A radar target detection method in sea clutter background based on alternating direction method

Abstract

The present invention discloses a radar target detection method in a sea clutter background based on the alternating direction method (ADM). The method comprises the following steps: selecting two radar image frames for normalization; reorganizing the normalized two radar image frames into multiple matrices by column; performing foreground and background separation on each reorganized matrix based on the alternating direction multiplier method, decomposing the matrix into a low-rank matrix and a sparse matrix, wherein the low-rank matrix corresponds to the sea clutter component and the sparse matrix corresponds to the target of interest component; setting an ADM iteration stop condition for each reorganized matrix; arranging the separated low-rank matrices by column to form a restoration matrix, and detecting the sea clutter result; and arranging the separated sparse matrices by column to form a restoration matrix, and detecting the target of interest result. The present invention reduces the difficulty of the hyperparameter selection problem faced by applying matrix separation technology to clutter suppression, effectively achieving target extraction and clutter suppression.

Description

Technical Field

[0001] The present invention relates to the technical field of radar target detection, and in particular to a radar target detection method under a sea clutter background based on an alternating direction method. Background Art

[0002] The tasks of developing, utilizing, protecting, and managing the ocean are becoming increasingly arduous. Successfully completing these tasks relies on information obtained from various ocean monitoring radar systems. However, the non-uniform, non-stationary, and non-Gaussian characteristics of sea surface clutter make it prone to producing extremely strong echoes (wave spikes). These strong echoes are very similar to the characteristics of target echoes. If they are not effectively filtered out, they will cause a large number of false alarms. Therefore, ocean monitoring radars must have strong sea clutter suppression capabilities to better detect various targets. Therefore, how to effectively separate the clutter background from targets of interest at sea based on actual application scenarios and measure data to improve the target detection performance of marine radars is of great practical value.

[0003] Currently, traditional physical methods are widely used to study the statistical characteristics of sea clutter, and a series of targeted clutter suppression techniques, such as generalized likelihood ratio detectors and adaptive matched filtering, have been developed. However, sea clutter is a naturally occurring object that appears to be a random process, but its essential characteristics cannot be fully described by statistical models. Furthermore, the nonlinear random variations of sea clutter, influenced by numerous factors such as the environment and radar equipment parameters, lead to problems with existing sea clutter filtering methods based on statistical models, such as long filter response times and insufficient clutter suppression capabilities. Summary of the Invention

[0004] In order to overcome the defects and shortcomings of the existing technology, the present invention provides a radar target detection method under sea clutter background based on the alternating direction method, which realizes sea clutter suppression and target extraction of interest. From the perspective of image processing, the present invention transforms the problem of separating clutter and targets in ocean monitoring radar images into a matrix decomposition problem. In view of the low rank characteristics of the sea clutter background and the sparse characteristics of the moving targets, the alternating direction method (ADM) is adopted to directly regard the two parts with different characteristics (i.e., clutter and target) as two variables. Through alternating iteration, the foreground (target) and background (clutter) of the radar image under the sea clutter background are effectively separated. The technology overcomes the uncertainty problem caused by the statistical characteristics of sea clutter and the difficulty in analyzing complex detail features. At the same time, it fully exploits the low rank of sea clutter and the sparsity of moving targets in the radar image, and directly solves the sparse component (target) and low rank component (clutter) of the radar image through alternating iteration. This reduces the difficulty of hyperparameter selection problem faced by applying matrix separation technology to clutter suppression, facilitates the rapid and effective realization of target extraction and clutter suppression, and lays a good foundation for subsequent target feature extraction.

[0005] A second object of the present invention is to provide a radar target detection system in a sea clutter background based on an alternating direction method.

[0006] A third object of the present invention is to provide a storage medium.

[0007] A fourth object of the present invention is to provide a computing device.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] The present invention provides a radar target detection method in a sea clutter background based on an alternating direction method, comprising the following steps:

[0010] Select two frames of radar images for normalization;

[0011] Reorganize the two normalized radar image frames into multiple matrices by column;

[0012] Based on the alternating direction multiplication method, each reorganized matrix is ​​separated into a foreground and background, and decomposed into a low-rank matrix and a sparse matrix, wherein the low-rank matrix corresponds to the sea clutter component and the sparse matrix corresponds to the target of interest component;

[0013] Set the ADM iteration stop condition for each reorganized matrix;

[0014] Arrange the separated low-rank matrix into a recovery matrix by columns and detect the sea clutter results;

[0015] The separated sparse matrix is ​​arranged column by column into a recovery matrix to detect the target results of interest.

[0016] As a preferred technical solution, two frames of radar images are selected for normalization, and two consecutive frames or two frames of images with a preset interval are selected.

[0017] As a preferred technical solution, the normalization formula is expressed as:

[0018]

[0019] Among them, Ymax represents the maximum pixel value, Ymin represents the minimum pixel value, and Y i Represents the i-th frame image.

[0020] As a preferred technical solution, the low-rank matrix is ​​specifically expressed as:

[0021]

[0022] The sparse matrix is ​​represented as:

[0023]

[0024] in, The subscript k is the number of iterations, Respectively represent the approximate values ​​of the low-rank part L and the sparse part S of the k+1th M, is the singular value threshold operator, is a soft threshold operator.

[0025] As a preferred technical solution, the update formula of the soft threshold operator is:

[0026]

[0027]

[0028] The update formula of the singular value threshold operator is:

[0029]

[0030]

[0031]

[0032] Among them, x ij is the element in the i-th row and j-th column of the matrix X.

[0033] As a preferred technical solution, the iterative formula is specifically expressed as:

[0034]

[0035] Among them, 0<ε<1,ε=10 -8 .

[0036] In order to achieve the above second purpose, the present invention adopts the following technical solutions:

[0037] A radar target detection system under sea clutter background based on alternating direction method, comprising: a normalization module, a matrix reorganization module, a separation module, an iterative condition setting module and a restoration module;

[0038] The normalization module is used to select two frames of radar images for normalization;

[0039] The matrix reorganization module is used to reorganize the two normalized radar image frames into multiple matrices by column;

[0040] The separation module is used to perform foreground and background separation on each reorganized matrix based on the alternating direction multiplication method, and decompose it into a low-rank matrix and a sparse matrix, wherein the low-rank matrix corresponds to the sea clutter component and the sparse matrix corresponds to the target of interest component;

[0041] The iteration condition setting module is used to set the ADM iteration stop condition of each reorganized matrix;

[0042] The restoration module is used to arrange the separated low-rank matrix into a restoration matrix in columns to detect sea clutter results, and to arrange the separated sparse matrix into a restoration matrix in columns to detect target results of interest.

[0043] In order to achieve the third purpose above, the present invention adopts the following technical solutions:

[0044] A storage medium stores a program, which, when executed by a processor, implements the above-mentioned radar target detection method in a sea clutter background based on the alternating direction method.

[0045] In order to achieve the fourth purpose above, the present invention adopts the following technical solutions:

[0046] A computing device includes a processor and a memory for storing a program executable by the processor. When the processor executes the program stored in the memory, the radar target detection method in a sea clutter background based on the alternating direction method is implemented.

[0047] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0048] (1) The present invention applies the alternating direction method to the field of ocean detection radar clutter suppression, and utilizes the low rank of clutter and the sparsity of target signals to separate clutter from targets, avoiding the description of complex details and features of sea clutter. It solves the problem that the actual clutter environment usually does not conform to the uniform clutter assumption, overcomes the disadvantage of the uniform Gaussian clutter background of the generalized likelihood ratio detector, and has stronger applicability.

[0049] (2) The present invention adopts the ADM algorithm to solve the matrix separation problem, converts the sea clutter low-rank matrix and the target sparse matrix into two variables for alternating iteration, and can extract the target of interest while suppressing the sea clutter, rather than just suppressing the sea clutter. Compared with the solution using the accelerated proximal gradient method (APG) algorithm, the use of the ADM algorithm does not require an internal loop for each parameter, overcomes the problems of the APG with many intermediate parameters and slow running speed, and improves the speed and accuracy of matrix separation. The present application has innovations in the application scenarios of matrix separation, the development of sea clutter suppression technology and target extraction technology, and is an efficient sea clutter suppression method. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 Schematic diagram of the framework of the radar target detection method in sea clutter background based on the alternating direction method of the present invention;

[0051] Figure 2Schematic diagram of the implementation process of the radar target detection method under sea clutter background based on the alternating direction method of the present invention;

[0052] Figure 3 This is a schematic diagram of the clutter suppression effect of a real ocean monitoring radar image according to the present invention;

[0053] Figure 4 for Figure 3 A partial enlarged schematic diagram. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0055] Example 1

[0056] like Figure 1 、 Figure 2 As shown, this embodiment provides a radar target detection method under sea clutter background based on the alternating direction method. In order to effectively suppress sea clutter and extract the target, this embodiment considers decomposing a matrix M reconstructed from a radar image into a low-rank matrix part L (background clutter) and a sparse matrix part S (target of interest), and M, L, Where m×2 means the matrix has m rows and 2 columns. The problem can be modeled as the following optimization problem:

[0057]

[0058] stL+S=M

[0059] Here, rank(·) represents the rank of the matrix, and ||·||0 represents the zero norm, that is, the number of non-zero elements in the corresponding vector or matrix.

[0060] Since rank(L) and ||S||0 are both nonlinear and non-convex, it is difficult to optimize them in practice. Therefore, the above problem can be relaxed into the following optimization model, which is also called robust principal component analysis (RPCA):

[0061]

[0062] stL+S=M

[0063] Among them, ||·|| * Represents the nuclear norm, which can be expressed by the sum of r singular values ​​σ, that is, The (1,1) norm is expressed as but When the singular values ​​of L are reasonably distributed and the non-zero elements of the sparse matrix are evenly distributed, the above model can recover the original low-rank matrix with a probability close to 1.

[0064] This embodiment uses matrix separation to solve the problem of radar target extraction in a sea clutter background, avoiding the discussion of the complex details of sea clutter. It can extract targets of interest while suppressing sea clutter, rather than just suppressing sea clutter. This reduces the difficulty of hyperparameter selection faced by applying matrix separation technology to clutter suppression, and is conducive to quickly and effectively achieving target extraction and clutter suppression.

[0065] The specific steps include:

[0066] S1: First, two consecutive radar images Normalize (here m represents the number of rows of the radar image, and n represents the number of columns of the radar image);

[0067] In this embodiment, since the measured radar image data contains a large number of targets to be measured and their positions change at all times, two consecutive frames can be directly selected. If the targets to be measured between the two frames are approximately stationary, two frames of images with an interval greater than 30 can also be selected.

[0068] The two frames of images in this embodiment are Y1 and Y2, which are normalized. The normalization formula is:

[0069]

[0070] In the formula, taking the i-th frame image as an example, its maximum pixel value is defined as Ymax and its minimum pixel value is defined as Ymin.

[0071] S2: Take the jth column from each of the normalized images Y′1 and Y′2 and Merge into a new matrix M j ,and Where j = 1, 2, 3...n. For example, M1 is composed of the first column vectors of Y'1 and Y'2 in sequence, M2 is composed of the second column vectors of Y'1 and Y'2 in sequence, and so on. Then M n It is composed of the nth column vectors of Y′1 and Y′2 in sequence.

[0072] S3: Use the alternating direction multiplication method to separate the foreground and background of each M matrix, and decompose it into a low-rank matrix L (background clutter) and a sparse matrix S (target of interest), and L, That is, each M j Perform ADM decomposition to obtain the corresponding L j and S j .

[0073] Taking the jth M matrix as an example, use ADM to solve the following model:

[0074]

[0075] stL+S=M

[0076] The specific ADM solution steps are:

[0077] S31: First, find the approximate value of the low-rank matrix L, which corresponds to the clutter component of the radar image:

[0078]

[0079] S32: Calculate the approximate value of the sparse matrix S, which corresponds to the target component of the radar image:

[0080]

[0081] S33: Note that the matrix W is required in both steps S31 and S32. k , the transition matrix W is given below k Iterative solution method:

[0082]

[0083] Wherein, subscript k is the number of iterations, and They represent the approximate values ​​of the low-rank part L and the sparse part S of the k+1th M respectively. The initial values ​​of L, S, and W are all zeros(m,2). ρ k =uλ, u=10. is the singular value threshold operator, is a soft threshold operator.

[0084] make Then the soft threshold operator [S t (X)] k The update formula is as follows:

[0085] x ij is the element in the i-th row and j-th column of the matrix X

[0086] make Then the singular value threshold operator The update formula is as follows:

[0087] ① First perform singular value decomposition on Y,

[0088] [U, ∑, V] = svd(Y)

[0089] ② Then let the above [S t (X)] k in X=∑, then:

[0090]

[0091] S34: For each M j The ADM iteration stopping condition is:

[0092] (1) Maximum number of iterations p, p = 200;

[0093] (2) Satisfy the following formula:

[0094]

[0095] Where 0<ε<1,ε=10 -8 , that is, we get j Corresponding and Repeat for M j+1 Perform S31, S32, S33, and S34 until j>n is satisfied, then stop the entire iteration process to obtain n groups. and

[0096] S4: Then extract the corresponding columns from the separated low-rank matrix L and arrange them in sequence to form a restoration matrix. The restored image of Y1 is composed of the first columns of n L matrices arranged in sequence, and the restored image of Y2 is composed of the second columns of n L matrices arranged in sequence.

[0097] Specifically expressed as:

[0098] n The i-th column is arranged in sequence to form a new matrix A i , the restored image A1 of Y1 is composed of n The first column of the matrix is ​​arranged in sequence, and the restored image A2 of Y2 is composed of n The second column of the matrix is ​​arranged sequentially.

[0099] The present invention uses ADM technology to solve the RPCA problem of target detection in a sea clutter background. Compared with the existing method of solving RPCA using the accelerated proximal gradient method (APG), this method requires fewer parameters to be iterated (APG requires 5 iterations each time, while ADM only requires 3), can achieve higher accuracy, and requires only lower storage space. At the same time, the ADM technology does not require internal loops for each parameter, and is faster.

[0100] As shown in Table 1 below, for two frames of 3001*601 radar images, APG and ADM are iterated 50 times respectively. Taking the first frame of the radar image with a rank of 601 as an example, a detailed comparison of the performance of APG and ADM is obtained:

[0101] Table 1 APG and ADM performance comparison

[0102] algorithm Accuracy rank 0 norm Number of iterations Runtime APG 0.050420 601 1453408 50 5.26s ADM 0.004055 601 919260 50 3.79s

[0103] In order to illustrate the timeliness and adaptability of this embodiment, Figure 3-Figure 4 As shown in the figure, the experiment was conducted using actual sea surface echo scattering data. From left to right in the figure are the original image, foreground image, and background image. The two radar images are target position and trajectory images with the same date. Figure 3 、 Figure 4 The results of ADM processing on two 3001*601 radar images clearly show that the sea clutter is filtered out and the target outline is clear.

[0104] Example 2

[0105] This embodiment provides a radar target detection system under sea clutter background based on the alternating direction method, comprising: a normalization module, a matrix reorganization module, a separation module, an iteration condition setting module and a restoration module;

[0106] In this embodiment, the normalization module is used to select two frames of radar images for normalization;

[0107] In this embodiment, the matrix reorganization module is used to reorganize the two normalized radar image frames into multiple matrices by column;

[0108] In this embodiment, the separation module is used to perform foreground and background separation on each reorganized matrix based on the alternating direction multiplier method, decomposing it into a low-rank matrix and a sparse matrix, wherein the low-rank matrix corresponds to the sea clutter component and the sparse matrix corresponds to the target of interest component;

[0109] In this embodiment, the iteration condition setting module is used to set the ADM iteration stop condition of each reorganized matrix;

[0110] In this embodiment, the restoration module is used to arrange the separated low-rank matrix into a restoration matrix in columns to detect the sea clutter results, and to arrange the separated sparse matrix into a restoration matrix in columns to detect the target of interest results.

[0111] Example 3

[0112] This embodiment provides a storage medium, which may be a ROM, RAM, magnetic disk, optical disk, or other storage medium. The storage medium stores one or more programs. When the programs are executed by a processor, the radar target detection method in a sea clutter background based on the alternating direction method of embodiment 1 is implemented.

[0113] Example 4

[0114] This embodiment provides a computing device, which may be a desktop computer, a laptop computer, a smart phone, a PDA handheld terminal, a tablet computer, or other terminal device with a display function. The computing device includes a processor and a memory, wherein the memory stores one or more programs. When the processor executes the programs stored in the memory, the radar target detection method in a sea clutter background based on the alternating direction method of embodiment 1 is implemented.

[0115] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A radar target detection method in sea clutter background based on alternating direction method, characterized in that: The steps include: Select two frames of radar images for normalization; Reorganize the two normalized radar images into multiple matrices by column, and take the jth column from each of the normalized images Y′1 and Y′2 and Merge into a new matrix M j , where j = 1, 2, 3...n, n represents the number of columns of the radar image; Based on the alternating direction multiplication method, each reorganized matrix is ​​separated into a foreground and background, and decomposed into a low-rank matrix and a sparse matrix, wherein the low-rank matrix corresponds to the sea clutter component and the sparse matrix corresponds to the target of interest component; The ADM algorithm is used to solve the matrix separation problem, and the sea clutter low-rank matrix and the target sparse matrix are converted into two variables for alternating iteration. Use ADM to solve the following model: stL+S=M The low-rank matrix is ​​specifically expressed as: The sparse matrix is ​​represented as: in, The subscript k is the number of iterations, Respectively represent the approximate values ​​of the low-rank part L and the sparse part s of the k+1th M, is the singular value threshold operator, is the soft threshold operator; Set the ADM iteration stop condition for each reorganized matrix; The iterative formula is specifically expressed as: Where, 0<ε<1,ε=10 -8 ; Arrange the separated low-rank matrix into a recovery matrix by columns and detect the sea clutter results; Extract the corresponding columns from the separated low-rank matrix L and arrange them in sequence to form a restoration matrix. The restored image of Y1 is composed of the first columns of n L matrices arranged in sequence, and the restored image of Y2 is composed of the second columns of n L matrices arranged in sequence. Y1 and Y2 represent the two selected radar images. The separated sparse matrix is ​​arranged column by column into a recovery matrix to detect the target results of interest.

2. The radar target detection method in sea clutter background based on alternating direction method according to claim 1 is characterized in that: The two frames of radar images are selected for normalization, and two consecutive frames or two frames of images with a preset interval are selected.

3. The radar target detection method in sea clutter background based on alternating direction method according to claim 1 or 2, wherein the normalization formula is expressed as: in, Ymax represents the maximum pixel value, Ymin represents the minimum pixel value, i Represents the i-th frame image.

4. The radar target detection method in sea clutter background based on alternating direction method according to claim 1, characterized in that: The update formula of the soft threshold operator is: The update formula of the singular value threshold operator is: Among them, x ij is the element in the i-th row and j-th column of the matrix X.

5. A radar target detection system in sea clutter background based on alternating direction method, characterized in that: Used to implement the radar target detection method under sea clutter background based on the alternating direction method as described in claims 1-4, comprising: a normalization module, a matrix reorganization module, a separation module, an iterative condition setting module and a restoration module; The normalization module is used to select two frames of radar images for normalization; The matrix reorganization module is used to reorganize the two normalized radar image frames into multiple matrices by column; The separation module is used to perform foreground and background separation on each reorganized matrix based on the alternating direction multiplication method, and decompose it into a low-rank matrix and a sparse matrix, wherein the low-rank matrix corresponds to the sea clutter component and the sparse matrix corresponds to the target of interest component; The iteration condition setting module is used to set the ADM iteration stop condition of each reorganized matrix; The restoration module is used to arrange the separated low-rank matrix into a restoration matrix in columns to detect sea clutter results, and to arrange the separated sparse matrix into a restoration matrix in columns to detect target results of interest.

6. A storage medium storing a program, characterized in that: When the program is executed by a processor, the radar target detection method in a sea clutter background based on the alternating direction method as described in any one of claims 1 to 4 is implemented.

7. A computing device comprising a processor and a memory for storing a program executable by the processor, characterized in that When the processor executes the program stored in the memory, the radar target detection method in a sea clutter background based on the alternating direction method as claimed in any one of claims 1 to 4 is implemented.

Images