A radar target structure feature extraction method and system
By using super-resolution reconstruction model and alternating direction method in radar system, the structural characteristics of radar targets are extracted, and the problem of poor radar target extraction accuracy in the existing technology is solved, and higher quality images are achieved Input and feature extraction.
Patent Information
- Application Number
- CN202210201688.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-03
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2042-03-03
AI Technical Summary
The existing radar target structure feature extraction method is poor in the face of diversified targets, complex observation environments and insufficient data utilization, and is vulnerable to electromagnetic environment interference and data pollution, resulting in information loss and false targets. The appearance of
The super-resolution reconstruction model based on radar echo data, discrete Fourier measurement matrix and wavelet transform basis is adopted to construct optimization problems through sparse constraint terms and model approximation terms, and combined with alternating direction method and reconstruction algorithm, the radar super-resolution image is obtained. , and image preprocessing it to extract target length and shape features.
It significantly improves the extraction accuracy of radar target structure features, enhances image resolution, reduces the impact of noise, and improves the extraction accuracy of target length and shape features.
Smart Images

Figure CN114581303B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of signal processing, and in particular to a method and system for extracting structural features of a radar target. Background Art
[0002] Radar occupies an important position in the fields of information acquisition and perception due to its unique advantages such as all-day, all-weather and long-range operation. Radar target feature extraction refers to the use of radar systems to obtain echo signals of targets in air, space, ground, ocean and other environments, and obtain key information such as the target's geometry, motion state, target type, working status, etc. through signal processing tools such as noise suppression, imaging, and feature analysis, thus providing conditions for target identification and interpretation.
[0003] The structural characteristics of radar targets are important content of target characteristics, including length, dispersion characteristics, geometric shape, etc., which reflect the scale characteristics of the target and are the key basis for target identification. The accuracy of radar target structural feature extraction not only depends on the radar system itself, but is also closely related to radar target characteristics, observation environment, and data utilization. The existing target structural feature methods are mainly developed and designed for traditional ideal conditions. Under the conditions of diversified radar targets, complex observation environment, and insufficient data utilization, they will face many problems and challenges from the aspects of targets, environment, and data utilization.
[0004] First, in terms of radar targets, there will be a problem of diversified target types. Radar targets include satellites, space stations, spacecraft and meteorites, and each type of target presents different structural characteristics. For example, the platform and payload structure of satellite targets are often complex, especially when they perform maneuvering and changing orbits, sailboard deployment and other actions, it becomes very difficult to obtain their structural characteristics; secondly, in terms of observation environment, there will be problems such as complex electromagnetic environment and high intensity of electronic countermeasures. Under ideal conditions, the radar observation effect depends only on the observed target and the radar system; complex observation environment will sharply reduce the observation effect. For example, complex electromagnetic environment will cause distortion, loss, phase error, etc. of radar echo; electromagnetic wave suppression interference will directly lead to the failure of receiving echo, and electromagnetic wave deception interference will lead to the appearance of false electromagnetic targets. These non-ideal radar signals pose a huge challenge to the subsequent target feature extraction. Finally, in terms of data utilization, there will be a problem of high data pollution. Data will be polluted in the links of data acquisition, transmission, storage, etc., which will cause information loss and increase the difficulty of information extraction.
[0005] Faced with the above bottleneck problems, there is an urgent need to develop new radar target structure feature extraction methods. Summary of the invention
[0006] In order to overcome the deficiencies of the prior art, an object of the present invention is to provide a radar target structure feature extraction method and system to solve the problem of poor accuracy in the prior art radar target feature extraction.
[0007] To achieve the above object, the present invention provides the following solutions:
[0008] A radar target structure feature extraction method, comprising:
[0009] A super-resolution reconstruction model is constructed based on radar echo data, discrete Fourier measurement matrix and wavelet transform basis;
[0010] Given the algorithm parameters, the reconstruction algorithm is used to solve the super-resolution reconstruction model to obtain the radar super-resolution image;
[0011] Performing image preprocessing on the radar super-resolution image to obtain a preprocessed image;
[0012] The target length feature and the target shape feature are extracted according to the preprocessed image.
[0013] Preferably, the super-resolution reconstruction model is constructed based on radar echo data, discrete Fourier measurement matrix and wavelet transform basis, including:
[0014] According to the acquired radar image and sparse transformation matrix, a sparse constraint term min||s||1=min||Ψ is constructed. -1 x||1; where x is the radar image, Ψ represents the sparse transformation matrix, s is the sparse transformation parameter, where x=Ψs;
[0015] Construct a model approximation term according to the radar image, the radar echo data and the discrete Fourier measurement matrix Wherein, E is the radar echo data, Φ is the discrete Fourier measurement matrix, and Ψ is the wavelet transform basis;
[0016] Combine the sparse constraint term and the model approximation term to establish an optimization problem to construct the super-resolution reconstruction model Among them, λ>0 is the regularization parameter.
[0017] Preferably, the given algorithm parameters and the use of a reconstruction algorithm to solve a super-resolution reconstruction model to obtain a radar super-resolution image include:
[0018] The super-resolution reconstruction model is transformed once to obtain a transformed model Where r is the transformation parameter;
[0019] Get the augmented Lagrangian function of the primary transformation model Where μ is the augmented variable, δ>0 is the Lagrangian multiplier;
[0020] Based on the augmented Lagrangian function, the primary transformation model is transformed twice to obtain a secondary transformation model Among them, k is the secondary transformation model parameter;
[0021] Given the algorithm parameters, the alternating direction method is used to solve the secondary transformation model to obtain the radar super-resolution image.
[0022] Preferably, given the algorithm parameters, solving the secondary conversion model using the alternating direction method to obtain the radar super-resolution image includes:
[0023] Given the algorithm parameters λ>0, δ>0, and the convergence condition parameter η>0;
[0024] Initialize the initial vector s0 = 0 and the step index n = 0;
[0025] Based on s n+1 =H λδ (s n +Ψ T Φ T (Er k -δμ k -ΦΨs n ))Iterate until the convergence condition ||s is met n+1 -s n ||2≤η, get the solution s of the first optimization subproblem k+1 ; where H λδ is the threshold shrinkage operator;
[0026] Using the least squares algorithm to find The closed form solution of the second optimization subproblem is obtained. k+1 ;
[0027] based on Iterate until the convergence condition ||s is met n+1 -s n ||2≤η, we get the solution μ of the third optimization subproblem k+1 ;
[0028] Repeatedly iterate the solution s of the first optimization subproblem k+1 , the solution to the second optimization subproblem is obtained k+1 And get the solution μ of the third optimization subproblem k+1 steps until the convergence condition ||s is met k+1 -s k ||2≤τ, we get Ψ s k+1, where τ>0, Ψs k+1 is the radar super-resolution image.
[0029] Preferably, the performing image preprocessing on the radar super-resolution image to obtain a preprocessed image includes:
[0030] The radar super-resolution image is limited to obtain the pre-processed image Among them, x(m,n) is a two-dimensional radar image formed by the radar super-resolution image, m and n are the serial numbers of the distance unit and the azimuth unit respectively, and G1 and G2 are the first limiting threshold and the second limiting threshold respectively.
[0031] Preferably, the extracting target length features and target shape features according to the preprocessed image comprises:
[0032] Assume that the number of target scattering resolution units extracted in the range and azimuth directions of the super-resolution image is m respectively. z and m h , then the target distance length and azimuth length are L z =m z *ρ z , L h =m h *ρ h ; where ρ r is the distance resolution, ρ a is the azimuth resolution;
[0033] Calculate the target length feature according to the target distance length and the azimuth length
[0034] Acquire the area where the target is located in the preprocessed image;
[0035] Calculate the area occupied by the target area and count the perimeter of the target area;
[0036] Counting the smallest bounding rectangle of the area where the target is located, and obtaining the first side length and the second side length of the bounding rectangle;
[0037] The target shape feature is calculated according to the area, the perimeter, the first side length, and the second side length.
[0038] Preferably, calculating the target shape feature according to the area, the perimeter, the first side length and the second side length comprises:
[0039] Calculate circularity Wherein A is the area; L is the perimeter;
[0040] Calculate rectangularity Wherein R1 is the first side length; R2 is the second side length;
[0041] Calculate slenderness
[0042] A radar target structure feature extraction system, comprising:
[0043] A model building unit, used for building a super-resolution reconstruction model based on radar echo data, a discrete Fourier measurement matrix and a wavelet transform basis;
[0044] A model solving unit is used to give algorithm parameters and solve the super-resolution reconstruction model using a reconstruction algorithm to obtain a radar super-resolution image;
[0045] A preprocessing unit, used for performing image preprocessing on the radar super-resolution image to obtain a preprocessed image;
[0046] The extraction unit is used to extract the target length feature and the target shape feature according to the preprocessed image.
[0047] Preferably, the model building unit specifically includes:
[0048] Sparse constraint construction module, used to construct sparse constraint term min||s||1=min||Ψ based on the acquired radar image and sparse transformation matrix -1 x||1; where x is the radar image, Ψ represents the sparse transformation matrix, s is the sparse transformation parameter, where x=Ψs;
[0049] A model approximation construction module is used to construct a model approximation term based on the radar image, the radar echo data and the discrete Fourier measurement matrix Wherein, E is the radar echo data, Φ is the discrete Fourier measurement matrix, and Ψ is the wavelet transform basis;
[0050] A joint module is used to combine the sparse constraint term and the model approximation term to establish an optimization problem to construct the super-resolution reconstruction model Among them, λ>0 is the regularization parameter.
[0051] Preferably, the model solving unit specifically includes:
[0052] A primary conversion module is used to perform a primary conversion on the super-resolution reconstruction model to obtain a primary conversion model. Where r is the transformation parameter;
[0053] Function acquisition module, used to obtain the augmented Lagrangian function of the primary conversion model Where μ is the augmented variable, δ>0 is the Lagrangian multiplier;
[0054] A secondary transformation module is used to perform a secondary transformation on the primary transformation model based on the augmented Lagrangian function to obtain a secondary transformation model. Among them, k is the secondary transformation model parameter;
[0055] The solution module is used to solve the secondary transformation model using the alternating direction method given the algorithm parameters to obtain the radar super-resolution image.
[0056] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0057] The present invention discloses a method and system for extracting radar target structural features, the method comprising: constructing a super-resolution reconstruction model based on radar echo data, a discrete Fourier measurement matrix and a wavelet transform basis; given algorithm parameters, solving the super-resolution reconstruction model using a reconstruction algorithm to obtain a radar super-resolution image; performing image preprocessing on the radar super-resolution image to obtain a preprocessed image; extracting target length features and target shape features according to the preprocessed image. The present invention mainly utilizes super-resolution imaging technology to improve the accuracy of radar target structural feature extraction, and focuses on the design of super-resolution imaging technology based on compressed sensing, including super-resolution reconstruction modeling and reconstruction algorithm design. Super-resolution imaging technology can improve the resolution of radar images, thereby providing high-quality image input for target structural feature extraction, and significantly improves the accuracy of extracting structural features such as target length and shape. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0059] Figure 1 A flowchart of a radar target structure feature extraction method in an embodiment provided by the present invention;
[0060] Figure 2 A flow chart of radar target structure feature extraction in an embodiment provided by the present invention;
[0061] Figure 3 A flow chart of an alternating optimization algorithm in an embodiment provided by the present invention;
[0062] Figure 4 A diagram of a ship target structure in an embodiment of the present invention;
[0063] Figure 5A schematic diagram of a traditional radar image of a ship target in an embodiment provided by the present invention;
[0064] Figure 6 A schematic diagram of a super-resolution radar image of a ship target in an embodiment provided by the present invention. DETAILED DESCRIPTION
[0065] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0066] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.
[0067] The terms "first", "second", "third" and "fourth" in the specification and claims of the present application and the drawings are used to distinguish different objects rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a series of steps, processes, methods, etc. are not limited to the listed steps, but may optionally include steps that are not listed, or may optionally include other step elements inherent to these processes, methods, products or devices.
[0068] The present invention is based on a series of new signal analysis tools proposed by the mathematics community in recent years to achieve modeling, processing and interpretation of non-ideal signals. In particular, the compressed sensing theory solves underdetermined systems by constraining the solution space of ill-posed inverse problems, which can be directly applied to radar imaging under incomplete signal conditions (such as phase distortion, partial echo loss, etc.). At the same time, the introduction of new constraint functions such as group sparseness and low rank, as well as new signal representation methods such as "local-nonlocal" convolution framework and bilinear dictionary learning make super-resolution imaging of incomplete radar signals possible, providing conditions for high-precision extraction of radar target structural features.
[0069] The purpose of the present invention is to provide a radar target structure feature extraction method and system to solve the problem of poor accuracy of existing radar target feature extraction.
[0070] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0071] Figure 1 For a flow chart of the radar target structure feature extraction method in the embodiment provided by the present invention, please refer to Figure 1 The present invention provides a radar target structure feature extraction method, comprising:
[0072] Step 100: constructing a super-resolution reconstruction model based on radar echo data, discrete Fourier measurement matrix and wavelet transform basis;
[0073] Step 200: Given algorithm parameters, a super-resolution reconstruction model is solved using a reconstruction algorithm to obtain a radar super-resolution image;
[0074] Step 300: performing image preprocessing on the radar super-resolution image to obtain a preprocessed image;
[0075] Step 400: extracting target length features and target shape features according to the preprocessed image.
[0076] Preferably, the super-resolution reconstruction model is constructed based on radar echo data, discrete Fourier measurement matrix and wavelet transform basis, including:
[0077] According to the acquired radar image and sparse transformation matrix, a sparse constraint term min||s||1=min||Ψ is constructed. -1 x||1; where x is the radar image, Ψ represents the sparse transformation matrix, s is the sparse transformation parameter, where x=Ψs;
[0078] Construct a model approximation term according to the radar image, the radar echo data and the discrete Fourier measurement matrix Wherein, E is the radar echo data, Φ is the discrete Fourier measurement matrix, and Ψ is the wavelet transform basis;
[0079] Combine the sparse constraint term and the model approximation term to establish an optimization problem to construct the super-resolution reconstruction model Among them, λ>0 is the regularization parameter.
[0080] Figure 2 The radar target structure feature extraction flow chart in the embodiment provided by the present invention is as follows: Figure 2 As shown, the first step in this embodiment is super-resolution reconstruction modeling, which includes constructing sparse constraint terms, constructing model approximation terms and constructing a super-resolution reconstruction model.
[0081] Specifically, for constructing sparse constraint terms, the sparsity of the target directly leads to the sparsity of its radar image. The sparsity of the image can be understood from two aspects: the sparsity of the image itself and the sparsity in the transform domain. A sparse image refers to an image in which the proportion of pixels with significantly non-zero grayscale values in the entire image is very small. Since zero grayscale value appears as black in the image, a sparse image usually appears as an image that is approximately 'all black'. For targets under the background of the sky and space, due to the particularity of the observation scene, the proportion of the imaging target in the entire scene is small, so the radar image appears as a near-black background with only a few bright spots. At this time, the radar image itself exhibits strong sparsity characteristics. For targets under the background of land and ocean, since the imaging scene is relatively complex, the sparsity of its radar image can be exploited by wavelet transform. In order to unify the sparsity exploitation of radar images under different backgrounds, the matrix Ψ is used to represent the sparse transformation matrix, and the sparse transformation can be expressed as x=Ψs, where x is the radar image. If the image itself is sparse, the sparse transformation matrix is the unit matrix, and x=s; if the image is sparse in the transform domain, the sparse transformation matrix is selected as the wavelet transform basis.
[0082] There are many ways to measure sparsity. Here we choose to use the vector norm to measure its sparsity. The l0 norm is usually used to measure its sparsity, but solving the l0 norm optimization problem is an NP-hard problem. Therefore, the l1 norm is used to measure its sparsity. At this time, the sparse constraint term is constructed as min||s||1=min||Ψ -1 x||1.
[0083] Further, for constructing the model approximation term, this embodiment is based on the local field principle of the GTD theory. In the high-frequency limit, the diffraction field only depends on the physical properties and geometric properties in a very small area near the diffraction point, and has nothing to do with the geometric shape of the object far away from the diffraction point. Theoretical calculations and experimental measurements show that the total electromagnetic scattering of the target in the high-frequency region can be considered to be synthesized by the electromagnetic scattering at certain local positions. These local scatterings are usually called equivalent multi-scattering centers. It can be seen that in the high-frequency region, the echo model can be written as:
[0084] Where κ and θ are wave number variables and angle variables respectively, K is the number of equivalent scattering centers on the target, σ i is the radar cross section of the i-th target, x i With y i is the two-dimensional spatial coordinate of the i-th target. When only considering the frequency-dimensional signal under a certain observation angle (i.e., fixing a certain angle variable θ), the above formula can be simplified to:
[0085]
[0086] Where f = κc / 2, R i =x i cosθ+y i sinθ. Discrete the frequency f=f0+nΔf, n=0,1,...,N-1, Δf is the sampling interval. Ignoring the constant phase, the above formula can be written in the following discrete form:
[0087] Let ω i =2ΔfR i / c, usually the size of radar observation targets is small, ω i ∈[0,1). For the convenience of calculation, ω is discretized into ω m =m / M,m=0,1,...,M-1, let Φ=[φ0,φ1,...,φ M-1 ] is the discrete Fourier measurement matrix, where:
[0088] Then the radar echo signal can be written as the following linear model: E = Φx.
[0089] Among them, E=[E(0),E(1),…,E(N-1)] T ,and represents the radar cross section vector corresponding to the discretization of ω, that is, the radar image to be reconstructed. Considering that the observation generally contains noise, the model approximation term is constructed as
[0090] Optionally, in this embodiment, the distinguishing reconstruction model is to combine the sparse constraint term and the model approximation term to establish an optimization problem:
[0091] Among them, λ>0 is the regularization parameter. The optimal solution of the above model can be obtained by designing the reconstruction algorithm in the next step. Suppose the sparse solution of s is Then the estimation of radar image x is Available through To obtain.
[0092] Preferably, the step 200 includes:
[0093] The super-resolution reconstruction model is transformed once to obtain a transformed model Where r is the transformation parameter;
[0094] Get the augmented Lagrangian function of the primary transformation model Where μ is the augmented variable, δ>0 is the Lagrangian multiplier;
[0095] Based on the augmented Lagrangian function, the primary transformation model is transformed twice to obtain a secondary transformation model Among them, k is the secondary transformation model parameter;
[0096] Given the algorithm parameters, the alternating direction method is used to solve the secondary transformation model to obtain the radar super-resolution image.
[0097] Preferably, given the algorithm parameters, solving the secondary conversion model using the alternating direction method to obtain the radar super-resolution image includes:
[0098] Given the algorithm parameters λ>0, δ>0, and the convergence condition parameter η>0;
[0099] Initialize the initial vector s0 = 0 and the step index n = 0;
[0100] Based on s n+1 =H λδ (s n +Ψ T Φ T (Er k -δμ k -ΦΨs n ))Iterate until the convergence condition ||s is met n+1 -s n ||2≤η, get the solution s of the first optimization subproblem k+1 ; where H λδ is the threshold shrinkage operator;
[0101] Using the least squares algorithm to find The closed form solution of the second optimization subproblem is obtained. k+1 ;
[0102] based on Iterate until the convergence condition ||s is met n+1 -s n ||2≤η, we get the solution μ of the third optimization subproblem k+1 ;
[0103] Repeatedly iterate the solution s of the first optimization subproblem k+1 , the solution to the second optimization subproblem is obtained k+1 And get the solution μ of the third optimization subproblem k+1 steps until the convergence condition ||s is met k+1 -s k ||2≤τ, we get Ψ s k+1 , where τ>0, Ψ s k+1 is the radar super-resolution image.
[0104] In this embodiment, since the optimization problem of the super-resolution reconstruction model is convex, a global optimal solution can be obtained. The original augmented Lagrangian method is used to design the algorithm, and the specific steps are as follows:
[0105] The first aspect is model conversion: for the super-resolution model, it is converted into:
[0106]
[0107] The augmented Lagrangian function of the above formula is:
[0108]
[0109] Where μ is the augmented variable, δ>0 is the Lagrangian multiplier, and the model is further transformed into:
[0110]
[0111] The model is subsequently solved using the alternating direction method.
[0112] The second aspect is solved by the alternating direction method:
[0113] Step 1: Solve the first optimization sub-problem. The first optimization sub-problem is a standard compressed sensing reconstruction model, so it can be solved using an iterative threshold algorithm. The specific steps are as follows:
[0114] 1) Given algorithm parameters: λ>0, δ>0, convergence condition parameter η>0;
[0115] 2) Initialization: initial vector s0 = 0, step index n = 0,
[0116] 3) Iteration: s n+1 =H λδ (s n +Ψ T Φ T (Er k -δμ k -ΦΨs n )) until the convergence condition ||s is met n+1 -s n ||2≤η, where H in the iteration step λδ is the threshold shrinkage operator.
[0117] 4) Output the solution s of the first optimization subproblem k+1 .
[0118] Step 2: Solve the second optimization subproblem. The second optimization subproblem consists of two smooth terms, so the least squares algorithm can be used to find its closed form solution:
[0119]
[0120] Step 3: Solve the third optimization subproblem. Considering that the third optimization subproblem is a direct iterative calculation, there is no need to design an algorithm.
[0121] Step 4: Alternating optimization. Based on the solution algorithms of the above three optimization sub-problems, the alternating direction method can be used to achieve the final solution of the model. See the flowchart for Figure 3 , the specific steps are:
[0122] 1) Given algorithm parameters: λ>0, η>0, τ>0, δ>0;
[0123] 2) Use the iterative hard threshold algorithm to solve the first optimization subproblem and obtain s k+1 ;
[0124] 3) Solve the second optimization subproblem using the least squares algorithm to obtain r k+1 ;
[0125] 4) Calculation
[0126] 5) Repeat iterative steps 2)-4) until the convergence condition ||s is met k+1 -s k ||2≤τ.
[0127] 6) Output Ψ s k+1 is the final solution of the model, i.e., the radar super-resolution image.
[0128] Preferably, the performing image preprocessing on the radar super-resolution image to obtain a preprocessed image includes:
[0129] The radar super-resolution image is limited to obtain the pre-processed image Among them, x(m,n) is a two-dimensional radar image formed by the radar super-resolution image, m and n are the serial numbers of the distance unit and the azimuth unit respectively, and G1 and G2 are the first limiting threshold and the second limiting threshold respectively.
[0130] Preferably, the extracting target length features and target shape features according to the preprocessed image comprises:
[0131] Assume that the number of target scattering resolution units extracted in the range and azimuth directions of the super-resolution image is m respectively. z and m h , then the target distance length and azimuth length are L z =m z *ρ z , Lh =m h *ρ h ; where ρ r is the distance resolution, ρ a is the azimuth resolution;
[0132] Calculate the target length feature according to the target distance length and the azimuth length
[0133] Acquire the area where the target is located in the preprocessed image;
[0134] Calculate the area occupied by the target area and count the perimeter of the target area;
[0135] Counting the smallest bounding rectangle of the area where the target is located, and obtaining the first side length and the second side length of the bounding rectangle;
[0136] The target shape feature is calculated according to the area, the perimeter, the first side length, and the second side length.
[0137] Preferably, calculating the target shape feature according to the area, the perimeter, the first side length and the second side length comprises:
[0138] Calculate circularity Wherein A is the area; L is the perimeter;
[0139] Calculate rectangularity Wherein R1 is the first side length; R2 is the second side length;
[0140] Calculate slenderness
[0141] Specifically, the structural feature extraction in this embodiment is to first pre-process the radar super-resolution image, and then extract the structural features of the target, that is, the target length feature and shape feature. The steps include the following:
[0142] Image preprocessing:
[0143] For radar super-resolution images, the radar resolution unit is generally much smaller than the target's geometric size. In this case, the target can be equivalent to a combination of several scattering points. Due to the different scattering characteristics of each part of the target, the equivalent scattering points on it are strong or weak, and the dynamic range is sometimes very large. Although the amplitude strength of each point on the image reflects the scattering characteristics of each part of the target, the stronger scattering points tend to suppress the ability of the weaker scattering points on the target to express the target's geometric structure. In order to enhance the expression of weak scattering points on the image and suppress noise at the same time, the image can be limited. The specific method is: let x(m,n) be the resulting two-dimensional radar image, m and n are the serial numbers of the distance unit and the azimuth unit respectively, then
[0144]
[0145] Among them, G1 and G2 are the clipping thresholds. The above formula indicates that the image is clipped, which can highlight the weaker scattering points and suppress noise.
[0146] Target length feature extraction:
[0147] After preprocessing, the number of target scattering resolution units extracted from the super-resolution image in range and azimuth directions is set to be m respectively. z 、m h , then the target distance length and azimuth length are
[0148] L z =m z *ρ z , L h =m h *ρ h
[0149] Among them, ρ r is the distance resolution, ρ a is the azimuth resolution. Then the estimated length of the target in the imaging plane is
[0150]
[0151] Target shape feature extraction:
[0152] In order to describe the shape characteristics of the target, the morphological processing of the image is used. By performing target detection on the image, the area where the target is located is obtained, and the area occupied by the area is calculated and set as A. At the same time, the perimeter of the area is counted and recorded as L; on the other hand, the smallest circumscribed rectangle of the area is counted and its side lengths are set as R1 and R2 respectively. Then the shape characteristics of the target such as roundness, rectangularity, and slenderness are calculated according to the following formulas:
[0153] The circularity reflects the degree to which an object is close to a circle and also reflects the compactness of the area. The calculation method is:
[0154]
[0155] Rectangularity is defined as the ratio of the area of the target region to the area of the smallest circumscribed rectangle of the region, and is calculated as follows:
[0156]
[0157] The slenderness is defined as the ratio of the shortest side to the longest side of the minimum bounding rectangle of the target area, and is calculated as:
[0158]
[0159] Compared with traditional radar target structure feature extraction, the target structure features extracted by this method have higher accuracy.
[0160] This embodiment also provides an example of the actual application process. Assume that the radar parameters are set to: carrier frequency 10 GHz, bandwidth 1 GHz, central viewing angle 90°, where the target is a ship, and the structure is shown in FIG. Figure 4 , simulate and generate ship echo data under radar parameters, and then extract target structure features from the ship echo data. The specific steps are:
[0161] S1: Super-resolution reconstruction modeling
[0162] The super-resolution reconstruction model combines the sparse constraint term and the model approximation term to establish an optimization problem:
[0163]
[0164] Among them, E is the radar echo data, Φ is the discrete Fourier measurement matrix, and Ψ is the wavelet transform basis.
[0165] S2: Refactoring Algorithm Design
[0166] Given the algorithm parameters: λ = 0.5, δ = 0.7, η = 0.1, τ = 0.01, the reconstruction algorithm is used to solve the super-resolution reconstruction model, and the radar super-resolution image is obtained. The radar image is compared with the radar image obtained by the traditional radar imaging method. The results are shown in Figure 5 and Figure 6 , we can find the target resolution of radar super-resolution image ( Figure 6 The side lobes are effectively suppressed, the structure of the cabin is more obvious, the details are richer, and the Figure 4 The shapes of the targets correspond one to one.
[0167] S3: Structural feature extraction
[0168] First, the radar super-resolution image is preprocessed, and then the structural features of the target, namely the target length features and shape features, are extracted. The results are shown in Table 1, which shows the feature extraction results of the ship target. It can be found that the feature extraction results of the present invention are closer to the true value than the traditional method, and therefore are better than the traditional method.
[0169] Table 1
[0170]
[0171] This embodiment also provides a radar target structure feature extraction system, including:
[0172] A model building unit, used for building a super-resolution reconstruction model based on radar echo data, a discrete Fourier measurement matrix and a wavelet transform basis;
[0173] A model solving unit is used to give algorithm parameters and solve the super-resolution reconstruction model using a reconstruction algorithm to obtain a radar super-resolution image;
[0174] A preprocessing unit, used for performing image preprocessing on the radar super-resolution image to obtain a preprocessed image;
[0175] The extraction unit is used to extract the target length feature and the target shape feature according to the preprocessed image.
[0176] Preferably, the model building unit specifically includes:
[0177] Sparse constraint construction module, used to construct sparse constraint term min||s||1=min||Ψ based on the acquired radar image and sparse transformation matrix -1 x||1; where x is the radar image, Ψ represents the sparse transformation matrix, s is the sparse transformation parameter, where x=Ψs;
[0178] A model approximation construction module is used to construct a model approximation term based on the radar image, the radar echo data and the discrete Fourier measurement matrix Wherein, E is the radar echo data, Φ is the discrete Fourier measurement matrix, and Ψ is the wavelet transform basis;
[0179] A joint module is used to combine the sparse constraint term and the model approximation term to establish an optimization problem to construct the super-resolution reconstruction model Among them, λ>0 is the regularization parameter.
[0180] Preferably, the model solving unit specifically includes:
[0181] A primary conversion module is used to perform a primary conversion on the super-resolution reconstruction model to obtain a primary conversion model. Where r is the transformation parameter;
[0182] Function acquisition module, used to obtain the augmented Lagrangian function of the primary conversion model Where μ is the augmented variable, δ>0 is the Lagrangian multiplier;
[0183] A secondary transformation module is used to perform a secondary transformation on the primary transformation model based on the augmented Lagrangian function to obtain a secondary transformation model. Among them, k is the secondary transformation model parameter;
[0184] The solution module is used to solve the secondary transformation model using the alternating direction method given the algorithm parameters to obtain the radar super-resolution image.
[0185] The beneficial effects of the present invention are as follows:
[0186] The present invention mainly uses super-resolution imaging technology to improve the accuracy of radar target structure feature extraction, and focuses on the design of super-resolution imaging technology based on compressed sensing, including super-resolution reconstruction modeling and reconstruction algorithm design. Super-resolution imaging technology can improve the resolution of radar images, thereby providing high-quality image input for target structure feature extraction, which will significantly improve the extraction accuracy of target length, shape and other structural features.
[0187] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0188] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of the present invention. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.
Claims
1. A radar target structure feature extraction method, characterized in that: include: A super-resolution reconstruction model is constructed based on radar echo data, discrete Fourier measurement matrix and wavelet transform basis, including: According to the acquired radar image and sparse transformation matrix, a sparse constraint term min||s||1=min||Ψ is constructed. -1 x||1; where x is the radar image, Ψ represents the sparse transformation matrix, s is the sparse transformation parameter, where x=Ψs; Construct a model approximation term according to the radar image, the radar echo data and the discrete Fourier measurement matrix Wherein, E is the radar echo data, Φ is the discrete Fourier measurement matrix, and Ψ is the wavelet transform basis; Combine the sparse constraint term and the model approximation term to establish an optimization problem to construct the super-resolution reconstruction model Among them, λ>0 is the regularization parameter; Given the algorithm parameters, the reconstruction algorithm is used to solve the super-resolution reconstruction model to obtain the radar super-resolution image, which includes: The super-resolution reconstruction model is transformed once to obtain a transformed model Where r is the transformation parameter; Get the augmented Lagrangian function of the primary transformation model Where μ is the augmented variable, δ>0 is the Lagrangian multiplier; Based on the augmented Lagrangian function, the primary transformation model is transformed twice to obtain a secondary transformation model Among them, k is the secondary transformation model parameter; Given the algorithm parameters, the alternating direction method is used to solve the secondary transformation model to obtain the radar super-resolution image; Performing image preprocessing on the radar super-resolution image to obtain a preprocessed image; The target length feature and the target shape feature are extracted according to the preprocessed image.
2. The radar target structure feature extraction method according to claim 1, characterized in that: Given the algorithm parameters, solving the secondary conversion model using the alternating direction method to obtain the radar super-resolution image includes: Given the algorithm parameters λ>0, δ>0, and the convergence condition parameter η>0; Initialize the initial vector s0 = 0 and the step index n = 0; Based on s n+1 =H λδ (s n +Ψ T Φ T (Er k -δμ k -ΦΨs n ))Iterate until the convergence condition ||s is met n+1 -s n ||2≤η, get the solution s of the first optimization subproblem k+1 ; where H λδ is the threshold shrinkage operator; Using the least squares algorithm to find The closed form solution of the second optimization subproblem is obtained. k+1 ; based on Iterate until the convergence condition ||s is met n+1 -s n ||2≤η, we get the solution μ of the third optimization subproblem k+1 ; Repeatedly iterate the solution s of the first optimization subproblem k+1 , the solution to the second optimization subproblem is obtained k+1 And get the solution μ of the third optimization subproblem k+1 steps until the convergence condition ||s is met k+1 -s k ||2≤τ, we get Ψs k+1 , where τ>0, Ψs k+1 is the radar super-resolution image.
3. The radar target structure feature extraction method according to claim 1, characterized in that: The performing image preprocessing on the radar super-resolution image to obtain a preprocessed image includes: The radar super-resolution image is limited to obtain the pre-processed image Among them, x(m,n) is a two-dimensional radar image formed by the radar super-resolution image, m and n are the serial numbers of the distance unit and the azimuth unit respectively, and G1 and G2 are the first limiting threshold and the second limiting threshold respectively.
4. The radar target structure feature extraction method according to claim 1, characterized in that: The step of extracting target length features and target shape features according to the preprocessed image comprises: Assume that the number of target scattering resolution units extracted in the range and azimuth directions of the super-resolution image is m respectively. z and m h , then the target distance length and azimuth length are L z =m z * z , L h =m h * h ; where ρ r is the distance resolution, ρ a is the azimuth resolution; Calculate the target length feature according to the target distance length and the azimuth length Acquire the area where the target is located in the preprocessed image; Calculate the area occupied by the target area and count the perimeter of the target area; Counting the smallest bounding rectangle of the area where the target is located, and obtaining the first side length and the second side length of the bounding rectangle; The target shape feature is calculated according to the area, the perimeter, the first side length, and the second side length.
5. The radar target structure feature extraction method according to claim 4, characterized in that: Calculating the target shape feature according to the area, the perimeter, the first side length, and the second side length comprises: Calculate circularity Wherein A is the area; L is the perimeter; Calculate rectangularity Wherein R1 is the first side length; R2 is the second side length; Calculate slenderness 6. A radar target structure feature extraction system, characterized in that: include: The model building unit is used to build a super-resolution reconstruction model based on radar echo data, discrete Fourier measurement matrix and wavelet transform basis, specifically including: Sparse constraint construction module, used to construct sparse constraint term min||s||1=min||Ψ based on the acquired radar image and sparse transformation matrix -1 x||1; where x is the radar image, Ψ represents the sparse transformation matrix, s is the sparse transformation parameter, where x=Ψs; A model approximation construction module is used to construct a model approximation term based on the radar image, the radar echo data and the discrete Fourier measurement matrix Wherein, E is the radar echo data, Φ is the discrete Fourier measurement matrix, and Ψ is the wavelet transform basis; A joint module is used to combine the sparse constraint term and the model approximation term to establish an optimization problem to construct the super-resolution reconstruction model Among them, λ>0 is the regularization parameter; The model solving unit is used to give algorithm parameters and use the reconstruction algorithm to solve the super-resolution reconstruction model to obtain the radar super-resolution image, which specifically includes: A primary conversion module is used to perform a primary conversion on the super-resolution reconstruction model to obtain a primary conversion model. Where r is the transformation parameter; The function acquisition module is used to obtain the augmented Lagrangian function of the primary conversion model, including: A primary conversion module is used to perform a primary conversion on the super-resolution reconstruction model to obtain a primary conversion model. Where r is the transformation parameter; Function acquisition module, used to obtain the augmented Lagrangian function of the primary conversion model Where μ is the augmented variable, δ>0 is the Lagrangian multiplier; A secondary transformation module is used to perform a secondary transformation on the primary transformation model based on the augmented Lagrangian function to obtain a secondary transformation model. Among them, k is the secondary transformation model parameter; A solution module, used for given the algorithm parameters, solving the secondary transformation model by using the alternating direction method to obtain the radar super-resolution image; wherein μ is an augmented variable, and δ>0 is a Lagrangian multiplier; A secondary transformation module is used to perform a secondary transformation on the primary transformation model based on the augmented Lagrangian function to obtain a secondary transformation model. Among them, k is the secondary transformation model parameter; A solution module, used for solving the secondary transformation model by using an alternating direction method given the algorithm parameters, to obtain the radar super-resolution image; A preprocessing unit, used for performing image preprocessing on the radar super-resolution image to obtain a preprocessed image; The extraction unit is used to extract the target length feature and the target shape feature according to the preprocessed image.
Citation Information
Patent Citations
Target micro-Doppler clustering estimation method of forward and backward TVAR model
CN113960539A
Apparatus, method, and system for providing super-resolution target estimation via atomic norm optimization
EP3933432A1
System and Method for Radar Imaging Using Distributed Arrays and Compressive Sensing
US20170010352A1
