0140] Similar to the one-step simplification, the two-step GLRT detection formula is simplified based on the maximum likelihood criterion as follows:
[0141]
[0142] in, is the transpose of the second whitened covariance matrix, is the projection matrix based on the steering vector, is the second whitened covariance matrix, that is, the whitened two-step covariance matrix, is the actual steering vector of the target whitening, is the actual guidance vector of the second whitening of the target, that is, the actual guidance vector of the target whitening two-step method, The actual guidance vector of the second whitening target The conjugate transpose of R r is the actual test unit data matrix after unitary transformation, S r is the known real covariance matrix, and ξ is the set detection threshold used to ensure constant false alarm.
[0143] Simplifying the above formula into a fractional optimization problem is:
[0144]
[0145] in, is the transpose of the real test unit data matrix after unitary transformation, g r is the real steering vector after unitary transformation, is the transpose of the real steering vector after the unitary transformation.
[0146] Similarly, the above formula can be whitened to obtain the second final detection formula:
[0147]
[0148] in, is the second whitened covariance matrix, is the transpose of the second whitened covariance matrix, S r is a known real covariance matrix, D p is the first unitary matrix, G * is the conjugate of the concatenated partial steering matrix, is the whitening partial steering matrix, is the target whitening actual steering vector, and Δθ is the target cosine deviation angle information.
[0149] S105 . Solve the first final detection formula and the second final detection formula respectively according to the rotation coordinate method to construct a first robust detector and a second robust detector.
[0150] Specifically, the first final detection formula and the second final detection formula are solved respectively according to the rotation coordinate method to construct a first robust detector and a second robust detector, including:
[0151] Step C1: Solve the first final detection formula according to the rotation coordinate method, that is, use the rotation coordinate method to solve the parameter optimization problem obtained by the one-step method. First, initialize the number of iterations n = 0; initialize the optimal azimuth angle θ (n) =[Δu (n) ,Δv (n) ] T =[0,0] T , where θ (n) Represents the azimuth after the nth iteration; initializes the deflection vector in Will and Δθ (n) Substitute the detection formula under the one-step method, that is, the first final detection formula, to obtain the nth detection value t (n) :
[0152]
[0153] in, Δu (n) is the value of the first cosine offset of the nth iteration, Δv (n) is the value of the second cosine offset of the nth iteration, is the actual steering vector of the target whitening, is the whitening partial steering matrix, Δθ (n) is the target cosine deviation angle information of the nth iteration, is the first whitened covariance matrix, is the transpose of the first whitened covariance matrix.
[0154] Step C2: After initialization, let n=n+1 represent the nth iteration, assuming that Δv is an invariant, and calculate the optimal value of the first direction cosine offset Δu as:
[0155]
[0156] in, Δu * is the optimal value of the first direction cosine offset Δu, and argmax is the parameter Δu where the maximum value of the right formula is returned* , is the bias Δv steering vector after the n-1th iteration, is the first whitened covariance matrix, is the actual target guidance vector containing only the cosine component u direction, D p is the first unitary matrix, Find the conjugate of the partial derivative of the target-oriented vector model with respect to the first true target direction position u.
[0157] Step C3: Calculate the deviation Δu steering vector after the nth iteration Assuming Δu is a constant, the optimal value of the second direction cosine offset Δv is calculated as:
[0158]
[0159] Where Δv * is the optimal value of the second direction cosine offset Δv, is the bias Δu steering vector after the nth iteration, is the actual guidance vector of the target containing only the cosine component v direction.
[0160] Step C4: Calculate the deviation Δv steering vector after the nth iteration Update azimuth angle θ (n) =[Δu * ,Δv * ] T , update the nth detection value under the one-step detection formula.
[0161] Step C5: Assume that the judgment threshold for the end of the cycle of the rotation coordinate method is If Then the first optimal azimuth angle, that is, the optimal azimuth angle under the one-step method, is obtained as like Then return to step C2 to calculate the iteration.
[0162] t (n-1) is the n-1th detection value.
[0163] Step C6: Optimize the optimal azimuth angle under the one-step method by rotating the coordinates and the second optimal azimuth, i.e. the optimal azimuth under the two-step method Then the expression of the first robust detector, that is, the robust detector for joint target detection and orientation estimation in one-step method, is:
[0164]
[0165] in, R r is the real test unit data matrix after unitary transformation, is the transpose of the real test unit data matrix after unitary transformation, is the real estimated covariance matrix after unitary transformation, is the first target actual guidance vector after optimization, i.e., the target actual guidance vector after one-step optimization, is the first target amplitude, that is, the target amplitude estimated by the one-step method, is the condition for the existence of target echo signal, is the condition that there is no target echo signal, ξ is the set detection threshold for ensuring constant false alarm, D p is the first unitary matrix, J is the inverse diagonal matrix, g is the estimated steering vector of the target, G is the partial steering matrix after splicing, is the first optimal azimuth, S r is a known real covariance matrix, It is the conjugate transpose of the optimized first target actual steering vector, that is, the conjugate transpose of the target actual steering vector after one-step optimization.
[0166] Similarly, the expression of the second robust detector, that is, the robust detector for joint target detection and orientation estimation in the two-step method, is:
[0167]
[0168] in, is the transpose of the second whitened covariance matrix, is the optimized second partial derivative projection matrix, that is, the partial derivative projection matrix optimized by the two-step method, is the second whitening covariance matrix, Δu is the first direction cosine offset, Δv is the second direction cosine offset, is the actual guidance vector of the second target after optimization, i.e., the actual guidance vector of the target after two-step optimization. is the transpose of the actual guidance vector of the optimized second target, is the second optimal azimuth, i.e., the azimuth after two-step optimization.
[0169] The optimization steps of the two-step method are the same as the one-step method, the only difference is the optimization detection formula.
[0170] S106. Detect whether there is a target according to the Monte Carlo method, the first robust detector and the second robust detector.
[0171] Specifically, according to the Monte Carlo method, the first robust detector and the second robust detector, detecting whether a target exists includes:
[0172] Step C1: In the absence of a target echo signal, generate noise data using the Monte Carlo method.
[0173] exist That is, when there is no target, assuming the false alarm probability of the experiment is P fa , then the total number of Monte Carlo experiments is 100 / P fa Using simulation software, a set of noise data is randomly generated in each Monte Carlo method experiment.
[0174] Step C2: Input the noise data into the first robust detector and the second robust detector to output a plurality of detection thresholds.
[0175] 100 / P fa After Monte Carlo experiments, we get 100 / P fa A detection threshold.
[0176] Step C3: Select a preset detection threshold from multiple detection thresholds.
[0177] Specifically, it will be 100 / P fa The detection thresholds are sorted from small to large, and the 100th detection threshold is selected as the detection threshold of the current detector.
[0178] Step C4: In the presence of a target echo signal, a Monte Carlo method is used to generate noise data with a target guidance vector.
[0179] exist That is, the detection statistic of the detector calculated under the condition of target, and 100 / P fa Monte Carlo experiments are performed, and in each Monte Carlo experiment, a set of noise data with target-oriented vectors is randomly generated.
[0180] Step C5: Input the noise data with the target-oriented vector into the first robust detector and the second robust detector to output detection values.
[0181] Step C6: Determine whether the detection value is greater than a preset detection threshold. If so, it is detected that a target exists; if not, it is detected that a target does not exist.
[0182] The detection value is compared with the corresponding preset detection threshold to determine whether the detection value is greater than the preset detection threshold. If so, the target is detected to exist; if not, the target is detected to not exist.
[0183] Figure 2 The three-dimensional image of the cosine similarity between the target guidance vector and the actual guidance vector is schematically shown, wherein Δu is the cosine offset in the first direction, Δv is the cosine offset in the second direction, Δu and Δv are the cosine offsets of the target guidance vector, and the z axis in the three-dimensional coordinate axis is the cosine similarity between the target guidance vector and the actual guidance vector. Figure 2As shown, when Δu and Δv are less than a certain threshold (i.e., the 3dB width of the radar transmitting beam), the target steering vector has a high similarity with the actual steering vector.
[0184] Figure 3 The probability curves of each detector detection when the training data is sufficient and insufficient are schematically shown. Figure 3 (a) with Figure 3 (b) is a comparison curve of radar detection performance under different mismatch conditions when the amount of training data is sufficient in the two-dimensional array radar. The amount of training data is twice the dimension of the two-dimensional array system, that is, K = 2NM = 32; Figure 3 (c) with Figure 3 (d) is a comparison curve of radar detection performance under different mismatch conditions when the amount of training data is insufficient in the two-dimensional array radar. The number of training data K is the system dimension K = NM = 16. The array elements of the two-dimensional array are set to N = M = 4, N is the number of row elements, M is the number of column elements, and the constraint parameter α = β = 0.5; in addition Figure 3 (a) with Figure 3 In (c), Δu=Δv=0 represents no mismatch, Δu is the first direction cosine offset, Δv is the second direction cosine offset, Figure 3 (b) with Figure 3 In (d), Δu = Δv = 0.3 represents the mismatch of the steering vector. d represents the detection probability value, the horizontal axis is the signal-to-interference-noise ratio, and the signal-to-interference-noise ratio is modeled as Where a is the signal amplitude, g a is the target-oriented vector model, and S is the target true covariance matrix.
[0185] Figure 3 (a) with Figure 3In (b), in the absence of mismatch, the detection probabilities of the first robust detector proposed in the present invention, namely the one-step persymmetric coordinate detector (PerCD-1S) and the second robust detector, namely the two-step persymmetric coordinate detector (PerCD-2S), are optimal. Their detection probabilities under the same signal to interference plus noise ratio (SINR) can differ by up to about 0.3 from those of other comparative detectors. The detection probabilities of GLRT and adaptive matched filter (AMF) are second, and the detection performance of subspace detector (Spatial Domain, SD) and subspace adaptive matched filter (Spatial Domain Adaptive Matched Filter, SD-AMF) is the lowest. In the case of mismatch, the detectors PerCD-1S and PerCD-2S proposed in the present invention still have the best detection performance. The detection performance curves of the non-mismatched detectors GLRT and AMF begin to decline due to the mismatch of the steering vector, which are lower than those of the subspace detectors SD and SD-AMF. Figure 3 (c) with Figure 3 In (d), no matter in the case of mismatch or no mismatch, the detection performance of the one-step joint detector PerCD-1S and the two-step joint detector PerCD-2S of the present invention is the best, and the detection probability thereof can differ from other comparison detectors by about 0.9 at most, and compared with the detection probability when the training data is sufficient, the detection probability of the detector proposed in the present invention at the same SINR when K=16 is at most attenuated by only about 0.3; since the amount of training data is reduced to half of the original amount, the detection performance of the no-mismatch detector GLRT and AMF, and the subspace detector SD and SD-AMF are all at a relatively low level. Therefore Figure 3 It shows that when the amount of training data is insufficient and the steering vector is mismatched, the first robust detector and the second robust detector of the present invention can still maintain good detection performance.
[0186] Figure 4 The MSE curve of the detector of the present invention is schematically shown, specifically the MSE curve of the first robust detector PerCD-1S under different training data, wherein the angle error is defined as Where R M is the number of Monte Carlo experiments, Δθ=[Δu,Δv] is the true target cosine offset, is the estimated target cosine offset. Figure 4(a) It can be seen that, when the steering vector is not mismatched, as the number of training samples increases, the estimated performance curve of the detector proposed by the present invention decreases faster; Figure 4 (b) shows that when the steering vector is mismatched, the MSE curve value is higher than that when there is no mismatch under the same SINR, but the MSE gradually decreases as the amount of training data increases. Figure 4 From the MSE curve, it can be seen that the detector proposed in the present invention has a better estimation performance in terms of orientation estimation, and the more training samples there are, the faster the convergence speed of the MSE curve.
[0187] Based on the above Figure 1 It can be seen from the implementation method that when the radar echo is received, the embodiment of the present invention constructs a target-oriented vector model in full increment form according to the three-dimensional spatial position of the target and the target azimuth position estimated by the radar; constructs a binary detection model according to the target-oriented vector model, the noise component in the test data, the noise component in the training data and the signal amplitude, and determines the covariance matrix of the target according to the noise component in the test data and the noise component in the training data. The binary detection model is used to determine the test data and the training data; determines the joint probability density function of the test data matrix and the training data matrix according to the inverse diagonal matrix, the covariance matrix of the target, the estimated covariance matrix, the signal amplitude matrix, the first unitary matrix and the second unitary matrix; determines the first preset detection formula and the second preset detection formula of the inverse diagonal parameter according to the joint probability density function and the generalized likelihood ratio test criterion, and performs score optimization on the first preset detection formula and the second preset detection formula to obtain the corresponding first final detection formula and the second final detection formula; solves the first final detection formula and the second final detection formula respectively according to the rotating coordinate method to construct the first robust detector and the second robust detector; detects whether the target exists according to the Monte Carlo method, the first robust detector and the second robust detector. In this way, a binary detection model of joint target detection and azimuth estimation based on the target-guided vector is constructed by using the target-guided vector model constructed in full incremental form. When the target is detected, the first final detection formula and the second final detection formula are solved by using the rotation coordinate method, and the target cosine offset can be adaptively output, thereby slowing down the sharp decline in the detection performance of the two-dimensional array radar when the target-guided vector is mismatched; the first preset detection formula and the second preset detection formula are derived by using the inverse diagonal matrix, the first unitary matrix and the second unitary matrix, and the inverse diagonal structure is used so that half of the parameters are needed to characterize the covariance matrix of the unknown target, which greatly reduces the dependence on the number of training data, so that the sensitivity of target detection is higher when the training data is reduced and the guidance vector is mismatched.
[0188] Based on the same inventive concept, as an implementation of the above-mentioned target detection method based on the inverse diagonal joint robust detector, an embodiment of the present invention also provides a target detection device based on the inverse diagonal joint robust detector. Figure 5 is a structural diagram of the device in the embodiment of the present invention, see Figure 5 As shown, the device may include:
[0189] The first construction module 501 is used to construct a target guidance vector model in a full increment form according to the three-dimensional spatial position of the target and the target azimuth position estimated by the radar when the radar echo is received;
[0190] A second construction module 502 is used to construct a binary detection model according to the target-oriented vector model, the noise component in the test data, the noise component in the training data, and the signal amplitude, and determine the covariance matrix of the target according to the noise component in the test data and the noise component in the training data, and the binary detection model is used to determine the test data and the training data;
[0191] A determination module 503 is used to determine a joint probability density function of a test data matrix and a training data matrix according to the inverse diagonal matrix, the covariance matrix of the target, the estimated covariance matrix, the signal amplitude matrix, the first unitary matrix and the second unitary matrix, wherein the test data matrix is determined by the test data and the training data matrix is determined by the training data;
[0192] A score optimization module 504 is used to determine a first preset detection formula and a second preset detection formula of the inverse diagonal parameter according to a joint probability density function and a generalized likelihood ratio test criterion, and to perform score optimization on the first preset detection formula and the second preset detection formula to obtain a corresponding first final detection formula and a second final detection formula;
[0193] A third construction module 505 is used to solve the first final detection formula and the second final detection formula respectively according to the rotation coordinate method to construct a first robust detector and a second robust detector;
[0194] The detection module 506 is used to detect whether there is a target according to the Monte Carlo method, the first robust detector and the second robust detector.
[0195] The first construction module 501 is specifically used to convert the three-dimensional spatial position of the target into the cosine space when receiving the radar echo, so as to obtain the real target direction position, wherein the real target direction position includes the first real target direction position and the second real target direction position; form a preset steering vector of the target according to the first real target direction position, the second real target direction position, the number of row array elements and the number of column array elements in the uniform rectangular distribution array; determine a first direction cosine offset according to the first real target direction position and the first estimated target azimuth position, and determine a second direction cosine offset according to the second real target direction position and the second estimated target azimuth position; construct a target steering vector model according to the first estimated target azimuth position, the second estimated target azimuth position, the preset steering vector of the target, the first direction cosine offset and the second direction cosine offset; the target azimuth position estimated by the radar includes the first estimated target azimuth position and the second estimated target azimuth position.
[0196] The first construction module 501 constructs a target steering vector model according to the first estimated target azimuth position, the second estimated target azimuth position, the preset steering vector of the target, the first direction cosine offset and the second direction cosine offset, including: replacing the first real target direction position in the preset steering vector of the target with the first estimated target azimuth position, and replacing the second real target direction position with the second estimated target azimuth position to obtain the estimated steering vector of the target; using the first direction cosine offset and the second direction cosine offset, Taylor expansion is performed on the first estimated target azimuth position and the second estimated target azimuth position in the estimated steering vector of the target to obtain the target steering vector model; the preset steering vector of the target includes the first real target direction position and the second real target direction position.
[0197] The first construction module 501 uses the first direction cosine offset and the second direction cosine offset to perform Taylor expansion on the first estimated target orientation position and the second estimated target orientation position in the estimated steering vector of the target to obtain a target steering vector model, including: using the first direction cosine offset and the second direction cosine offset to perform Taylor expansion on the first estimated target orientation position and the second estimated target orientation position in the estimated steering vector of the target, and using the following formula to obtain the target steering vector model:
[0198]
[0199]
[0200] Among them, g a (Δu,Δv) is the target-oriented vector model, is the estimated steering vector of the target, Δu is the first direction cosine offset, Δv is the second direction cosine offset, is the first estimated target position, is the second estimated target position, is a complex matrix space of dimension Z×1, where Z is the product of the number of row array elements and the number of column array elements, Z=NM, where N is the number of row array elements and M is the number of column array elements. Calculate the partial derivative of the target-oriented vector model for the first true target direction position u, The partial derivative of the target-oriented vector model for the second real target direction position v is obtained, j is the imaginary unit, λ0 is the wavelength of the radar transmission wave, For each antenna element pair from The weighted response of the signal in the direction, x N-1 is the coordinate position of the N-1th row element in the two-dimensional array, For each antenna element pair from The weighted response of the signal in the direction, ⊙ is the Kronecker product, y M-1 is the coordinate position of the M-1th row element in the two-dimensional array, It is Hadamard.
[0201] Determination module 503 is specifically used to determine the target true covariance matrix according to the conjugate matrix corresponding to the inverse diagonal matrix and the target covariance matrix; determine the test data matrix according to the test data, and determine the training data matrix according to the training data; determine the preset joint probability density function of the test data matrix and the training data matrix according to the target true covariance matrix, the estimated covariance matrix, the adjacency matrix corresponding to the test data matrix, and the adjacency matrix corresponding to the signal amplitude matrix; use the first unitary matrix and the second unitary matrix to perform unitary transformation on the adjacency matrix corresponding to the signal amplitude matrix, the adjacency matrix corresponding to the test data matrix, the estimated covariance matrix, and the covariance matrix of the target, respectively, to obtain the unitary transformed real amplitude matrix, the unitary transformed real test unit data matrix, the unitary transformed real estimated covariance matrix, and the unitary transformed real steering vector; substitute the unitary transformed real amplitude matrix, the unitary transformed real test unit data matrix, the unitary transformed real estimated covariance matrix, and the unitary transformed real steering vector into the preset joint probability density function to obtain the joint probability density function.
[0202] In the score optimization module 504, the first final detection formula is expressed as:
[0203]
[0204] in, is the actual steering vector of the target whitening, is the whitening partial steering matrix, Δθ is the target cosine deviation angle information, is the first whitened covariance matrix, is the transpose of the first whitened covariance matrix,
[0205] D p is the first unitary matrix, J is the inverse diagonal matrix, g * is the conjugate of the estimated steering vector of the target, is the real estimated covariance matrix after unitary transformation, G * is the conjugate of the concatenated partial steering matrix, R r is the actual test unit data matrix after unitary transformation, φ is the variable, I is the unit matrix, is the transpose of the actual test unit data matrix after unitary transformation, α is the first constraint parameter, β is the second constraint parameter, Δu is the first direction cosine offset, and Δv is the second direction cosine offset;
[0206] The second final detection formula is expressed as:
[0207]
[0208] in, is the second whitened covariance matrix, is the transpose of the second whitened covariance matrix, S r is a known real covariance matrix, D p is the first unitary matrix.
[0209] The third building block 505, the expression of the first robust detector is:
[0210]
[0211] in, R r is the real test unit data matrix after unitary transformation, is the transpose of the real test unit data matrix after unitary transformation, is the real estimated covariance matrix after unitary transformation, is the actual guidance vector of the first target after optimization, is the first target amplitude, is the condition for the existence of target echo signal, is the condition that there is no target echo signal, ξ is the set detection threshold for ensuring constant false alarm, D p is the first unitary matrix, J is the inverse diagonal matrix, g is the estimated steering vector of the target, G is the partial steering matrix after splicing, is the first optimal azimuth, S r is a known real covariance matrix, is the conjugate transpose of the optimized actual steering vector of the first target;
[0212] The expression of the second robust detector is:
[0213]
[0214] in, is the transpose of the second whitened covariance matrix, is the optimized second partial derivative projection matrix, is the second whitening covariance matrix, Δu is the first direction cosine offset, Δv is the second direction cosine offset, is the actual guidance vector of the optimized second target, is the transpose of the actual guidance vector of the optimized second target, is the second best azimuth.
[0215] The detection module 506 is specifically used to generate noise data using the Monte Carlo method under the condition that there is no target echo signal; input the noise data into the first robust detector and the second robust detector to output multiple detection thresholds; select a preset detection threshold from the multiple detection thresholds; generate noise data with a target-oriented vector using the Monte Carlo method under the condition that there is a target echo signal; input the noise data with the target-oriented vector into the first robust detector and the second robust detector to output a detection value; determine whether the detection value is greater than the preset detection threshold, if so, detect that a target exists, if not, detect that a target does not exist.
[0216] It should be pointed out here that: the above description of the embodiment of the target detection device based on the inverse diagonal joint robust detector is similar to the description of the embodiment of the target detection method based on the inverse diagonal joint robust detector, and has similar beneficial effects as the embodiment of the target detection method based on the inverse diagonal joint robust detector. For technical details not disclosed in the embodiment of the target detection device based on the inverse diagonal joint robust detector of the embodiment of the present invention, please refer to the description of the embodiment of the target detection method based on the inverse diagonal joint robust detector of the present invention for understanding.
[0217] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. A target detection method based on an inverse diagonal joint robust detector, characterized in that: The target detection method comprises: When receiving the radar echo, a target guidance vector model is constructed in full increment form according to the three-dimensional spatial position of the target and the target azimuth position estimated by the radar; Constructing a binary detection model according to the target-oriented vector model, the noise component in the test data, the noise component in the training data, and the signal amplitude, and determining a covariance matrix of the target according to the noise component in the test data and the noise component in the training data, wherein the binary detection model is used to determine the test data and the training data; Determine a joint probability density function of a test data matrix and a training data matrix according to an inverse diagonal matrix, a covariance matrix of the target, an estimated covariance matrix, a signal amplitude matrix, a first unitary matrix, and a second unitary matrix, wherein the test data matrix is determined by the test data and the training data matrix is determined by the training data; Determine a first preset detection formula and a second preset detection formula of the inverse diagonal parameter according to the joint probability density function and the generalized likelihood ratio test criterion, and perform score optimization on both the first preset detection formula and the second preset detection formula to obtain corresponding first final detection formula and second final detection formula; Solving the first final detection formula and the second final detection formula respectively according to the rotation coordinate method to construct a first robust detector and a second robust detector; Whether the target exists is detected according to the Monte Carlo method, the first robust detector and the second robust detector.
2. The target detection method according to claim 1, characterized in that: The target azimuth position estimated by the radar includes a first estimated target azimuth position and a second estimated target azimuth position, and constructing a target guidance vector model in a full increment form according to the three-dimensional spatial position of the target and the target azimuth position estimated by the radar includes: When the radar echo is received, the three-dimensional spatial position of the target is converted into a cosine space to obtain a true target direction position, wherein the true target direction position includes a first true target direction position and a second true target direction position; According to the first real target direction position, the second real target direction position, the number of row array elements and the number of column array elements in the uniform rectangular distribution array, a preset target guidance vector is formed; Determine a first direction cosine offset according to the first true target direction position and the first estimated target azimuth position, and determine a second direction cosine offset according to the second true target direction position and the second estimated target azimuth position; The target steering vector model is constructed according to the first estimated target azimuth position, the second estimated target azimuth position, a preset steering vector of the target, the first direction cosine offset, and the second direction cosine offset.
3. The target detection method according to claim 2, characterized in that: The preset steering vector of the target includes the first real target direction position and the second real target direction position, and constructing the target steering vector model according to the first estimated target azimuth position, the second estimated target azimuth position, the preset steering vector of the target, the first direction cosine offset and the second direction cosine offset includes: Replacing the first real target direction position in the preset steering vector of the target with the first estimated target orientation position, and replacing the second real target direction position with the second estimated target orientation position, so as to obtain an estimated steering vector of the target; The first direction cosine offset and the second direction cosine offset are used to perform Taylor expansion on the first estimated target orientation position and the second estimated target orientation position in the estimated steering vector of the target to obtain the target steering vector model.
4. The target detection method according to claim 3, characterized in that: The step of performing Taylor expansion on the first estimated target orientation position and the second estimated target orientation position in the estimated steering vector of the target by using the first direction cosine offset and the second direction cosine offset to obtain the target steering vector model includes: The first direction cosine offset and the second direction cosine offset are used to perform Taylor expansion on the first estimated target orientation position and the second estimated target orientation position in the estimated steering vector of the target, and the target steering vector model is obtained using the following formula: Among them, g a (Δu, Δv) is the target-oriented vector model, is the estimated steering vector of the target, Δu is the first direction cosine offset, Δv is the second direction cosine offset, is the first estimated target position, is the second estimated target position, is a complex matrix space of dimension Z×1, Z is the product of the number of row array elements and the number of column array elements, Z=NM, N is the number of row array elements, M is the number of column array elements, Calculate the partial derivative of the target-oriented vector model with respect to the first real target direction position u, Calculate the partial derivative of the target-oriented vector model for the second real target direction position v, j is an imaginary unit, λ0 is the wavelength of the radar transmission wave, For each antenna element pair from The weighted response of the signal in the direction, x N-1 is the coordinate position of the N-1th row element in the two-dimensional array, For each antenna element pair from The weighted response of the signal in the direction, ⊙ is the Kronecker product, y M-1 is the coordinate position of the M-1th row element in the two-dimensional array, It is Hadamard.
5. The target detection method according to claim 1, characterized in that: Determining a joint probability density function of a test data matrix and a training data matrix according to the inverse diagonal matrix, the covariance matrix of the target, the estimated covariance matrix, the signal amplitude matrix, the first unitary matrix, and the second unitary matrix includes: Determine the target true covariance matrix according to the conjugate matrix corresponding to the inverse diagonal matrix and the covariance matrix of the target; Determine the test data matrix according to the test data, and determine the training data matrix according to the training data; Determine a preset joint probability density function of the test data matrix and the training data matrix according to the target true covariance matrix, the estimated covariance matrix, the adjacency matrix corresponding to the test data matrix, and the adjacency matrix corresponding to the signal amplitude matrix; Using the first unitary matrix and the second unitary matrix, unitary transformation is performed on the adjacency matrix corresponding to the signal amplitude matrix, the adjacency matrix corresponding to the test data matrix, the estimated covariance matrix and the covariance matrix of the target, respectively, to obtain a unitary transformed real amplitude matrix, a unitary transformed real test unit data matrix, a unitary transformed real estimated covariance matrix and a unitary transformed real steering vector; Substitute the real amplitude matrix after the unitary transformation, the real test unit data matrix after the unitary transformation, the real estimated covariance matrix after the unitary transformation and the real steering vector after the unitary transformation into the preset joint probability density function to obtain the joint probability density function.
6. The target detection method according to claim 5, characterized in that: Substituting the unitary transformed real amplitude matrix, the unitary transformed real test unit data matrix, the unitary transformed real estimated covariance matrix and the unitary transformed real steering vector into the preset joint probability density function to obtain the joint probability density function includes: Substitute the unitary transformed real amplitude matrix, the unitary transformed real test unit data matrix, the unitary transformed real estimated covariance matrix and the unitary transformed real steering vector into the preset joint probability density function, and use the following formula to obtain the joint probability density function: in, is the joint probability density function, R is the test data matrix, R s is the training data matrix, is the condition without target echo signal, is the condition with target echo signal, S r is a known real covariance matrix, is the known actual test data information matrix, Z is the product of the number of row array elements and the number of column array elements, Z = NM, N is the number of row array elements, M is the number of column array elements, K is the number of training data, D p is the first unitary matrix, S is the target true covariance matrix, is the matrix conjugate transpose, For the real part, J is the inverse diagonal matrix, To find the imaginary part, is the real estimated covariance matrix after the unitary transformation, R r is the actual test unit data matrix after the unitary transformation, g r is the real steering vector after the unitary transformation, is the transpose of the real amplitude matrix after the unitary transformation.
7. The target detection method according to claim 6, characterized in that: The first final detection formula is expressed as: in, is the actual steering vector of the target whitening, is the whitening partial steering matrix, Δθ is the target cosine deviation angle information, is the first whitened covariance matrix, is the transpose of the first whitened covariance matrix, D p is the first unitary matrix, J is the inverse diagonal matrix, g * is the conjugate of the estimated steering vector of the target, is the real estimated covariance matrix after the unitary transformation, G * is the conjugate of the concatenated partial steering matrix, R r is the actual test unit data matrix after the unitary transformation, φ is a variable, I is the unit matrix, is the transpose of the actual test unit data matrix after the unitary transformation, α is the first constraint parameter, β is the second constraint parameter, Δu is the first direction cosine offset, and Δv is the second direction cosine offset; The second final detection formula is expressed as: in, is the second whitened covariance matrix, is the transpose of the second whitened covariance matrix, S r is the known real covariance matrix, D p is the first unitary matrix.
8. The target detection method according to claim 7, characterized in that: The expression of the first robust detector is: in, R r is the real test unit data matrix after the unitary transformation, is the transpose of the real test unit data matrix after the unitary transformation, is the real estimated covariance matrix after the unitary transformation, is the actual guidance vector of the first target after optimization, is the first target amplitude, is the condition for the existence of target echo signal, is the condition that there is no target echo signal, ξ is the set detection threshold for ensuring constant false alarm, D p is the first unitary matrix, J is the inverse diagonal matrix, g is the estimated steering vector of the target, G is the partial steering matrix after splicing, is the first optimal azimuth, S r is the known real covariance matrix, is the conjugate transpose of the optimized actual steering vector of the first target; The expression of the second robust detector is: in, is the transpose of the second whitened covariance matrix, is the optimized second partial derivative projection matrix, is the second whitening covariance matrix, Δu is the first direction cosine offset, Δv is the second direction cosine offset, is the actual guidance vector of the optimized second target, is the transpose of the optimized actual steering vector of the second target, is the second best azimuth.
9. The target detection method according to claim 1, characterized in that: The detecting whether the target exists according to the Monte Carlo method, the first robust detector and the second robust detector comprises: In the absence of a target echo signal, generating noise data using the Monte Carlo method; inputting the noise data into the first robust detector and the second robust detector to output a plurality of detection thresholds; Selecting a preset detection threshold from the multiple detection thresholds; In the presence of the target echo signal, using the Monte Carlo method to generate noise data with a target-guided vector; inputting the noise data with the target-oriented vector into the first robust detector and the second robust detector to output a detection value; Determine whether the detection value is greater than the preset detection threshold, if so, detect that the target exists, if not, detect that the target does not exist.
10. An object detection device based on an inverse diagonal joint robust detector, characterized in that: The device comprises: The first construction module is used to construct a target guidance vector model in a full increment form according to the three-dimensional spatial position of the target and the target azimuth position estimated by the radar when the radar echo is received; A second construction module is used to construct a binary detection model according to the target-oriented vector model, the noise component in the test data, the noise component in the training data and the signal amplitude, and determine the covariance matrix of the target according to the noise component in the test data and the noise component in the training data, wherein the binary detection model is used to determine the test data and the training data; A determination module, configured to determine a joint probability density function of a test data matrix and a training data matrix according to an inverse diagonal matrix, a covariance matrix of the target, an estimated covariance matrix, a signal amplitude matrix, a first unitary matrix, and a second unitary matrix, wherein the test data matrix is determined by the test data and the training data matrix is determined by the training data; A score optimization module, used for determining a first preset detection formula and a second preset detection formula of the inverse diagonal parameter according to the joint probability density function and the generalized likelihood ratio test criterion, and performing score optimization on both the first preset detection formula and the second preset detection formula to obtain a corresponding first final detection formula and a second final detection formula; A third construction module is used to solve the first final detection formula and the second final detection formula respectively according to the rotation coordinate method to construct a first robust detector and a second robust detector; A detection module is used to detect whether the target exists according to the Monte Carlo method, the first robust detector and the second robust detector.
Citation Information
Patent Citations
A radar target adaptive detection method based on anti-symmetric transformation
CN108535708A
Distributed target detection method based on antisymmetric transformation and parameter adjustability
CN108919219A
Mismatch robust subspace signal detection method and device
CN112149516A
Radar target detection method and system based on symmetric power spectral density
CN114994632A
