Knowledge-aided broadband radar target detector and design method in non-uniform clutter
By dividing broadband radar signals and clutter signals into subbands, establishing auxiliary data models and building knowledge auxiliary detectors, the problem of broadband radar target detection in non-uniform clutter environments is solved, and more accurate target detection and performance are achieved promote.
Patent Information
- Application Number
- CN202210103315.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-01-27
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2042-01-27
AI Technical Summary
The prior art is difficult to effectively detect broadband radar targets in non-uniform cluttered environments, especially in the case of distance migration. The existing methods cannot effectively deal with the target detection problem of broadband radars in the context of clutter.
By dividing the broadband radar target signal and clutter signal into several subbands, a clutter model of auxiliary data of each subband is established, and the clutter probability density function and covariance matrix relationship model of each subband is obtained, combining the posterior probability density Function construction knowledge-assisted detectors, including maximum posterior ratio detectors, Rao detectors and Wald detectors, are used for broadband radar target detection in non-uniform clutter environments.
In a non-uniform clutter environment, the accuracy and performance of broadband radar target detection is improved, and the target detection in distance migration is effectively handled, which improves the performance of the detector.
Smart Images

Figure CN114609594B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of radar technology, and in particular relates to a knowledge-assisted broadband radar target detector in non-uniform clutter and a design method thereof. Background Art
[0002] Broadband radar has the advantages of strong anti-interference ability, high range resolution, and high measurement accuracy. It is an important direction for the development of modern radar. Broadband radar is widely used in synthetic aperture and inverse synthetic aperture radar imaging, radar high-precision measurement, and radar target recognition. Detection is the basic function of radar, and so is broadband radar. Broadband radar has unique advantages in target detection in a clutter environment because clutter is composed of a large number of randomly distributed scattering centers. As the radar range resolution increases, the number of scattering centers in each range unit decreases proportionally, and the clutter power also decreases. However, for artificial targets, such as missiles, aircraft, vehicles, and ships, they are composed of a small number of sparse scattering centers. The improvement of range resolution does not reduce the echo power of strong scatterers, but the signal-to-clutter ratio (SCR) of the target echo signal is also improved.
[0003] Due to the improvement of range resolution, the target echo of broadband radar will be distributed in several adjacent range cells, forming range-extended targets and range-distributed targets. Most of the existing research focuses on the detection of range-extended targets.
[0004] In modern radar target detection, coherent processing technology is usually used. Doppler analysis based on coherent multi-pulses can detect moving targets from stationary clutter. For narrowband radars, due to the relatively large size of the range cell, the range change caused by target motion is usually less than one range cell within the coherent processing interval (CPI). Therefore, Doppler processing can be performed directly on multiple echoes of a range cell. This method is also used in most range-extended target detectors. For wideband radars, due to the small range resolution cell, the range change caused by target motion may span multiple range cells within the coherent processing interval, which is called range walk or range offset. In this case, the range offset of the target needs to be corrected first, so that the target returns from different pulse ranges, and then Doppler processing is performed. Keystone transform is a common linear range migration correction algorithm, which has been applied to radar imaging and high-speed target detection as well as wideband radar target feature extraction. However, there are few related literatures on wideband radar target detection considering range migration in the background of clutter. Among them, in one prior art, the problem of range-extended target with range migration detection in composite Gaussian clutter is solved, and the problem is implemented in the range-frequency domain; in another prior art, range-extended and range-Doppler-extended targets with range migration detection in partially uniform clutter are designed respectively; and in yet another prior art, a standard linear model of a broadband radar target with range migration is derived, and a knowledge-assisted detector without auxiliary data is designed.
[0005] The performance of radar target detectors depends mainly on the power and clutter covariance matrix (CCM) of the clutter. In a homogeneous environment, the power and covariance matrix of the clutter are estimated from auxiliary data, which are assumed to contain no target echoes and have the same statistical distribution as the clutter of the unit under test (CUT) or the original data containing both target and clutter. Unfortunately, the homogeneity assumption does not hold in many practical situations, especially for wideband radars. This non-homogeneity means that the power and covariance matrix of the clutter in different range bins are different. The non-homogeneity of wideband radar clutter is caused by the relatively small number of scatterers in a range bin.
[0006] Therefore, how to achieve detection of broadband radar targets in non-uniform clutter environments has become a technical problem that needs to be solved urgently. Summary of the invention
[0007] In order to solve the above problems existing in the prior art, the present invention provides a knowledge-assisted broadband radar target detector in non-uniform clutter and a design method. The technical problem to be solved by the present invention is achieved by the following technical solutions:
[0008] The embodiment of the present invention provides a method for designing a knowledge-assisted broadband radar target detector in non-uniform clutter, comprising the steps of:
[0009] S1. Divide the broadband radar target signal into several sub-bands to obtain each pulse signal of each sub-band during the coherent processing interval;
[0010] S2, dividing the broadband radar non-uniform clutter signal into several sub-bands, and modeling the clutter covariance matrix as a random matrix to obtain the clutter of the auxiliary data in each sub-band, wherein the clutter of the auxiliary data obeys a composite Gaussian distribution;
[0011] S3, obtaining the probability density function of the clutter of the auxiliary data in each sub-band, and establishing a clutter covariance matrix relationship model and a clutter power relationship model between the original data and the auxiliary data;
[0012] S4, establishing a broadband radar target detection problem model according to each pulse signal and the clutter of the auxiliary data;
[0013] S5. Combining the broadband radar target detection problem model, the clutter of the auxiliary data, the probability density function, the clutter covariance matrix relationship model and the clutter power relationship model, a knowledge-assisted detector is constructed using a posterior probability density function.
[0014] In one embodiment of the present invention, step S1 comprises:
[0015] S11, using a multi-pulse coherent processing model to process the broadband radar target echo to obtain m discrete baseband echo signals after broadband radar pulse compression:
[0016]
[0017] Where L represents the adjacent distance unit, represents the complex amplitude of the target echo in the lth distance unit, B represents the broadband radar transmission bandwidth, k represents the detection window distance unit, T s represents the coherent processing interval, ΔR represents the size of the range unit, c represents the speed of light, v represents the radial velocity, m represents the number of broadband radars, and T r represents the pulse repetition interval, j represents the complex unit, and f c represents the signal frequency, K represents the length of the detection window, and M represents the number of pulses during the coherent processing interval;
[0018] S12, dividing the discrete baseband echo signal into several sub-bands by a discrete Fourier transform sub-band filter bank, and obtaining each pulse signal of each sub-band and during the coherent processing interval:
[0019]
[0020] Where n represents the nth subband, t represents the tth pulse, Represents the complex amplitude of the target echo in the range unit after being divided into N sub-bands, represents the complex amplitude of the target echo in the lth range unit, N represents the number of sub-bands, Sr represents the discrete baseband echo signal after broadband radar pulse compression, k represents the detection window range unit, j represents the complex unit, and f c represents the signal frequency, c represents the speed of light, v represents the radial velocity, m represents the number of broadband radars, T r represents the pulse repetition interval, B represents the wideband transmitted by the wideband radar;
[0021] S13, performing vector processing on each pulse signal of each sub-band during the coherent processing interval to obtain a target benefit of each sub-band:
[0022] s n =H n a
[0023] in, p w represents the Doppler steering vector at the wth frequency point, a represents a variable, q n Represents a variable.
[0024] In one embodiment of the present invention, step S2 comprises:
[0025] S21, dividing the broadband radar non-uniform clutter signal into a plurality of sub-bands by using a discrete Fourier transform sub-band filter bank;
[0026] S22. Modeling the clutter covariance matrix as a random matrix in a Bayesian framework to obtain clutter of the auxiliary data in each subband, wherein the clutter of the auxiliary data obeys a composite Gaussian distribution:
[0027]
[0028] Among them, c nj represents the clutter of the auxiliary data in the nth subband, j=1,…,J, M represents M pulses, J represents J auxiliary data units in each subband, τ nj is a non-negative random variable representing the clutter power, Represented by the covariance matrix R nj is a complex Gaussian vector of unit variance conditional on .
[0029] In one embodiment of the present invention, step S3 includes:
[0030] S31, obtaining the probability density function of the clutter of the auxiliary data in each sub-band with the non-negative random variable of the clutter power and the covariance matrix as conditions:
[0031]
[0032] Among them, c nj represents the clutter of the auxiliary data in the nth subband, τ nj A non-negative random variable representing the clutter power, R nj represents the covariance matrix, M represents M pulses, J represents J auxiliary data units in each subband, and N represents the number of subbands;
[0033] S32, based on the fact that the clutter covariance matrix of the original data is the average value of the clutter covariance matrix of the auxiliary data, a probability density function of the covariance matrix of the auxiliary data and a probability density function of the covariance matrix of the original data are established to form the clutter covariance matrix relationship model, and a probability density function of the clutter power of the auxiliary data and a probability density function of the clutter power of the original data are established to form the clutter power relationship model, wherein,
[0034] The probability density function of the covariance matrix of the auxiliary data is:
[0035]
[0036] Among them, ν nj >M+1 represents the degree of freedom, M represents the number of pulses during the coherent processing interval, and R nj represents the auxiliary data clutter covariance matrix, R n0 represents the original data clutter covariance matrix, N represents the number of subbands, n represents the nth subband, and J represents that each subband has J auxiliary data units;
[0037] The probability density function of the covariance matrix of the original data is:
[0038]
[0039] Among them, n >M+1 represents the degree of freedom, Σ n represents the precision matrix, R n0 represents the raw data clutter covariance matrix, M represents the number of pulses during the coherent processing interval;
[0040] The probability density function of the auxiliary data clutter power is:
[0041]
[0042] Among them, τ nj A non-negative random variable representing the auxiliary data clutter power, τ n0 A non-negative random variable representing the original data clutter power, q njrepresents the shape parameter, N represents the number of subbands, n represents the nth subband, and J represents that each subband has J auxiliary data units;
[0043] The probability density function of the original data clutter power is:
[0044]
[0045] Among them, τ n0 A non-negative random variable representing the original data clutter power, β n represents the scale parameter of the inverse gamma distribution, p n Represents the shape parameter.
[0046] In one embodiment of the present invention, the broadband radar target detection problem model is:
[0047]
[0048] Among them, H1 represents the situation including target signal and clutter signal, x n represents the measured data vector of the original data in the nth subband, p w represents the Doppler steering vector at the wth frequency point, a represents a variable, q n Represents a variable, c n0 represents the clutter of the unit under test in the nth subband, z nj represents the auxiliary data vector of the nth subband, c nj represents the noise of the auxiliary data in the nth subband, n represents the nth subband, N represents the number of subbands, J represents that each subband has J auxiliary data units, and H0 represents the case where the noise signal is included.
[0049] In one embodiment of the present invention, the knowledge-assisted detector includes one or more of a knowledge-assisted maximum a posteriori ratio detector, a knowledge-assisted Rao detector, and a knowledge-assisted Wald detector.
[0050] In one embodiment of the present invention, when the knowledge-assisted detector is the knowledge-assisted maximum a posteriori ratio detector, step S5 comprises:
[0051] S51, combining the broadband radar target detection problem model, the clutter of the auxiliary data, the probability density function, the clutter covariance matrix relationship model and the clutter power relationship model, based on the maximum a posteriori ratio test framework of the radar system, constructing the knowledge-assisted maximum a posteriori ratio detector:
[0052]
[0053] Among them, R n0represents the original data clutter covariance matrix, τ n0 represents the non-negative random variable of the original data clutter power, a represents the variable, H1 represents the case including the target signal and the clutter signal, x n represents the measured data vector of the original data in the nth subband, z n ( 1,…,J ) represents the auxiliary data vector of the nth subband, and H0 represents the case where a clutter signal is included;
[0054] S52, obtain parameter τ based on prior knowledge and auxiliary data n0 and R n0 The maximum a posteriori probability ratio estimate is:
[0055]
[0056] in, Represents parameter R n0 The maximum a posteriori probability estimate of Represents the parameter τ n0 The maximum a posteriori probability estimate, R n0 represents the original data clutter covariance matrix, τ n0 A non-negative random variable representing the original data clutter power, z n(1,…,J) represents the auxiliary data vector of the nth subband, n represents the nth subband, and N represents the number of subbands;
[0057] According to the parameter τ n0 and R n0 The maximum a posteriori probability ratio estimate computes the estimate of the parameter a:
[0058]
[0059] in, represents the estimated value of parameter a, p w represents the Doppler steering vector at the wth frequency point, q n represents a variable, x n A measured data vector representing the original data in the nth subband;
[0060] S53, in the case of including target signal and clutter signal, by using To replace the method of a, we get the parameter τ n0 and R n0 The estimated value of .
[0061] In one embodiment of the present invention, the knowledge-assisted Rao detector is:
[0062]
[0063] in, p w represents the Doppler steering vector at the wth frequency point, q n Represents a variable, It represents the parameter τ when the auxiliary data and the measured data in the unit under test are given. n0 The maximum a posteriori probability estimate of It represents the parameter R under given auxiliary data and measured data in the unit under test. n0 The maximum a posteriori probability estimate, x n represents the measured data vector of the original data in the nth subband, n represents the nth subband, and N represents the number of subbands.
[0064] In one embodiment of the present invention, the knowledge-assisted Wald detector is:
[0065]
[0066] in, p w represents the Doppler steering vector at the wth frequency point, q n Represents a variable, It represents the parameter τ when the auxiliary data and the measured data in the unit under test are given. n0 The maximum a posteriori probability estimate of It represents the parameter R under given auxiliary data and measured data in the unit under test. n0 The maximum a posteriori probability estimate, x n represents the measured data vector of the original data in the nth subband, n represents the nth subband, and N represents the number of subbands.
[0067] Another embodiment of the present invention provides a knowledge-assisted broadband radar target detector in non-uniform clutter, which is designed by the design method described in the above embodiment.
[0068] Compared with the prior art, the present invention has the following beneficial effects:
[0069] The design method of this embodiment is based on the clutter of each pulse signal in each sub-band and the auxiliary data in each sub-band during the coherent processing interval, and proposes a knowledge-assisted detector for broadband radar target detection in a non-uniform clutter environment, which solves the problem of broadband radar target detection in a non-uniform clutter environment. The detector is effective for broadband radar target detection in a clutter background, and in certain scenarios, the performance of the detector is improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0070] Figure 1 A schematic flow chart of a method for designing a knowledge-assisted broadband radar target detector in non-uniform clutter provided by an embodiment of the present invention;
[0071] Figure 2 A schematic diagram of the relationship between the detection probability and SCR of a detector in an environment with different degrees of clutter heterogeneity provided by an embodiment of the present invention;
[0072] Figure 3 A schematic diagram of the relationship between the detection probability and SCR of a detector in an environment with different accuracies of clutter prior knowledge provided by an embodiment of the present invention;
[0073] Figure 4 A schematic diagram of the relationship between the detection probability of a detector and the SCR under different heterogeneous conditions and the number of auxiliary units provided by an embodiment of the present invention;
[0074] Figure 5 A probability schematic diagram of a detection curve of a detector when parameters of a prior distribution do not match provided by an embodiment of the present invention;
[0075] Figure 6 A schematic diagram of a SAR image for generating clutter provided by an embodiment of the present invention;
[0076] Figure 7 A schematic diagram of a detector detection probability curve in the case of synthetic clutter in a SAR image provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0077] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.
[0078] Embodiment 1
[0079] This embodiment uses a knowledge-assisted (KA) method to offset the impact of non-uniformity on detector performance. The knowledge-assisted (KA) method is widely used by utilizing the environmental prior knowledge from the remote sensing information of the region of interest or the previous scanning data. Most knowledge-assisted (KA) methods are based on Bayes' theorem, in which the clutter power and the clutter covariance matrix are not assumed to be deterministic like the classical method, but are regarded as random variables and random matrices, whose prior distribution integrates the prior knowledge of the environment.
[0080] Based on this, this embodiment describes the design of a knowledge-assisted detector for range-migrating radar targets under the background of wideband radar clutter. The design idea of the knowledge-assisted detector in this embodiment is as follows: First, the target echo and clutter of the wideband radar are divided into several sub-bands, and the target range migration in the model of the target echo within a coherent processing interval is considered at the same time. Second, the target high-resolution range image (HRRP) is used as the target signal to be detected. Since the energy distribution of the target echo is small, the performance loss of the detector designed by this method is small. Third, the clutter power of the original data and the auxiliary data are both inverse gamma distribution, the expected value of the auxiliary data clutter power is assumed to be the original data clutter power, and the expected value of the original data clutter power is assumed to be a priori value; the clutter covariance matrix of the original data and the auxiliary data is assumed to follow a complex inverse Wishart distribution, the auxiliary data is the clutter covariance matrix of the original data, and the precision matrix of the original data is assumed to be a priori value. Fourth, based on the above theory, a knowledge-assisted maximum a posteriori ratio detector, a knowledge-assisted Wald detector, and a knowledge-assisted Rao detector are designed using a two-step method for wideband radar target detection in non-uniform clutter.
[0081] See also Figure 1 , Figure 1 A flow chart of a method for designing a knowledge-assisted wideband radar target detector in non-uniform clutter provided by an embodiment of the present invention. This embodiment is described by taking a knowledge-assisted maximum a posteriori ratio detector as an example. The design method includes the following steps:
[0082] S1. Divide the broadband radar target signal into several sub-bands, obtain each pulse signal of each sub-band and during the coherent processing interval, so as to establish a signal model of the broadband radar target.
[0083] S11. Processing the broadband radar target echo using a multi-pulse coherent processing model to obtain m discrete baseband echo signals compressed by broadband radar pulses.
[0084] Specifically, considering that the broadband radar transmission bandwidth is B, and the range point spread function (PSF) is a sinc function, the width of the main lobe of the range point spread function (PSF), that is, the size of the range unit is ΔR = c / (2B) and is related to the speed of light; in order to suppress clutter and integrate the energy of the target echo, the broadband radar adopts a multi-pulse coherent processing model. Assume that the pulse repetition interval (PRI) is T r , and there are M pulses during the coherent processing interval (CPI); the target is expanded into L adjacent range cells and moves at a radial velocity V, and the discrete baseband echo signals after m broadband radar pulse compression are obtained as follows:
[0085]
[0086] Where L represents the adjacent distance unit, represents the complex amplitude of the target echo in the lth distance unit, B represents the broadband radar transmission bandwidth, k represents the detection window distance unit, T s represents the coherent processing interval, ΔR represents the size of the range unit, c represents the speed of light, v represents the radial velocity, m represents the number of broadband radars, and T r represents the pulse repetition interval, j represents the complex unit, and f c represents the signal frequency, K represents the length of the detection window, and M represents the number of pulses during the coherent processing interval.
[0087] From the signal model in (1), it can be seen that the echo of a wideband radar is different from that of a narrowband radar because the range migration of the target echo envelope during the coherent processing interval spans multiple range cells. In (1), the target moves vmT during the coherent processing interval. r / ΔR distance cells. Then the length of the range window should satisfy K ≥ K moving +L to include complete target echo data.
[0088] S12. Divide the discrete baseband echo signal into several subbands through a discrete Fourier transform subband filter bank to obtain each pulse signal of each subband during a coherent processing interval.
[0089] Specifically, it is difficult to perform range migration directly in the range domain. Therefore, for the wideband radar target model, processing range migration in the range-frequency domain can consider dividing the wideband radar target echo signal into several narrowband signals without performing range migration. Using this method, it is assumed that the wideband radar target echo signal is divided into N subbands by a discrete Fourier transform (DFT) subband filter bank. According to the frequency domain sampling theorem, N should satisfy N ≥ K and the nth subband and the tth pulse during the coherent processing interval can be expressed as:
[0090]
[0091] Where n represents the nth subband, t represents the tth pulse, Represents the complex amplitude of the target echo in the range unit after being divided into N sub-bands, represents the complex amplitude of the target echo in the lth range unit, N represents the number of sub-bands, Sr represents the discrete baseband echo signal after broadband radar pulse compression, k represents the detection window range unit, j represents the complex unit, and f c represents the signal frequency, c represents the speed of light, v represents the radial velocity, m represents the number of broadband radars, T r represents the pulse repetition interval, and B represents the wideband radar outgoing bandwidth.
[0092] Define the following two variables:
[0093] a=[a0 a1 …a L-1 ] T (3)
[0094]
[0095] Therefore, s(n,m) in formula (2) can be rewritten as follows:
[0096]
[0097] The Doppler steering vector p at the wth frequency point w for:
[0098] p w =[exp(-jω w 0),exp(-jω w 1),…,exp(-jω w (M-1))] T (6)
[0099] Among them, ω n =4πvT r (f c +nB / N) / c.
[0100] According to the above definition, the target return of the nth sub-band can be expressed as the following linear model:
[0101] s n =H n a (7)
[0102] in, p w represents the Doppler steering vector at the wth frequency point, a represents a variable, q n Represents a variable.
[0103] S2. Divide the broadband radar non-uniform clutter signal into several sub-bands, and model the clutter covariance matrix as a random matrix, obtain the clutter of the auxiliary data in each sub-band to establish a non-uniform clutter model, wherein the clutter of the auxiliary data obeys a composite Gaussian distribution.
[0104] S21. Divide the broadband radar non-uniform clutter signal into a plurality of sub-bands by using a discrete Fourier transform sub-band filter bank.
[0105] Specifically, since the target echo is divided into N sub-bands by discrete Fourier transform, the clutter is also processed in the same way. When the gap between two adjacent sub-bands is not less than the bandwidth of a sub-band, the clutter is basically unrelated to each other. For the model of this embodiment, when the number of sub-bands N is equal to the length of the detection window K, the above conditions remain unchanged.
[0106] S22. Model the clutter covariance matrix as a random matrix in a Bayesian framework to obtain the clutter of the auxiliary data in each subband.
[0107] Specifically, the data in the detection window is called the original data, and the data assumed to contain only clutter and used to estimate the clutter covariance matrix (CCM) is called the auxiliary data. In heterogeneous scenarios, the power and structure of the clutter covariance matrix (CCM) of different distance units are different from each other, and in homogeneous and non-homogeneous scenarios, the clutter covariance matrix (CCM) of the original data cannot be directly estimated with the auxiliary data using the sample covariance matrix (SCM). To overcome this problem, the clutter covariance matrix is modeled as a random matrix in a Bayesian framework.
[0108] Assume that the clutter in the unit under test (CUT) in the nth subband is Each subband has J auxiliary data units, and the clutter of the auxiliary data in the nth subband is j=1,…,J, the clutters are independent of each other, then the clutter of the auxiliary data obeys the composite Gaussian distribution expressed as:
[0109]
[0110] Among them, c nj represents the clutter of the auxiliary data in the nth subband, j=1,…,J, M represents M pulses, J represents J auxiliary data units in each subband, τ nj is a non-negative random variable representing the clutter power, Represented by the covariance matrix R nj is a complex Gaussian vector of unit variance conditional on .
[0111] S3. Obtain the probability density function of the clutter of the auxiliary data in each sub-band, and establish a clutter covariance matrix relationship model and a clutter power relationship model between the original data and the auxiliary data.
[0112] S31. Obtain a probability density function of clutter of the auxiliary data in each sub-band with a non-negative random variable of clutter power and a covariance matrix as conditions.
[0113] Specifically, the non-negative random variable τ of the clutter power nj and the covariance matrix R njThe probability density function of the clutter conditional on is:
[0114]
[0115] Among them, c nj represents the clutter of the auxiliary data in the nth subband, τ nj A non-negative random variable representing the clutter power, R nj represents the covariance matrix, M represents M pulses, J represents that each subband has J auxiliary data units, and N represents the number of subbands.
[0116] S32, based on the fact that the clutter covariance matrix of the original data is the average value of the clutter covariance matrix of the auxiliary data, a probability density function of the covariance matrix of the auxiliary data and a probability density function of the covariance matrix of the original data are established to form the clutter covariance matrix relationship model, and a probability density function of the clutter power of the auxiliary data and a probability density function of the clutter power of the original data are established to form the clutter power relationship model.
[0117] Although the clutter covariance matrix of the auxiliary data is different from the clutter covariance matrix of the original data in the heterogeneous scene, in practice, they are related to each other in a potential relationship in many scenes, and this property ensures that the auxiliary data can be used to infer the clutter covariance matrix of the original data used in the detector. A flexible method based on a statistical model is adopted, assuming that the clutter covariance matrix of the original data is the average value of the clutter covariance matrix of the auxiliary data, and this method successfully describes the relationship between the two. According to this method, the present embodiment establishes a clutter covariance matrix relationship model between the original data and the auxiliary data (i.e., the probability density function of the covariance matrix of the auxiliary data and the probability density function of the covariance matrix of the original data). At the same time, a model of the clutter power relationship between the original data and the auxiliary data (i.e., the probability density function of the auxiliary data clutter power and the probability density function of the original data clutter power) is also added to enhance the model's description performance of environmental heterogeneity.
[0118] Specifically, assuming that the power parameter τ of the auxiliary data of the nth subband is nj is τ n0 is the inverse gamma distribution conditional on the scale parameter, and the covariance matrix R nj R n0 is the complex inverse Wishart distribution under the precision matrix condition, then the probability density function of the auxiliary data clutter power is:
[0119]
[0120] Among them, τ nj A non-negative random variable representing the auxiliary data clutter power, τ n0 A non-negative random variable representing the original data clutter power, qnj represents the shape parameter, N represents the number of subbands, n represents the nth subband, and J represents that each subband has J auxiliary data units.
[0121] The probability density function of the covariance matrix of the auxiliary data is:
[0122]
[0123] Among them, ν nj >M+1 represents the degree of freedom, M represents the number of pulses during the coherent processing interval, and R nj represents the auxiliary data clutter covariance matrix, R n0 represents the original data clutter covariance matrix, N represents the number of subbands, n represents the nth subband, and J represents that each subband has J auxiliary data units.
[0124] In equations (10) and (11), according to the properties of the inverse gamma distribution and the complex inverse Wishart distribution, the parameter q nj and ν nj Controls the degree of heterogeneity of the clutter environment, i.e., the parameter q nj and ν nj The smaller it is, the more heterogeneous the clutter environment is.
[0125] In order to utilize the prior knowledge of clutter in the Bayesian framework, it is necessary to specify the prior distribution of the power of the original data and the clutter covariance matrix. Similar to the auxiliary data, the power and clutter covariance matrix of the original data are also assumed to be inverse gamma distribution and complex inverse Wishart distribution, respectively. The specific form of the probability density function of the original data clutter power is as follows:
[0126]
[0127] Among them, τ n0 A non-negative random variable representing the original data clutter power, β n represents the scale parameter of the inverse gamma distribution, p n Represents the shape parameter.
[0128] The probability density function of the covariance matrix of the original data is:
[0129]
[0130] Among them, n >M+1 represents the degree of freedom, Σ n represents the precision matrix, R n0 represents the raw data clutter covariance matrix, and M represents the number of pulses during the coherent processing interval.
[0131] In formula (12) and formula (13), β n and Σn Parameters play an important role in clutter and can be estimated from previous data or constructed based on environmental knowledge. n and n Controls the accuracy of prior knowledge, p n and n The more there are, the more accurate the prior knowledge is.
[0132] S4. Establishing a broadband radar target detection problem model according to each pulse signal and the clutter of the auxiliary data.
[0133] Specifically, let x n is the measured data vector of the original data of the nth subband, z nj , j = 1, ..., J is the auxiliary data vector of the nth subband, and then the broadband radar target detection problem model can be restated as:
[0134]
[0135] Among them, H1 represents the situation including target signal and clutter signal, x n represents the measured data vector of the original data in the nth subband, p w represents the Doppler steering vector at the wth frequency point, a represents a variable, q n Represents a variable, c n0 represents the clutter of the unit under test in the nth subband, z nj represents the auxiliary data vector of the nth subband, c nj represents the noise of the auxiliary data in the nth subband, n represents the nth subband, N represents the number of subbands, J represents that each subband has J auxiliary data units, and H0 represents the case where the noise signal is included.
[0136] In the case of H0 and H1, τ n0 and R n0 The probability density function of the measured data under the condition can be expressed by the following unified expression:
[0137]
[0138] Among them, x n0 Represents τ n0 and R n0 is the measurement data of the condition, τ n0 A non-negative random variable representing the original data clutter power, R n0 represents the original data clutter covariance matrix, q represents H0 or H1, a represents the variable, x n represents the measured data vector of the original data in the nth subband, p wrepresents the Doppler steering vector at the wth frequency point, q n represents a variable, N represents the number of subbands, and n represents the nth subband.
[0139] S5. Combining the broadband radar target detection problem model, the clutter of the auxiliary data, the probability density function, the clutter covariance matrix relationship model and the clutter power relationship model, a knowledge-assisted detector is constructed using a posterior probability density function.
[0140] Specifically, the likelihood ratio test (LRT) of the wideband radar target detection problem model in equation (14) is given by:
[0141]
[0142] Among them, x 0,…,N-1 ,Z 0,…,N-1 They respectively represent the measured data vector of the original data in the nth subband and the auxiliary data vector of the nth subband.
[0143] Since the clutter in different sub-bands is independent, the distribution of the observed data is the product of the distribution functions of all sub-band observed data:
[0144]
[0145] Among them, x n Represents the measured data vector of the original data in the nth subband, Z n Represents the auxiliary data vector for the nth subband.
[0146] However, the likelihood function of the subband is too complex to be calculated using a closed form, and the LRT in equation (16) cannot be directly implemented. Therefore, according to the target and clutter model of this embodiment, the following method is used to design a detector for heterogeneous broadband radar target detection.
[0147] Specifically, the knowledge-assisted detector includes one or more of a knowledge-assisted maximum a posteriori ratio detector, a knowledge-assisted Rao detector, and a knowledge-assisted Wald detector.
[0148] When the knowledge-assisted detector is the knowledge-assisted maximum a posteriori ratio detector, step S5 comprises:
[0149] S51. Combining the broadband radar target detection problem model, the clutter of the auxiliary data, the probability density function, the clutter covariance matrix relationship model and the clutter power relationship model, based on the maximum a posteriori ratio test framework of the radar system, construct the knowledge-assisted maximum a posteriori ratio detector.
[0150] Specifically, since the likelihood function is too complex to be directly calculated according to formula (17), this embodiment alternately uses the posterior probability density function to replace the likelihood function.
[0151] Given the measured data vector x of the original data n and auxiliary data vector z n(1,…,J) After that, the parameter τ n0 and R n0 The posterior distribution of is as follows:
[0152]
[0153] According to the knowledge of Bayesian theory, given the auxiliary data, the parameter τ n0 and R n0 The posterior distribution of is:
[0154]
[0155] In formula (19),
[0156]
[0157] In formula (20),
[0158]
[0159] In formula (21),
[0160]
[0161] in, represents the conjugate transpose of the auxiliary data vector of the nth subband, q nj represents the auxiliary data shape parameter, ν nj represents the auxiliary data degrees of freedom, z nj represents the auxiliary data vector of the nth subband, Γ represents the gamma function, z n(1,…,J) represents each auxiliary data vector in the nth subband.
[0162] Therefore, based on the maximum a posteriori ratio (MAP) test framework of radar systems, the knowledge-assisted maximum a posteriori ratio test (KAMAPRT) for wideband radar is as follows:
[0163]
[0164] Among them, R n0 represents the original data clutter covariance matrix, τ n0 represents the non-negative random variable of the original data clutter power, a represents the variable, H1 represents the case including the target signal and the clutter signal, x n represents the measured data vector of the original data in the nth subband, zn(1,…,J) represents each auxiliary data vector in the nth subband, and H0 represents the case where a clutter signal is included.
[0165] For p(R n0 ,τ n0 ,qa,H1|x n ,z n(1,…,J) ) Take the logarithm, T KAMAPRT It can be rewritten as follows:
[0166]
[0167] S52, obtain parameter τ based on prior knowledge and auxiliary data n0 and R n0 The maximum a posteriori probability ratio estimate.
[0168] Specifically, in order to realize the knowledge-assisted maximum a posteriori ratio test, the parameters a and τ need to be n0 and R n0 Make an estimate.
[0169] Substituting equations (9) to (13), (15) and (19) into equation (18), we can obtain:
[0170]
[0171] In the case of H0, given the auxiliary data and the measured data in the unit under test, let and is the parameter τ n0 and R n0 The estimated value under the maximum a posteriori probability ratio criterion is
[0172]
[0173] In the H1 case, given the auxiliary data and the measured data in the unit under test, let and is the parameter τ n0 and R n0 The estimated value under the maximum a posteriori probability ratio criterion is
[0174]
[0175] According to formula (27), for the parameter τ n0 and R n0 The estimate of depends on a. In the case of H1, the estimate of parameter a depends on τ n0 and R n0 , so for parameters a, τ n0 and R n0To solve this problem, this embodiment uses an approximate method to obtain their estimates in two steps.
[0176] First, the parameter τ is obtained based on prior knowledge and auxiliary data n0 and R n0 Specifically, we can estimate the maximum a posteriori probability ratio by setting the parameter τ n0 and R n0 To maximize the posterior distribution, we use and To represent the parameter τ n0 and R n0 The maximum a posteriori probability estimate of
[0177]
[0178] in, Represents parameter R n0 The maximum a posteriori probability estimate of Represents the parameter τ n0 The maximum a posteriori probability estimate, R n0 represents the original data clutter covariance matrix, τ n0 A non-negative random variable representing the original data clutter power, z n(1,…,J) represents the auxiliary data vector of the nth subband, n represents the nth subband, and N represents the number of subbands.
[0179] Through formula (28) we can get and Then, according to the parameter τ n0 and R n0 The maximum a posteriori probability ratio estimate is:
[0180]
[0181] in, represents the estimated value of parameter a, x n The measured data vector representing the original data in the nth subband.
[0182] S53, in the case of including target signal and clutter signal, by using To replace the method of a, we get the parameter τ n0 and R n0 That is, in the case of H1, by using To replace a method to get the parameter τ n0 and R n0 The estimated value of .
[0183] Through the above design method, a knowledge-assisted wideband radar target detector in non-uniform clutter is obtained. The detector is a knowledge-assisted maximum a posteriori ratio detector, which is used for wideband radar target detection in non-uniform clutter environment.
[0184] The design method of this embodiment adopts a multi-pulse coherent processing model to describe the broadband radar range migrating target echo, obtains each pulse signal in each sub-band and during the coherent processing interval, and adopts a hierarchical Bayesian model to describe the non-uniform clutter, and obtains the clutter of the auxiliary data in each sub-band. On the basis of the above, a knowledge-assisted maximum a posteriori ratio detector for broadband radar target detection in a non-uniform clutter environment is proposed, which solves the problem of broadband radar target detection in a non-uniform clutter environment. The knowledge-assisted maximum a posteriori ratio detector is effective for broadband radar target detection under a clutter background, and the performance of the detector is improved in certain scenarios.
[0185] Embodiment 2
[0186] Based on Example 1, please combine Figure 1 , This embodiment is described by taking the knowledge-assisted Rao detector as an example.
[0187] The design method includes the steps of:
[0188] S1. Divide the broadband radar target signal into several sub-bands, obtain each pulse signal of each sub-band and during the coherent processing interval, so as to establish a signal model of the broadband radar target.
[0189] S2. Divide the broadband radar non-uniform clutter signal into several sub-bands, and model the clutter covariance matrix as a random matrix, obtain the clutter of the auxiliary data in each sub-band to establish a non-uniform clutter model, wherein the clutter of the auxiliary data obeys a composite Gaussian distribution.
[0190] S3. Obtain the probability density function of the clutter of the auxiliary data in each sub-band, and establish a clutter covariance matrix relationship model and a clutter power relationship model between the original data and the auxiliary data.
[0191] S4. Establishing a broadband radar target detection problem model according to each pulse signal and the clutter of the auxiliary data.
[0192] For the specific implementation method of steps S1 to S4, please refer to the first embodiment, which will not be described in detail in this embodiment.
[0193] S5. Combining the broadband radar target detection problem model, the clutter of the auxiliary data, the probability density function, the clutter covariance matrix relationship model and the clutter power relationship model, a knowledge-assisted detector is constructed using a posterior probability density function.
[0194] Specifically, the Rao detector is a detector for wideband radar target detection. In many applications, the Wald detector outperforms current radar target detectors based on the general likelihood transformation (GLRT) design. Because the maximum likelihood estimation (MLE) of unknown parameters is not required, the Rao detector requires less computation than the GLRT detector.
[0195] The specific design process of the Rao detector is:
[0196] Assume that the τ0 vector contains the clutter power of all sub-band main data in the following form:
[0197] τ0=[τ 00 τ 10 … τ (N-1)0 ] T (30)
[0198] Where τ0 represents the clutter power of all sub-band main data, τ (N-1)0 Indicates the clutter power of the N-1th sub-band main data.
[0199] Set parameters The Rao detector can be defined as follows:
[0200]
[0201] in, is the estimated value of parameter θ under the H0 hypothesis, represents the Fisher information matrix of parameter θ under the H0 hypothesis, that is, is the estimated value of parameter τ0 under the H0 hypothesis.
[0202] According to equation (18), the distribution of the measured data is very complex and the Rao detector cannot be directly obtained according to its definition. Therefore, a two-step method is used to design the knowledge-assisted Rao detector for wideband radar.
[0203] In the first step, assuming that the covariance matrix of the clutter is known and the measured data is Gaussian distributed, the Fisher information matrix of the parameter θ can be divided into the following form:
[0204]
[0205] Using the inverse formula of the block matrix, the block matrix I aa The inverse matrix of (θ) can be decomposed into:
[0206]
[0207] In formula (33),
[0208]
[0209] About f(x) with respect to parameter a * The partial derivatives of are as follows:
[0210]
[0211] The form under the H0 hypothesis is:
[0212]
[0213] Finally, the value defined in formula (27) and Substituting into equations (34), (37) and (31), we obtain the wideband radar knowledge-assisted Rao detector, which is specifically:
[0214]
[0215] in, Indicates H n The conjugate transpose of In the case of H0, given the auxiliary data and the measured data in the unit under test, for the parameter τ n0 The maximum a posteriori probability estimate of Indicates that under the condition of H0, given the auxiliary data and the measured data in the unit under test, for the parameter R n0 The maximum a posteriori probability estimate, x n represents the measured data vector of the original data in the nth subband, n represents the nth subband, and N represents the number of subbands.
[0216] Through the above design method, this embodiment obtains a knowledge-assisted wideband radar target detector in non-uniform clutter, which is a knowledge-assisted Rao detector used for wideband radar target detection in a non-uniform clutter environment.
[0217] The design method of this embodiment is based on the clutter of each pulse signal in each subband and the auxiliary data in each subband during the coherent processing interval, and proposes a knowledge-assisted Rao detector for broadband radar target detection in a non-uniform clutter environment, which solves the problem of broadband radar target detection in a non-uniform clutter environment. The knowledge-assisted Rao detector is effective for broadband radar target detection under a clutter background, and the detector has good performance in certain scenarios.
[0218] Embodiment 3
[0219] Based on Example 1, please combine Figure 1 , This embodiment is described by taking the knowledge-assisted Wald detector as an example.
[0220] The design method includes the steps of:
[0221] S1. Divide the broadband radar target signal into several sub-bands, obtain each pulse signal of each sub-band and during the coherent processing interval, so as to establish a signal model of the broadband radar target.
[0222] S2. Divide the broadband radar non-uniform clutter signal into several sub-bands, and model the clutter covariance matrix as a random matrix, obtain the clutter of the auxiliary data in each sub-band to establish a non-uniform clutter model, wherein the clutter of the auxiliary data obeys a composite Gaussian distribution.
[0223] S3. Obtain the probability density function of the clutter of the auxiliary data in each sub-band, and establish a clutter covariance matrix relationship model and a clutter power relationship model between the original data and the auxiliary data.
[0224] S4. Establishing a broadband radar target detection problem model according to each pulse signal and the clutter of the auxiliary data.
[0225] For the specific implementation method of steps S1 to S4, please refer to the first embodiment, which will not be described in detail in this embodiment.
[0226] S5. Combining the broadband radar target detection problem model, the clutter of the auxiliary data, the probability density function, the clutter covariance matrix relationship model and the clutter power relationship model, a knowledge-assisted detector is constructed using a posterior probability density function.
[0227] Specifically, the Wald detector is a detector for wideband radar target detection. In many applications, the performance of the Wald detector is better than the current radar target detector based on the general likelihood transformation (GLRT) design.
[0228] The Wald detector is defined as:
[0229]
[0230] in, is the estimated value of parameter θ under the H1 hypothesis, represents the Fisher function of the parameter θ under the H1 hypothesis
[0231]
[0232] The knowledge-assisted Wald detector is also designed in a two-step method, which is roughly the same as the knowledge-assisted Rao detector. By using the formula (29) To replace a, the definition in formula (27) and Substituting into equations (34), (37) and (39) we can obtain the wideband radar knowledge-assisted Wald detector, which is in the form of:
[0233]
[0234] in, Indicates H n The conjugate transpose of In the case of H1, given the auxiliary data and the measured data in the unit under test, for the parameter τ n0 The maximum a posteriori probability estimate of In the case of H1, given the auxiliary data and the measured data in the unit under test, for the parameter R n0 The maximum a posteriori probability estimate, x n represents the measured data vector of the original data in the nth subband, n represents the nth subband, and N represents the number of subbands.
[0235] Through the above design method, this embodiment obtains a knowledge-assisted wideband radar target detector in non-uniform clutter, which is a knowledge-assisted Wald detector used for wideband radar target detection in a non-uniform clutter environment.
[0236] The design method of this embodiment is based on the clutter of each pulse signal in each sub-band and the auxiliary data in each sub-band during the coherent processing interval, and proposes a knowledge-assisted Wald detector for broadband radar target detection in a non-uniform clutter environment, which solves the problem of broadband radar target detection in a non-uniform clutter environment. The knowledge-assisted Wald detector is effective for broadband radar target detection in a clutter background, and the detector has good performance in certain scenarios.
[0237] Embodiment 4
[0238] Based on the first, second and third embodiments, this embodiment verifies the performance of the knowledge-assisted maximum a posteriori ratio detector, the knowledge-assisted Rao detector and the knowledge-assisted Wald detector.
[0239] Since the probability density functions of the knowledge-assisted maximum a posteriori probability ratio detector, the knowledge-assisted Rao detector, and the knowledge-assisted Wald detector are too complex, it is impossible to calculate their detection probabilities (P D ) and false alarm probability (P FA ), so they are estimated through Monte Carlo experiments. In the experiment, the false alarm probability (P FA ) is set to P FA =10 -3 , after 100 / P FA =10 5After 10 independent experiments, the threshold is estimated. 4 Independent experiments have been conducted to estimate the detection probability of the detector under various signal-to-clutter ratios (SCR). According to equation (7), the broadband radar target echo signal is modeled as a linear model. When the clutter covariance matrix and clutter power are known in the H1 case, the GLRT distribution of the linear model signal is a non-central complex chi-square distribution. Therefore, the SCR is defined as the non-central parameter, that is,
[0240]
[0241] Assuming the precision matrix Σ n is exponential in shape with a lagged autocorrelation coefficient, namely:
[0242]
[0243] Among them, ρ n represents the lagged autocorrelation coefficient, i represents the variable, and j represents the variable.
[0244] The carrier frequency is f c =10GHz, bandwidth is B w =1GHz, that is, the distance cell size is 0.15m.
[0245] A. Analyze the detector threshold based on the clutter power and clutter covariance matrix.
[0246] For radar target detectors, it is usually expected that the false alarm rate is constant relative to the characteristics of the clutter power and the clutter covariance matrix. Since the constant false alarm rate characteristics of the detector designed in this embodiment cannot be analyzed using a closed formula, it is investigated through numerical simulation.
[0247] In the simulation, the adjacent distance unit L=8, the number of subbands N=16, the number of pulses during the coherent processing interval (CPI) M=8, the radial velocity of the target v=75m / s, and the raw data shape parameter p n =1.01, original data degrees of freedom ν n =10, auxiliary data shape parameter q nj =1.01, auxiliary data degrees of freedom ν nj = 10. First, the autocorrelation coefficient ρ of the fixed lag n = 0.75, let the parameter β controlling the clutter power n The thresholds of the three proposed detectors are shown in Table 1. From Table 1, it can be found that the knowledge-assisted maximum a posteriori probability ratio detector threshold has almost no effect on β. nThis means that the knowledge-assisted maximum a posteriori probability ratio detector is approximately a constant false alarm detection under clutter power; however, as the clutter power increases, the thresholds of the knowledge-assisted Rao detector and the knowledge-assisted Wald detector also increase.
[0248] Table 1 Detector thresholds for different clutter powers
[0249] <![CDATA[β n ]]> KAMAPRT KA Rao KA Wald 0.1 61.8921 10.5209 67.4577 1 62.7252 19.7601 69.4373 10 60.4420 42.9803 90.0866 100 62.3739 92.2926 140.541
[0250] Then, the fixed parameter β n =1, let the autocorrelation coefficient ρ n The thresholds of the three proposed detectors are shown in Table 2. From Table 2, it can be found that all the detection thresholds of the three proposed detectors have almost no relationship with the autocorrelation coefficient ρ. n changes, which means that they are approximately constant false alarm with the clutter covariance matrix.
[0251] Table 2 Detector thresholds for different clutter autocorrelation coefficients
[0252] <![CDATA[ρ n ]]> KAMAPRT KA Rao KA Wald 0.45 62.6806 19.5090 62.8292 0.6 61.6328 19.3182 69.0755 0.75 62.7252 19.7601 69.4373 0.9 61.3849 19.5031 69.3232
[0253] B. The degree of clutter heterogeneity verifies the performance of the detector
[0254] When τ nj The shape parameter q of the distribution nj τ n0 As the condition, R nj The degrees of freedom parameter ν of the distribution nj R n0 When the condition is reduced, the non-uniformity of the clutter increases. This embodiment will evaluate the detection performance of the detector under different heterogeneous environments. Specifically, set the adjacent range unit L = 8, the number of sub-bands N = 16, the number of pulses during the coherent processing interval M = 8, the number of auxiliary data units J = 8, the radial velocity of the target v = 75m / s, and the raw data shape parameter p n =1.01, original data degrees of freedom ν n =10.
[0255] See also Figure 2 , Figure 2 A schematic diagram of the relationship between the detection probability and SCR of a detector in an environment with different degrees of clutter heterogeneity provided by an embodiment of the present invention. Figure 2 Shows different q nj and ν nj The performance of the detector is due to p n and n The smaller the value of is, the lower the accuracy of the clutter prior information is. For the clutter environment in the detection unit and the reference unit, the same prior information is Figure 2 The performance of the MAPGLRT detector for object detection is also compared.
[0256] from Figure 2 It can be seen that the performance of KAMAPRT and KAWald detectors does not change significantly with the degree of environmental heterogeneity, and their performance characteristics are very close. In environments with high heterogeneity, the performance of KARao detector is relatively poor. However, as the heterogeneity decreases, the performance of KARao detector improves significantly and even exceeds KAMAPRT and KAWald. Figure 2 It can also be seen that MAPGLRT is not suitable for heterogeneous environments where the accuracy of clutter prior information is low.
[0257] C. Verify the performance of the detector using prior knowledge of clutter
[0258] Specifically, when τ n0 The shape parameter p of the distribution n and R n0 The degree of freedom parameter of the distribution is n When increases, the accuracy of clutter prior knowledge and the degree of clutter non-uniformity increase. Therefore, this embodiment will evaluate the detection performance of the detector under different accuracies of clutter prior knowledge.
[0259] Specifically, the adjacent distance units L=8, the number of subbands N=16, the number of pulses during the coherent processing interval M=8, the number of auxiliary data units J=8, the radial velocity of the target v=75m / s, and the auxiliary data shape parameter q nj =1.01, auxiliary data degrees of freedom ν nj =10.
[0260] See also Figure 3 , Figure 3 A schematic diagram of the relationship between the detection probability and SCR of a detector in an environment where the clutter prior knowledge has different accuracies is provided in an embodiment of the present invention. Figure 3 Shows different p n and n Worth the performance of the detector.
[0261] By comparison Figure 2 (a) Figure 3 (a) and Figure 3 (b) It can be seen that when the heterogeneity of clutter is the same, the performance of all detectors improves with the improvement of the accuracy of prior knowledge, especially for KARao detector and MAPGLRT detector. n =4,υ n When =32, the detection performance of KARao and MAPGLRT is close to that of KAPAPRT and KAWald.
[0262] D. Verify the performance of the detector using the number of auxiliary data units.
[0263] In this set of simulations, the effect of the number of reference units on the detector performance is evaluated. Specifically, the range unit L is set to 8, the number of subbands N is set to 16, the number of pulses during the coherent processing interval M is set to 8, the radial velocity of the target v is set to 75 m / s, and the raw data shape parameter p is set to n =1.01, original data degrees of freedom ν n = 10, the performance of the detector can be found in Figure 4 , Figure 4 A schematic diagram of the relationship between the detection probability of a detector and the SCR under different heterogeneous conditions and numbers of auxiliary units provided by an embodiment of the present invention.
[0264] Figure 4 (a) and Figure 4 (b) The effects of the number of reference units on the detection performance of the proposed detectors are compared when the clutter heterogeneity is high. The results show that when the number of reference units is small, the detection performance of KAMAPRT and KAWald decreases seriously, while the detection performance of KARao is better. Figure 4 (c) and Figure 4 (d) The effect of the number of reference units on the performance of the proposed detector is compared when the clutter heterogeneity is low; the results show that similar conclusions can be drawn as when the clutter heterogeneity is high, but the performance of the proposed detector is higher than Figure 4 (a) and Figure 4 (b), and the performance of KAWald exceeds that of KARao at high SCR.
[0265] E. Verifying the performance of the detector when the parameters are mismatched based on prior knowledge
[0266] In the prior art, it is assumed that the shape parameters and degrees of freedom parameters of the clutter power prior distribution and the clutter covariance matrix are precisely known, but in practice, they are estimated based on historical data and inevitably have some errors. Therefore, the impact of the errors of these parameters on the performance of the proposed detector is evaluated in this set of simulations.
[0267] Specifically, the adjacent distance units L=8, the number of subbands N=16, the number of pulses during the coherent processing interval M=8, the number of auxiliary data units J=8, the radial velocity of the target v=75m / s, and the original data shape parameter p n =2, original data degrees of freedom ν n =16, auxiliary data shape parameter q nj =2, auxiliary data degrees of freedom ν nj =16.
[0268] See also Figure 5 , Figure 5 A probability diagram of a detection curve of a detector when the parameters of a prior distribution do not match is provided in an embodiment of the present invention, wherein: Figure 5 (a) is the detection probability curve of the detector with precisely known hyperparameters, Figure 5 (b) and 5(c) are the result plots with mismatched hyperparameters.
[0269] from Figure 5 (a) It can be seen that when the clutter prior knowledge is accurate and the environment heterogeneity is not serious, when the hyperparameters of the clutter prior distribution are known, the three detectors have similar detection performance. Figure 5 (b) shows the performance of the proposed detector under the same environment, but the clutter hyperparameters are not exact and are set to p n =3,υ n =24,q nj =3, ν nj = 24. From the results, it can be found that the performance degradation of KAMAPRT and KAWald detectors is small, but the performance degradation of KARao detector is obvious. Figure 5 (c) shows that when the hyperparameter of the clutter is set to p n =1.01,υ n =10,q nj =1.01,ν nj =10. The results show that the performance of KAMAPRT detector is almost not degraded, the performance of KAWald detector is slightly degraded, but the performance of KARao is seriously degraded.
[0270] F. Verifying the performance of the detector using clutter synthesized from real SAR images
[0271] Specifically, a real X-band synthetic aperture radar (SAR) complex image is used to synthesize single-channel airborne pulse Doppler radar clutter to verify the performance of the proposed detector. The resolution of the SAR image is 0.3m*0.3m, as shown in Figure 6 As shown, Figure 6A schematic diagram of a SAR image for generating clutter provided by an embodiment of the present invention. Assume that the first side-looking airborne radar detects an area of the SAR image. Assume that the speed of the aircraft carrier is 100m / s and the flight altitude is 1000m. The beam width of the airborne radar is assumed to be 1°, the radar bandwidth is 250MHz, the pulse repetition frequency is 1000Hz, there are 8 pulses in the coherent processing interval, and the detector uses 16 reference units. First, clutter is generated based on the SAR image and the airborne radar parameters according to the Ward method, and then the Markov chain Monte Carlo method is used to estimate the prior distribution parameters of the clutter in equations (10) to (13). Next, assume that the error of the position of the aircraft carrier in the X, Y and Z directions is 0.1m, and generate clutter to verify the performance of the proposed detector. Assume that the artificially added target is extended in 8 distance units and the target speed is 150m / s.
[0272] Furthermore, the performance of the proposed detector is verified in the high non-uniformity region (region A in the SAR image) and the low non-uniformity region (region B in the SAR image). The detection curves are shown in Figure 2. Figure 7 As shown, Figure 7 A schematic diagram of a detector detection probability curve in the case of synthetic clutter in a SAR image provided by an embodiment of the present invention.
[0273] from Figure 7 It can be found from the results shown that KAMAPRT and KAWald are robust to the heterogeneity of the clutter, but the performance of KARao degrades in high heterogeneity areas, which is consistent with the results of simulation B.
[0274] In summary, this embodiment evaluates the detection performance of three broadband radar target detectors, KAMAPRT, KARao and KAWald, by simulating clutter with a probability model and synthesizing clutter with a real synthetic aperture radar (SAR) complex image. All three broadband radar target detectors handle the range offset problem of broadband radar targets. Among them, KAMAPRT is a constant false alarm detector with clutter power and clutter covariance matrix, but the other two detectors do not have this feature. KAMAPRT and KAWald algorithms have better robustness than KARao algorithm, but KARao algorithm has the smallest amount of calculation.
[0275] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the scope of protection of the present invention.
Claims
1. A knowledge-assisted wideband radar target detector design method in non-uniform clutter, characterized in that: Includes steps: S1. Divide the broadband radar target signal into several sub-bands to obtain each pulse signal of each sub-band during the coherent processing interval; S2, dividing the broadband radar non-uniform clutter signal into several sub-bands, and modeling the clutter covariance matrix as a random matrix to obtain the clutter of the auxiliary data in each sub-band, wherein the clutter of the auxiliary data obeys a composite Gaussian distribution; S3, obtaining the probability density function of the clutter of the auxiliary data in each sub-band, and establishing a clutter covariance matrix relationship model and a clutter power relationship model between the original data and the auxiliary data; Step S3 includes: S31, obtaining the probability density function of the clutter of the auxiliary data in each sub-band with the non-negative random variable of the clutter power and the covariance matrix as conditions: Among them, c nj represents the clutter of the auxiliary data in the nth subband, τ nj A non-negative random variable representing the clutter power, R nj represents the covariance matrix, M represents M pulses, J represents J auxiliary data units in each subband, and N represents the number of subbands; S32, based on the fact that the clutter covariance matrix of the original data is the average value of the clutter covariance matrix of the auxiliary data, a probability density function of the covariance matrix of the auxiliary data and a probability density function of the covariance matrix of the original data are established to form the clutter covariance matrix relationship model, and a probability density function of the clutter power of the auxiliary data and a probability density function of the clutter power of the original data are established to form the clutter power relationship model, wherein, The probability density function of the covariance matrix of the auxiliary data is: Among them, ν nj >M+1 represents the degree of freedom, M represents the number of pulses during the coherent processing interval, and R nj represents the auxiliary data clutter covariance matrix, R n0 represents the original data clutter covariance matrix, N represents the number of subbands, n represents the nth subband, and J represents that each subband has J auxiliary data units; The probability density function of the covariance matrix of the original data is: Among them, n >M+1 represents the degree of freedom, Σ n represents the precision matrix, R n0 represents the raw data clutter covariance matrix, M represents the number of pulses during the coherent processing interval; The probability density function of the auxiliary data clutter power is: Among them, τ nj A non-negative random variable representing the auxiliary data clutter power, τ n0 A non-negative random variable representing the original data clutter power, q nj represents the shape parameter, N represents the number of subbands, n represents the nth subband, and J represents that each subband has J auxiliary data units; The probability density function of the original data clutter power is: Among them, τ n0 A non-negative random variable representing the original data clutter power, β n represents the scale parameter of the inverse gamma distribution, p n represents the shape parameter; S4, establishing a broadband radar target detection problem model according to each pulse signal and the clutter of the auxiliary data; S5. Combining the broadband radar target detection problem model, the clutter of the auxiliary data, the probability density function, the clutter covariance matrix relationship model and the clutter power relationship model, a knowledge-assisted detector is constructed using the maximum a posteriori probability density function.
2. The method for designing a knowledge-assisted wideband radar target detector in non-uniform clutter according to claim 1, characterized in that: Step S1 includes: S11, using a multi-pulse coherent processing model to process the broadband radar target echo to obtain m discrete baseband echo signals after broadband radar pulse compression: Where L represents the adjacent distance unit, represents the complex amplitude of the target echo in the lth distance unit, B represents the broadband radar transmission bandwidth, k represents the detection window distance unit, T s represents the coherent processing interval, ΔR represents the size of the range unit, c represents the speed of light, v represents the radial velocity, m represents the number of broadband radars, and T r represents the pulse repetition interval, j represents the complex unit, and f c represents the signal frequency, K represents the length of the detection window, and M represents the number of pulses during the coherent processing interval; S12, dividing the discrete baseband echo signal into several sub-bands by a discrete Fourier transform sub-band filter bank, and obtaining each pulse signal of each sub-band and during the coherent processing interval: Where n represents the nth subband, t represents the tth pulse, Represents the complex amplitude of the target echo in the range unit after being divided into N sub-bands, represents the complex amplitude of the target echo in the lth range unit, N represents the number of sub-bands, Sr represents the discrete baseband echo signal after broadband radar pulse compression, k represents the detection window range unit, j represents the complex unit, and f c represents the signal frequency, c represents the speed of light, v represents the radial velocity, m represents the number of broadband radars, T r represents the pulse repetition interval, B represents the wideband transmitted by the wideband radar; S13, performing vector processing on each pulse signal of each sub-band during the coherent processing interval to obtain a target benefit of each sub-band: s n =H n a in, p w represents the Doppler steering vector at the wth frequency point, a represents a variable, q n Represents a variable.
3. The method for designing a knowledge-assisted wideband radar target detector in non-uniform clutter according to claim 1, characterized in that: Step S2 includes: S21, dividing the broadband radar non-uniform clutter signal into a plurality of sub-bands by using a discrete Fourier transform sub-band filter bank; S22. Modeling the clutter covariance matrix as a random matrix in a Bayesian framework to obtain clutter of the auxiliary data in each subband, wherein the clutter of the auxiliary data obeys a composite Gaussian distribution: Among them, c nj represents the clutter of the auxiliary data in the nth subband, M represents M pulses, J represents J auxiliary data units in each subband, τ nj is a non-negative random variable representing the clutter power, Represented by the covariance matrix R nj is a complex Gaussian vector of unit variance conditional on .
4. The method for designing a knowledge-assisted wideband radar target detector in non-uniform clutter according to claim 1, characterized in that: The broadband radar target detection problem model is: Among them, H1 represents the situation including target signal and clutter signal, x n represents the measured data vector of the original data in the nth subband, p w represents the Doppler steering vector at the wth frequency point, a represents a variable, q n Represents a variable, c n0 represents the clutter of the unit under test in the nth subband, z nj represents the auxiliary data vector of the nth subband, c nj represents the noise of the auxiliary data in the nth subband, n represents the nth subband, N represents the number of subbands, J represents that each subband has J auxiliary data units, and H0 represents the case where the noise signal is included.
5. The method for designing a knowledge-assisted broadband radar target detector in non-uniform clutter according to claim 1, characterized in that: The knowledge-assisted detector includes one or more of a knowledge-assisted maximum a posteriori ratio detector, a knowledge-assisted Rao detector, and a knowledge-assisted Wald detector.
6. The method for designing a knowledge-assisted wideband radar target detector in non-uniform clutter according to claim 5, characterized in that: When the knowledge-assisted detector is the knowledge-assisted maximum a posteriori ratio detector, step S5 comprises: S51, combining the broadband radar target detection problem model, the clutter of the auxiliary data, the probability density function, the clutter covariance matrix relationship model and the clutter power relationship model, based on the maximum a posteriori ratio test framework of the radar system, constructing the knowledge-assisted maximum a posteriori ratio detector: Among them, R n0 represents the original data clutter covariance matrix, τ n0 represents the non-negative random variable of the original data clutter power, a represents the variable, H1 represents the case including the target signal and the clutter signal, x n represents the measured data vector of the original data in the nth subband, z n(1,…,J) represents the auxiliary data vector of the nth subband, and H0 represents the case where a clutter signal is included; S52, obtain parameter τ based on prior knowledge and auxiliary data n0 and R n0 The maximum a posteriori probability estimate of : in, Represents parameter R n0 The maximum a posteriori probability estimate of Represents the parameter τ n0 The maximum a posteriori probability estimate, R n0 represents the original data clutter covariance matrix, τ n0 A non-negative random variable representing the original data clutter power, z n(1,…,J) represents the auxiliary data vector of the nth subband, n represents the nth subband, and N represents the number of subbands; According to the parameter τ n0 and R n0 The maximum a posteriori estimate of Calculate the estimate of parameter a: in, represents the estimated value of parameter a, p w represents the Doppler steering vector at the wth frequency point, q n represents a variable, x n A measured data vector representing the original data in the nth subband; S53, in the case of including target signal and clutter signal, by using To replace the method of a, we get the parameter τ n0 and R n0 The estimated value of .
7. The method for designing a knowledge-assisted wideband radar target detector in non-uniform clutter according to claim 5, characterized in that: The knowledge-assisted Rao detector is: in, p w represents the Doppler steering vector at the wth frequency point, q n Represents a variable, It represents the parameter τ when the auxiliary data and the measured data in the unit under test are given. n0 The maximum a posteriori probability estimate of It represents the parameter R under given auxiliary data and measured data in the unit under test. n0 The maximum a posteriori probability estimate, x n represents the measured data vector of the original data in the nth subband, n represents the nth subband, and N represents the number of subbands.
8. The method for designing a knowledge-assisted wideband radar target detector in non-uniform clutter according to claim 5, characterized in that: The knowledge-assisted Wald detector is: in, p w represents the Doppler steering vector at the wth frequency point, q n Represents a variable, It represents the parameter τ when the auxiliary data and the measured data in the unit under test are given. n0 The maximum a posteriori probability estimate of It represents the parameter R under given auxiliary data and measured data in the unit under test. n0 The maximum a posteriori probability estimate, x n represents the measured data vector of the original data in the nth subband, n represents the nth subband, and N represents the number of subbands.
9. A knowledge-assisted broadband radar target detector in non-uniform clutter, characterized in that: The method is designed according to any one of claims 1 to 8.
