A method, device and medium for target detection under non-uniform clutter background

Abstract

The embodiment of the present application discloses a method, device and medium for target detection under a non-uniform clutter background, which belongs to the field of radar signal processing technology and solves the problem of low accuracy of traditional target detection methods. A sample covariance matrix is ​​constructed based on radar slow-time echo data, and a binary hypothesis test problem based on the sample covariance matrix is ​​established in the Riemann manifold; modeling is performed based on the Riemann distance through the Riemann Gaussian distribution to obtain the statistical distribution characteristics of the sample covariance matrix on the Riemann manifold, so as to obtain a target detector based on the statistical distribution characteristics; based on the sample data of the reference unit obtained, the covariance matrix of the clutter is constructed, and the covariance matrix of the target signal is determined based on the radar slow-time echo data; based on the covariance matrix of the clutter, the covariance matrix of the target signal and the target detector, a detection statistic based on the Riemann distance is obtained, and the detection statistic is compared with the detection threshold to determine the detection target.

Description

Technical Field

[0001] The present application relates to the field of radar signal processing technology, and in particular to a method, device and medium for detecting a target under a non-uniform clutter background. Background Art

[0002] The problem of signal detection in clutter background is a research difficulty in the fields of radar target detection and communication. In actual detection scenarios, the distribution of clutter background is usually unknown. Traditional detection methods generally adopt a two-step generalized likelihood ratio test (GLRT) detection framework, that is, first, the optimal detector is obtained based on the Neyman-Pearson criterion under the assumption that the distribution is completely known. Then, for problems involving unknown parameters such as the clutter covariance matrix, the maximum likelihood (ML) estimation is used to replace the unknown parameters. This type of detector can only achieve good detection performance when the parameters such as the clutter covariance matrix are accurately estimated, and sufficient independent and identically distributed reference unit samples are required for parameter estimation.

[0003] However, for clustered targets such as drones and migratory organisms, the distances between different targets are usually close, and reference cells are prone to target interference. At the same time, high-resolution clutter usually has spatial non-uniform characteristics. This will limit the number of samples of reference cells with the same distribution, affect the estimation accuracy of parameters such as the clutter covariance matrix, and result in low accuracy of traditional target detection methods. Summary of the invention

[0004] The embodiments of the present application provide a method, device and medium for target detection under a non-uniform clutter background, which are used to solve the following technical problems: for cluster targets such as drones and migratory organisms, the distances between different targets are usually close, and target interference is prone to occur in reference units. At the same time, high-resolution clutter usually has spatial non-uniform characteristics. This will limit the number of samples of reference units with the same distribution, affect the estimation accuracy of parameters such as the clutter covariance matrix, and result in a low accuracy rate of traditional target detection methods.

[0005] The present application embodiment adopts the following technical solutions:

[0006] The embodiment of the present application provides a method for target detection under a non-uniform clutter background. It includes: constructing a sample covariance matrix based on the acquired radar slow-time dimension echo data, mapping the sample covariance matrix to a Riemann manifold, so as to establish a binary hypothesis test problem based on the sample covariance matrix in the Riemann manifold; determining the distance between the sample covariance matrices in the Riemann manifold through the Riemann metric, modeling based on the distance through the Riemann Gaussian distribution, obtaining the statistical distribution characteristics of the sample covariance matrix on the Riemann manifold, and obtaining a target detector based on the statistical distribution characteristics; constructing the covariance matrix of the clutter based on the acquired sample data of the reference unit, and determining the covariance matrix of the target signal based on the radar slow-time dimension echo data; wherein the reference unit is a sample unit whose distance from the target is less than a preset distance threshold; obtaining a detection statistic based on the Riemann distance based on the covariance matrix of the clutter, the covariance matrix of the target signal and the target detector, comparing the detection statistic with the detection threshold, and determining the detection target based on the comparison result.

[0007] The embodiment of the present application maps the sample covariance matrix to the Riemann manifold, making full use of the framework of Riemannian geometry to process high-dimensional, nonlinear data, making statistical analysis and hypothesis testing on the manifold more intuitive and effective. By establishing a binary hypothesis test problem based on the sample covariance matrix in the Riemann manifold, the statistical characteristics of the two cases where the target exists and does not exist are clearly distinguished, and the sample covariance matrix is ​​modeled through the Riemann Gaussian distribution to capture the statistical distribution characteristics of the data on the Riemann manifold, thereby more accurately describing the statistical behavior of the data and improving the accuracy of target detection. The target detector built based on the statistical distribution characteristics can use the statistical information of the sample covariance matrix to detect the target, improve the sensitivity and accuracy of target detection, and solve the problem of inaccurate target detection under the background of non-uniform clutter.

[0008] In one implementation of the present application, a binary hypothesis testing problem based on a sample covariance matrix is ​​established in a Riemann manifold, specifically including:

[0009] Function-based: ;

[0010] Construct a binary hypothesis testing problem on a Riemannian manifold, where; It means that the assumption is without a goal; It means that there is a goal in the assumption; represents the probability density function of the covariance matrix in the manifold of positive definite matrices; is the sample covariance matrix; for Assume that the sample covariance matrix is The distribution parameters of .

[0011] In one implementation of the present application, modeling is performed based on the distance through the Riemann-Gaussian distribution to obtain the statistical distribution characteristics of the sample covariance matrix on the Riemann manifold, specifically including: constructing the Riemann volume element corresponding to the sample covariance matrix based on the distance; wherein the Riemann volume element is the product of the lengths of each side of the cube; based on the Riemann volume element, determining the probability density function of the Riemann-Gaussian distribution corresponding to the sample covariance matrix, and determining the statistical distribution characteristics of the sample covariance matrix on the Riemann manifold based on the probability density function of the Riemann-Gaussian distribution.

[0012] In one implementation of the present application, after determining the probability density function of the Riemann-Gaussian distribution corresponding to the sample covariance matrix based on the Riemann volume infinitesimal, the method further includes:

[0013] Through the distribution of eigenvalues ​​and characteristic matrices, the conditions for samples to obey the Riemann-Gaussian distribution are determined;

[0014] The eigenvalue decomposition of the matrix is:

[0015] ;

[0016] The eigenvalues ​​satisfy the following joint probability density function:

[0017] ;

[0018] in, is a random sample matrix; is a diagonal matrix, Representation Matrix The logarithm of the eigenvalue, N is the number of pulses in the slow time dimension; the characteristic matrix composed of eigenvectors is an orthogonal matrix; In the case of Riemann-Gaussian distribution, is a uniformly distributed orthogonal matrix; is the normalization coefficient of the logarithmic distribution of the sample eigenvalues ​​of the Riemann-Gaussian distribution matrix; is the standard deviation of the Riemann-Gaussian distribution; p RGD is the joint probability density function.

[0019] The relationship between the standard deviation of the Riemann-Gaussian distribution and the number of samples is determined as:

[0020] ;

[0021] in, is the standard deviation of the Riemann-Gaussian distribution; is the sample size;

[0022] When the sample covariance matrix corresponding to the multivariate Gaussian observations conforms to the Wishart distribution, the joint probability density function of the eigenvalues ​​conforming to the Wishart distribution is:

[0023] ;

[0024] in, , represents the normalization coefficient; represents the logarithm of the eigenvalues ​​of the sample covariance matrix; is the sample size; f W is the joint probability density function that conforms to the eigenvalues ​​of the Wishart distribution; x i For the i Observation data; x j For the j Observation data.

[0025] Based on the joint probability density function of the eigenvalues ​​that conform to the Wishart distribution, it is determined that the sample covariance matrix approaches the Riemann-Gaussian distribution in the symmetric positive definite matrix manifold.

[0026] In one implementation of the present application, a target detector is obtained based on statistical distribution characteristics, specifically including: the target detector is:

[0027] ;

[0028] in, It means that the assumption is without a goal; It means that there is a goal in the assumption; Represents the probability density function of the covariance matrix in the manifold of positive definite matrices when there is a target; represents the probability density function of the covariance matrix in the manifold of positive definite matrices when there is no target; is the sample covariance matrix; for Assume that the sample covariance matrix is The distribution parameters of for Assume that the sample covariance matrix is The distribution parameters of is the detection threshold.

[0029] In one implementation of the present application, a covariance matrix of clutter is constructed based on sample data of a reference unit obtained, specifically including: constructing a log-likelihood function based on multiple independent samples obtained that obey a Riemann-Gaussian distribution; normalizing the coefficients of the log-likelihood function; obtaining a maximum likelihood estimation function of the mean based on the normalized log-likelihood function; iterating the maximum likelihood estimation function of the mean through a gradient descent algorithm to obtain a corresponding numerical solution, so as to obtain the covariance matrix of the clutter signal based on the corresponding numerical solution.

[0030] In one implementation of the present application, the covariance matrix of the target signal is determined based on the radar slow-time dimension echo data, specifically including: inputting the radar slow-time dimension echo data into a preset signal power maximum likelihood estimation function to obtain the maximum likelihood value corresponding to the target signal; based on the maximum likelihood value and the preset covariance matrix function, obtaining the covariance matrix of the target signal.

[0031] In one implementation of the present application, based on the covariance matrix of the clutter, the covariance matrix of the target signal and the target detector, a detection statistic based on the Riemann distance is obtained, specifically including: inputting the covariance matrix of the clutter and the covariance matrix of the target signal into the target detector to obtain a detection statistic based on the Riemann distance; wherein the detection statistic is:

[0032] ;

[0033] in, is the sample covariance matrix; is the covariance matrix of the clutter; is the covariance matrix of the target signal; It means that the assumption is without a goal; It means that there is a goal in the assumption; represents the Riemann distance, Indicates the detection threshold.

[0034] The embodiment of the present application provides a target detection device under a non-uniform clutter background, comprising: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: construct a sample covariance matrix based on acquired radar slow-time echo data, map the sample covariance matrix to a Riemann manifold, so as to establish a binary hypothesis test problem based on the sample covariance matrix in the Riemann manifold; determine the distance between the sample covariance matrices in the Riemann manifold through the Riemann metric, and calculate the distance based on the distance through the Riemann metric; Riemann-Gaussian distribution is used to model the sample covariance matrix on the Riemann manifold, and the statistical distribution characteristics of the sample covariance matrix on the Riemann manifold are obtained, so as to obtain a target detector based on the statistical distribution characteristics; based on the sample data of the reference unit obtained, the covariance matrix of the clutter is constructed, and the covariance matrix of the target signal is determined based on the radar slow-time echo data; wherein, the reference unit is a sample unit whose distance to the target is less than a preset distance threshold; based on the covariance matrix of the clutter, the covariance matrix of the target signal and the target detector, a detection statistic based on the Riemann distance is obtained, the detection statistic is compared with the detection threshold, and the detection target is determined based on the comparison result.

[0035] A non-volatile computer storage medium provided by an embodiment of the present application stores computer executable instructions, wherein the computer executable instructions are configured to: construct a sample covariance matrix based on acquired radar slow-time dimension echo data, map the sample covariance matrix to a Riemann manifold, so as to establish a binary hypothesis test problem based on the sample covariance matrix in the Riemann manifold; determine the distance between the sample covariance matrices in the Riemann manifold by Riemann metric, model the sample covariance matrix on the Riemann manifold based on the distance by Riemann Gaussian distribution, obtain the statistical distribution characteristics of the sample covariance matrix on the Riemann manifold, and obtain a target detector based on the statistical distribution characteristics; construct a clutter covariance matrix based on acquired reference unit sample data, and determine a target signal covariance matrix based on radar slow-time dimension echo data; wherein the reference unit is a sample unit whose distance from the target is less than a preset distance threshold; obtain a detection statistic based on the Riemann distance based on the clutter covariance matrix, the target signal covariance matrix and the target detector, compare the detection statistic with the detection threshold, and determine the detection target based on the comparison result.

[0036] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects: The embodiments of the present application map the sample covariance matrix to the Riemann manifold, making full use of the framework of Riemannian geometry to process high-dimensional, nonlinear data, so that statistical analysis and hypothesis testing on the manifold become more intuitive and effective. By establishing a binary hypothesis test problem based on the sample covariance matrix in the Riemann manifold, the statistical characteristics of the two cases where the target exists and does not exist are clearly distinguished, and the sample covariance matrix is ​​modeled through the Riemann Gaussian distribution to capture the statistical distribution characteristics of the data on the Riemann manifold, thereby more accurately describing the statistical behavior of the data and improving the accuracy of target detection. The target detector constructed based on the statistical distribution characteristics can use the statistical information of the sample covariance matrix to detect the target, improve the sensitivity and accuracy of target detection, and solve the problem of inaccurate target detection under the background of non-uniform clutter. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application 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 recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. In the drawings:

[0038] Figure 1 A flow chart of a method for detecting a target under a non-uniform clutter background provided by an embodiment of the present application;

[0039] Figure 2 A schematic diagram of specific information of measured meteorological clutter and target data collected by a Ku-band high-resolution phased array radar provided in an embodiment of the present application;

[0040] Figure 3 A schematic diagram of the marginal probability density function of the logarithm of the eigenvalue of a sample covariance matrix provided in an embodiment of the present application, wherein (a) is the eigenvalue r 1. Schematic diagram of marginal probability density function with 16 observed samples, (b) is the characteristic value r 2. Schematic diagram of marginal probability density function with 16 observed samples, (c) is the characteristic value r 1. Schematic diagram of marginal probability density function with 64 observed samples, (d) is the characteristic value r 2. Schematic diagram of marginal probability density function when the number of observed samples is 64;

[0041] Figure 4Schematic diagram of detection probability under different signal-to-noise ratios for an optimal detector based on the Neyman-Pearson criterion and a likelihood ratio detection algorithm based on the Riemann-Gaussian distribution provided in an embodiment of the present application, wherein (a) is a schematic diagram for an observation sample number of 9, (b) is a schematic diagram for an observation sample number of 18, (c) is a schematic diagram for an observation sample number of 45, and (d) is a schematic diagram for an observation sample number of 90;

[0042] Figure 5 A schematic diagram for comparing detection probabilities of different signal-to-clutter ratios of meteorological clutter data with existing methods provided in an embodiment of the present application, wherein (a) is a schematic diagram for comparing data set 1, (b) is a schematic diagram for comparing data set 2, and (c) is a schematic diagram for comparing data set 3;

[0043] Figure 6 A schematic diagram of a comparison of detection probabilities of different signal-to-clutter ratios based on public shore-based sea clutter data with existing methods provided in an embodiment of the present application, wherein Figure (a) is a schematic diagram of a comparison corresponding to the first data set, and Figure (b) is a schematic diagram of a comparison corresponding to the second data set;

[0044] Figure 7 A schematic diagram of the structure of a target detection device under a non-uniform clutter background provided in an embodiment of the present application.

[0045] Reference numerals:

[0046] 200 target detection device under non-uniform clutter background, 201 processor, 202 memory. DETAILED DESCRIPTION

[0047] The embodiments of the present application provide a method, device and medium for detecting a target under a non-uniform clutter background.

[0048] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0049] The technical solution proposed in the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0050] Figure 1 A flow chart of a method for detecting a target under a non-uniform clutter background provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the target detection method under non-uniform clutter background includes the following steps:

[0051] Step 101: construct a sample covariance matrix based on the acquired radar slow-time echo data, and map the sample covariance matrix to a Riemannian manifold to establish a binary hypothesis testing problem based on the sample covariance matrix in the Riemannian manifold.

[0052] In one embodiment of the present application, the embodiment of the present application uses radar slow-time echo to construct a sample covariance matrix and maps the observed data to a Riemann manifold.

[0053] Assume that the number of CPIs (Coherent Processing Intervals) for a single observation is , the number of slow-time pulses in a single CPI is . No. The slow-time echo in a CPI is expressed as , the observed data is expressed as ,in yes The zero-mean independent observation vector of Constructing the sample covariance matrix :

[0054] ;

[0055] Based on the sample covariance matrix , establish the binary hypothesis testing problem on the Riemann manifold:

[0056] ;

[0057] in; Represents the observation data matrix The conjugate transpose of ; It means that the assumption is without a goal; It means that there is a goal in the assumption; represents the probability density function of the covariance matrix in the manifold of positive definite matrices; is the sample covariance matrix; for Assume that the sample covariance matrix is The distribution parameters of .

[0058] Step 102: Determine the distance between sample covariance matrices in the Riemann manifold through Riemannian metric, model the distance through Riemann-Gaussian distribution, obtain the statistical distribution characteristics of the sample covariance matrix on the Riemann manifold, and obtain the target detector based on the statistical distribution characteristics.

[0059] In one embodiment of the present application, a Riemann volume element corresponding to the sample covariance matrix is ​​constructed based on the distance, wherein the Riemann volume element is the product of the lengths of the sides of the cube. Based on the Riemann volume element, a probability density function of the Riemann-Gaussian distribution corresponding to the sample covariance matrix is ​​determined, and the statistical distribution characteristics of the sample covariance matrix on the Riemann manifold are determined based on the probability density function of the Riemann-Gaussian distribution.

[0060] In one embodiment of the present application, the distance between sample covariance matrices in a Riemannian manifold is defined based on the Rao-Fisher metric.

[0061] Specifically, the Rao-Fisher metric is an affine invariant metric, based on which the structure of the Riemannian homogeneous space with negative curvature can be obtained. Every point Define its tangent space The inner product of , the Rao-Fisher metric is defined as:

[0062] ;

[0063] in represents the trace of the matrix. The square of the length ds 2 It can be expressed as:

[0064] ;

[0065] Furthermore, a geodesic is a curve with the shortest distance between two points on a manifold. The geodesic distance defines the geometric distance between two points on the manifold. When the Rao-Fisher metric is used, the Riemann distance is obtained. d R for:

[0066] ;

[0067] in Representation Matrix The characteristic value of represents any two symmetric positive definite matrices. Considering that the length can be defined on a Riemann manifold based on the Rao-Fisher metric, it can also be used to define the Riemann volume element. is the product of the lengths of the sides of the cube:

[0068] ;

[0069] in, Represents a square matrix The dimension of .

[0070] In one embodiment of the present application, the statistical distribution of the covariance matrix on the manifold is approximately modeled using the Riemann-Gaussian distribution.

[0071] Specifically, the probability density function of samples in the Riemann manifold can be defined by the Riemann volume element defined above. The mean is , the variance is The probability density function (PDF) of the Riemann-Gaussian distribution is:

[0072] ;

[0073] If the random variable on the symmetric positive definite matrix manifold The mean is , the standard deviation is The Riemann-Gaussian distribution of , then for the linear transformation , with the following properties:

[0074] ;

[0075] According to this transformation property, if ,but , that is, if we can get samples of Riemann-Gaussian distribution with the mean of the unit matrix, we can use linear transformation to get samples of any mean. Therefore, we only need to consider the method of sampling from the Riemann-Gaussian distribution with the mean of the unit matrix. Among them, represents the normalized coefficient of the Riemann-Gaussian distribution, represents a linear transformation, Represents the identity matrix.

[0076] The distribution of eigenvalues ​​and characteristic matrices gives the sample to follow the Riemann-Gaussian distribution Consider the matrix The eigenvalue decomposition of is:

[0077] ;

[0078] in is a diagonal matrix, Representation Matrix Logarithm of eigenvalues, characteristic matrix consisting of eigenvectors is an orthogonal matrix; For the matrix The conjugate transpose of . If the random sample It obeys the Riemann-Gaussian distribution, and its characteristic matrix should be an orthogonal matrix with uniform distribution. At the same time, its eigenvalue logarithm The following joint probability density function needs to be satisfied:

[0079] ;

[0080] in, is a random sample matrix; is a diagonal matrix, Representation Matrix Logarithm of eigenvalue; N is the number of pulses in the slow time dimension; the characteristic matrix composed of eigenvectors is an orthogonal matrix; In the case of Riemann-Gaussian distribution, is a uniformly distributed orthogonal matrix; is the normalization coefficient of the logarithmic distribution of the sample eigenvalues ​​of the Riemann-Gaussian distribution matrix; is the standard deviation of the Riemann-Gaussian distribution; p RGD is the joint probability density function.

[0081] First, consider the distribution of the characteristic matrix of the sample covariance matrix. It should be noted that according to random matrix theory, if the data matrix By independent standard Gaussian distribution If the sample is composed of , then its sample covariance matrix obeys Haar distribution, that is, it satisfies the uniform distribution on the orthogonal matrix. This conclusion shows that the distribution of the feature matrix meets the requirements of Riemann-Gaussian distribution samples.

[0082] Consider the relationship between the sample covariance matrix eigenvalue distribution and the Riemann-Gaussian distribution sample eigenvalue distribution requirements. Riemann-Gaussian distribution standard deviation With sample size The relationship is:

[0083] ;

[0084] For multivariate Gaussian observations, the sample covariance matrix follows the Wishart distribution, and the joint probability density function of its eigenvalues ​​is:

[0085] ;

[0086] in, , represents the normalization coefficient; represents the logarithm of the eigenvalues ​​of the sample covariance matrix; is the sample size; f W is the joint probability density function that conforms to the eigenvalues ​​of the Wishart distribution; x i For the i Observation data; x j For thej Observation data.

[0087] Furthermore, we can get the dimension The joint probability density function of the eigenvalues ​​of the finite sample covariance matrix increases with the number of observed samples. As increases, it will approach the joint probability density function of the Riemann-Gaussian distribution samples, that is:

[0088] ;

[0089] Therefore, as the number of observation samples increases As increases, the sample covariance matrix approaches the Riemann-Gaussian distribution in the symmetric positive definite matrix manifold. p W is the joint probability density function of the logarithm of the eigenvalues ​​of the Wishart distribution matrix; p RGD is the joint probability density function of the logarithm of the eigenvalues ​​of the Riemann-Gaussian distribution matrix.

[0090] Step 103: construct a covariance matrix of clutter based on the acquired sample data of the reference unit, and determine a covariance matrix of the target signal based on the radar slow-time dimension echo data.

[0091] In one embodiment of the present application, a log-likelihood function is constructed based on a plurality of independent samples that are obtained and obey the Riemann-Gaussian distribution, and the coefficients of the log-likelihood function are normalized. Based on the normalized log-likelihood function, a maximum likelihood estimation function of the mean is obtained. The maximum likelihood estimation function of the mean is iterated by a gradient descent algorithm to obtain a corresponding numerical solution, so as to obtain a covariance matrix of the clutter signal based on the corresponding numerical solution. Among them, the reference unit in the embodiment of the present application is a sample unit whose distance from the target is less than a preset distance threshold.

[0092] Specifically, is a Riemann-Gaussian distribution of independent samples, the log-likelihood function can be expressed as:

[0093] ;

[0094] Since the integral remains constant for linear transformations, the normalization coefficient is:

[0095] ;

[0096] It can be seen that the normalization coefficient and the mean It is irrelevant, that is, it can be expressed as:

[0097] ;

[0098] Therefore, the maximum likelihood estimate of the mean is:

[0099] ;

[0100] The numerical solution of this estimate can be calculated by the gradient descent method, that is, it is obtained by the following iteration:

[0101] ;

[0102] in, Indicates the step length.

[0103] In one embodiment of the present application, radar slow time dimension echo data is input into a preset signal power maximum likelihood estimation function to obtain a maximum likelihood value corresponding to the target signal. Based on the maximum likelihood value and a preset covariance matrix function, a covariance matrix of the target signal is obtained.

[0104] Specifically, the covariance matrix of the target signal is estimated :

[0105] ;

[0106] The maximum likelihood estimate of the signal power is It can be obtained by the following formula:

[0107] ;

[0108] in represents the Doppler steering vector of the target; for The conjugate transpose of ; is the inverse matrix of the estimated value of the clutter covariance matrix; For observation data The conjugate transpose of .

[0109] Step 104: Based on the covariance matrix of the clutter, the covariance matrix of the target signal and the target detector, a detection statistic based on the Riemann distance is obtained, the detection statistic is compared with the detection threshold, and the detection target is determined based on the comparison result.

[0110] In one embodiment of the present application, in a symmetric positive definite matrix manifold, the binary hypothesis testing problem based on the Riemann-Gaussian distribution is:

[0111] ;

[0112] in express Assume that the standard deviation of the Riemann-Gaussian distribution is . The likelihood ratio detector based on the Neyman-Pearson criterion is:

[0113] ;

[0114] in, It means that the assumption is without a goal; It means that there is a goal in the assumption; Represents the probability density function of the covariance matrix in the manifold of positive definite matrices when there is a target; represents the probability density function of the covariance matrix in the manifold of positive definite matrices when there is no target; is the sample covariance matrix; for Assume that the sample covariance matrix is The distribution parameters of for Assume that the sample covariance matrix is The distribution parameters of is the detection threshold.

[0115] Furthermore, the covariance matrix of the clutter and the covariance matrix of the target signal are input into the target detector, and the detection statistic based on the Riemann distance is obtained as follows:

[0116] ;

[0117] in, is the sample covariance matrix; is the covariance matrix of the clutter; is the covariance matrix of the target signal; It means that the assumption is without a goal; It means that there is a goal in the assumption; represents the Riemann distance, Indicates the detection threshold.

[0118] In practical applications, this threshold can be adjusted based on the false alarm rate requirements using the collected data in the clutter scene. Assume that the samples are obtained by Monte Carlo experiment statistics.

[0119] In this embodiment, the performance of the proposed detection method is verified by using the measured meteorological clutter data collected by the Ku-band high-resolution phased array radar and the sea clutter data in the IPIX Grimsby dataset. The Ku-band high-resolution phased array radar uses stepped frequency synthesis to obtain high-resolution range images with a synthesis bandwidth of 1 GHz. The pulse repetition interval of the stepped frequency frame is The detection range is 450~2100m. The IPIX Grimsby dataset was collected by the shore-based X-band IPIX radar system in the winter of 1998. The radar system has a carrier frequency of 9.39GHz, a pulse repetition frequency of 1kHz, and a grazing angle of 0.33 . Figure 2 A schematic diagram of specific information of measured meteorological clutter and target data collected by a Ku-band high-resolution phased array radar provided in an embodiment of the present application, such as Figure 2 As shown, different numbers correspond to different collection dates, and their weather, pitch angles, azimuth angles and data descriptions are all different.

[0120] Figure 3 A schematic diagram of the marginal probability density function of the logarithm of the eigenvalue of a sample covariance matrix provided in an embodiment of the present application, wherein (a) is the eigenvalue r 1. Schematic diagram of marginal probability density function with 16 observed samples, (b) is the characteristic value r 2. Schematic diagram of marginal probability density function with 16 observed samples, (c) is the characteristic value r 1. Schematic diagram of marginal probability density function with 64 observed samples, (d) is the characteristic value r 2. Schematic diagram of marginal probability density function with 64 observed samples; Figure 3 As shown, Take the example to verify the statistical distribution characteristics of the sample covariance matrix. Figure 3 The marginal probability density functions of the two eigenvalues ​​are given in the figure, and the number of observed samples is set to and . Compare the empirical probability density function obtained from the simulation sample data statistics with the theoretical probability density function obtained from the formula, such as Figure 3 As shown, with As increases, the marginal distribution obtained from the simulation data will gradually approach the theoretical distribution. When is large, the eigenvalue distribution of the sample covariance matrix will be close to the eigenvalue distribution of the Riemann-Gaussian distribution sample.

[0121] Figure 4 Schematic diagram of detection probability under different signal-to-noise ratios for an optimal detector based on the Neyman-Pearson criterion and a likelihood ratio detection algorithm based on the Riemann-Gaussian distribution provided in an embodiment of the present application, wherein (a) is a schematic diagram for an observation sample number of 9, (b) is a schematic diagram for an observation sample number of 18, (c) is a schematic diagram for an observation sample number of 45, and (d) is a schematic diagram for an observation sample number of 90; Figure 4 As shown, Figure 4 The detection probability of the likelihood ratio detector based on the Riemann-Gaussian distribution and the optimal detector based on the Neyman-Pearson criterion under different signal-to-clutter ratios (SCR) is compared in [1]. The dimension of the sample covariance matrix is , the number of observation samples is given by Increase to The false alarm rate is set to , the corresponding detection threshold is determined by Monte Carlo simulation. Figure 4 As shown, with As increases, the performance of the likelihood ratio detector based on the Riemann-Gaussian distribution gradually approaches the theoretical optimal detection performance.

[0122] In this embodiment, measured meteorological clutter data and sea clutter data are used to compare the detector based on Riemann-Gaussian distribution proposed in the embodiment of the present application with the existing detector based on total Bregman divergence (TBD), the detector based on Bures-Wasserstein distance, the detector based on KL divergence and the traditional generalized likelihood ratio detector.

[0123] Figure 5 A schematic diagram for comparing the detection probability of different signal-to-clutter ratios of meteorological clutter data with that of existing methods provided in an embodiment of the present application, wherein (a) is a schematic diagram for comparing data set 1, (b) is a schematic diagram for comparing data set 2, and (c) is a schematic diagram for comparing data set 3. Figure 5 The detection probability of the comparison method under different signal-to-clutter ratios is given in the figure. In the measured clutter data, the false alarm rate is set to .like Figure 5 As shown in the figure, the detection algorithm based on Riemann-Gaussian distribution can effectively improve the performance of moving target detection in the background of meteorological clutter.

[0124] Figure 6 A schematic diagram for comparing the detection probabilities of the existing methods under different signal-to-clutter ratios based on public shore-based sea clutter data is provided in an embodiment of the present application, wherein Figure (a) is a schematic diagram for comparing the first data set, and Figure (b) is a schematic diagram for comparing the second data set. Figure 6 The detection probabilities of the proposed method and the comparative method under different signal-to-clutter ratios in the measured sea clutter data are given in Fig. and The false alarm rate is , the number of reference units is .like Figure 6 As shown, the detector in the embodiment of the present application can effectively improve the target detection performance in the sea clutter background.

[0125] Figure 7 The schematic diagram of the structure of a target detection device under a non-uniform clutter background provided by an embodiment of the present application. Figure 7As shown, a target detection device 200 under a non-uniform clutter background includes: at least one processor 201; and a memory 202 that is communicatively connected to the at least one processor 201; wherein the memory 202 stores instructions that can be executed by the at least one processor 201, and the instructions are executed by the at least one processor 201, so that the at least one processor 201 can: construct a sample covariance matrix based on the acquired radar slow-time echo data, map the sample covariance matrix to a Riemann manifold, so as to establish a binary hypothesis test problem based on the sample covariance matrix in the Riemann manifold; determine the sample covariance matrix in the Riemann manifold through the Riemann metric. Distance, based on the distance, modeling is performed through Riemann-Gaussian distribution to obtain the statistical distribution characteristics of the sample covariance matrix on the Riemann manifold, so as to obtain the target detector based on the statistical distribution characteristics; based on the sample data of the reference unit obtained, the covariance matrix of the clutter is constructed, and the covariance matrix of the target signal is determined based on the radar slow-time echo data; wherein, the reference unit is a sample unit whose distance to the target is less than a preset distance threshold; based on the covariance matrix of the clutter, the covariance matrix of the target signal and the target detector, a detection statistic based on the Riemann distance is obtained, the detection statistic is compared with the detection threshold, and the detection target is determined based on the comparison result.

[0126] A non-volatile computer storage medium provided by an embodiment of the present application stores computer executable instructions, wherein the computer executable instructions are configured to: construct a sample covariance matrix based on acquired radar slow-time dimension echo data, map the sample covariance matrix to a Riemann manifold, so as to establish a binary hypothesis test problem based on the sample covariance matrix in the Riemann manifold; determine the distance between the sample covariance matrices in the Riemann manifold by Riemann metric, model the sample covariance matrix on the Riemann manifold based on the distance by Riemann Gaussian distribution, obtain the statistical distribution characteristics of the sample covariance matrix on the Riemann manifold, and obtain a target detector based on the statistical distribution characteristics; construct a clutter covariance matrix based on acquired reference unit sample data, and determine a target signal covariance matrix based on radar slow-time dimension echo data; wherein the reference unit is a sample unit whose distance from the target is less than a preset distance threshold; obtain a detection statistic based on the Riemann distance based on the clutter covariance matrix, the target signal covariance matrix and the target detector, compare the detection statistic with the detection threshold, and determine the detection target based on the comparison result.

[0127] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0128] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the embodiments of the present application may have various changes and variations. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A method for target detection under non-uniform clutter background, characterized in that: The method comprises: A sample covariance matrix is ​​constructed based on the acquired radar slow-time echo data, and the sample covariance matrix is ​​mapped to a Riemann manifold to establish a binary hypothesis testing problem based on the sample covariance matrix in the Riemann manifold; Determine the distance between the sample covariance matrices in the Riemann manifold by Riemannian metric, perform modeling based on the distance by Riemann-Gaussian distribution, obtain the statistical distribution characteristics of the sample covariance matrix on the Riemann manifold, and obtain the target detector based on the statistical distribution characteristics; Based on the sample data of the reference unit obtained, a covariance matrix of clutter is constructed, and based on the radar slow time dimension echo data, a covariance matrix of the target signal is determined; wherein the reference unit is a sample unit whose distance to the target is less than a preset distance threshold; Based on the covariance matrix of the clutter, the covariance matrix of the target signal and the target detector, a detection statistic based on the Riemann distance is obtained, the detection statistic is compared with a detection threshold, and a detection target is determined based on the comparison result; The modeling is performed based on the distance through the Riemann-Gaussian distribution to obtain the statistical distribution characteristics of the sample covariance matrix on the Riemann manifold, specifically including: Constructing a Riemann volume element corresponding to the sample covariance matrix based on the distance; wherein the Riemann volume element is the product of the lengths of the sides of a cube; Based on the Riemann volume element, the probability density function of the Riemann-Gaussian distribution corresponding to the sample covariance matrix is ​​determined, so as to determine the statistical distribution characteristics of the sample covariance matrix on the Riemann manifold based on the probability density function of the Riemann-Gaussian distribution.

2. The target detection method under non-uniform clutter background according to claim 1, characterized in that: The binary hypothesis testing problem based on the sample covariance matrix is ​​established in the Riemann manifold, specifically including: Function-based: ； Formulate binary hypothesis testing problems on Riemannian manifolds; in; It means that the assumption is without a goal; It means that there is a goal in the assumption; represents the probability density function of the covariance matrix in the manifold of positive definite matrices; is the sample covariance matrix; for Assume that the sample covariance matrix is The distribution parameters of .

3. The target detection method under non-uniform clutter background according to claim 1, characterized in that: After determining the probability density function of the Riemann-Gaussian distribution corresponding to the sample covariance matrix based on the Riemann volume element, the method further includes: Through the distribution of eigenvalues ​​and characteristic matrices, the conditions for samples to obey the Riemann-Gaussian distribution are determined; The eigenvalue decomposition of the matrix is: ； The eigenvalues ​​satisfy the following joint probability density function: ； in, is a random sample matrix; is a diagonal matrix, Representation Matrix Logarithm of eigenvalue; N is the number of pulses in the slow time dimension; the characteristic matrix composed of eigenvectors is an orthogonal matrix; In the case of Riemann-Gaussian distribution, is a uniformly distributed orthogonal matrix; is the normalization coefficient of the logarithmic distribution of the sample eigenvalues ​​of the Riemann-Gaussian distribution matrix; is the standard deviation of the Riemann-Gaussian distribution; p RGD is the joint probability density function; The relationship between the standard deviation of the Riemann-Gaussian distribution and the number of samples is determined as: ； in, is the standard deviation of the Riemann-Gaussian distribution; is the sample size; When the sample covariance matrix corresponding to the multivariate Gaussian observations conforms to the Wishart distribution, the joint probability density function of the eigenvalues ​​conforming to the Wishart distribution is: ； in, , represents the normalization coefficient; represents the logarithm of the eigenvalues ​​of the sample covariance matrix; is the sample size; f W is the joint probability density function that conforms to the eigenvalues ​​of the Wishart distribution; x i For the i Observation data; x j For the j Observation data; Based on the joint probability density function of the eigenvalues ​​conforming to the Wishart distribution, it is determined that the sample covariance matrix approaches a Riemann-Gaussian distribution in a symmetric positive definite matrix manifold.

4. The target detection method under non-uniform clutter background according to claim 1, characterized in that: The obtaining of the target detector based on the statistical distribution characteristics specifically includes: The target detector is: ； in, It means that the assumption is without a goal; It means that there is a goal in the assumption; Represents the probability density function of the covariance matrix in the manifold of positive definite matrices when there is a target; represents the probability density function of the covariance matrix in the manifold of positive definite matrices when there is no target; is the sample covariance matrix; for Assume that the sample covariance matrix is The distribution parameters of for Assume that the sample covariance matrix is The distribution parameters of is the detection threshold.

5. The method for target detection under non-uniform clutter background according to claim 1, characterized in that: The step of constructing a covariance matrix of clutter based on the acquired sample data of the reference unit specifically includes: Based on the obtained multiple independent samples that obey the Riemann-Gaussian distribution, a log-likelihood function is constructed; Performing coefficient normalization processing on the log-likelihood function; Based on the normalized log-likelihood function, a maximum likelihood estimation function of the mean is obtained; The maximum likelihood estimation function of the mean is iterated by a gradient descent algorithm to obtain a corresponding numerical solution, so as to obtain a covariance matrix of the clutter signal based on the corresponding numerical solution.

6. The method for target detection under non-uniform clutter background according to claim 1, characterized in that: Determining the covariance matrix of the target signal based on the radar slow-time echo data specifically includes: Inputting the radar slow-time dimension echo data into a preset signal power maximum likelihood estimation function to obtain a maximum likelihood value corresponding to the target signal; Based on the maximum likelihood value and a preset covariance matrix function, a covariance matrix of the target signal is obtained.

7. The method for target detection under non-uniform clutter background according to claim 1, characterized in that: The method of obtaining a detection statistic based on the Riemann distance based on the covariance matrix of the clutter, the covariance matrix of the target signal and the target detector specifically includes: Inputting the covariance matrix of the clutter and the covariance matrix of the target signal into the target detector to obtain a detection statistic based on the Riemann distance; Wherein, the detection statistic is: ； in, is the sample covariance matrix; is the covariance matrix of the clutter; is the covariance matrix of the target signal; It means that the assumption is without a goal; It means that there is a goal in the assumption; represents the Riemann distance, Indicates the detection threshold.

8. A target detection device under non-uniform clutter background, characterized in that: The device comprises a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the device is triggered to execute the method according to any one of claims 1 to 7.

9. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions can execute the method according to any one of claims 1 to 7.

Images