A method and system for extracting targets from SAR images

Abstract

The application discloses a SAR image target extraction method and system, and relates to the field of SAR image processing.The method comprises the following steps: performing singular value decomposition based on a SAR image to obtain singular vectors for expressing target structure characteristics; fitting a broken line singular vector by using the singular vectors; screening target-related broken line singular vectors by using peak-valley triangular area characteristics of the broken line singular vector; and reconstructing the target.The application realizes SAR image target extraction by means of singular vectors on the basis of singular value decomposition, can simultaneously realize target extraction and noise removal, does not involve conventional processing means such as filter denoising and edge detection, avoids the problems of image edge blurring caused by filter denoising and great influence of noise on edge detection in the conventional processing means, and does not exist a sliding window search process, so that the execution efficiency of the algorithm is improved.

Description

Technical Field

[0001] This invention relates to the field of SAR image processing, and more specifically, to a method and system for SAR image target extraction. Background Technology

[0002] SAR sensors are observation devices capable of operating around the clock and in all weather conditions. They are widely used in Earth and sea observation, and are carried by platforms including satellites, aircraft, and drones. The output information is SAR images, and extracting and identifying targets in these images is the ultimate goal of SAR sensor applications. Existing SAR image target extraction methods are mainly based on gray-level correlation features and generally require processing steps such as sliding window search, filtering and denoising, and edge detection. Existing technologies have the following shortcomings: 1. Ambiguity of energy-related features: Gray-level correlation features are the main features relied upon for SAR image processing, belonging to energy-related features, and do not necessarily correspond to targets. When noise energy is strong and target energy is weak, using gray-level correlation features may result in target loss. 2. Limitations of sliding window search: Sliding window search is a local processing method. The window size and features are predetermined, but the window characteristics may not match the target characteristics. An overly large window may mask local features, while an overly small window may add false features. Mismatched window features can lead to similar results. Furthermore, the sliding window search process is cumbersome and the algorithm's execution efficiency is low. 3. The contradiction between filtering and edge detection: Filtering and denoising can smooth the image, suppressing noise but also weakening or eliminating abrupt changes, making edge detection more difficult. On the other hand, direct edge detection preserves abrupt changes but also retains noise, increasing the error in differential operations during edge detection. Summary of the Invention

[0003] To address the aforementioned problems, the present invention aims to provide a novel SAR image target extraction technique, which addresses related technical issues in the field of target extraction within SAR image processing.

[0004] To achieve the above technical objectives, this application provides a SAR image target extraction method, comprising the following steps: Singular value decomposition is performed on SAR images to obtain singular vectors that represent the structural features of the target. Using singular vectors, fit to obtain the polyline singular vector; By utilizing the area characteristics of peak-valley triangles in the singular direction of the broken line, target-related singular vectors are screened for target reconstruction.

[0005] Preferably, when obtaining singular vectors, the horizontal structural features of the target are expressed by the left singular vector, and the vertical structural features of the target are expressed by the right singular vector, wherein the left singular vector comes from the Gram matrix XX. TOrthogonalization of eigenvectors, with right singular vectors derived from Gram matrix X. T The orthogonalization of the X feature vectors gives the SAR image matrix as X.

[0006] Preferably, when obtaining the piecewise singular vector, the endpoints of the piecewise fitting of the singular vector are determined by searching for peaks and valleys, and linear least squares fitting is used between adjacent peaks and valleys to obtain the piecewise singular vector. Alternatively, nonlinear least squares fitting is used to obtain the piecewise singular vector.

[0007] Preferably, when screening target-related singular vectors, the area of ​​the peak-valley triangle of the singular vector is used as a threshold to determine whether the singular vector is related to the target.

[0008] Preferably, when setting the threshold, the mean and standard deviation of the area of ​​the peak-valley triangle of all singular vectors of the broken line are statistically analyzed, and the mean plus the standard deviation is used as the threshold.

[0009] This invention also discloses a SAR image target extraction system, comprising: The singular value decomposition module is used to perform singular value decomposition based on SAR images to obtain singular vectors that represent the structural features of the target. The singular vector fitting module is used to fit and obtain a piecewise singular vector using singular vectors. The target reconstruction module is used to filter target-related singular vectors by utilizing the peak-valley triangle area feature of the polyline singular vectors and then reconstruct the target.

[0010] Preferably, the singular value decomposition module is further configured to express the horizontal structural features of the target using a left singular vector, and to express the vertical structural features of the target using a right singular vector, wherein the left singular vector is derived from the Gram matrix XX. T Orthogonalization of eigenvectors, with right singular vectors derived from Gram matrix X. T The orthogonalization of the X feature vectors gives the SAR image matrix as X.

[0011] Preferably, the singular vector fitting module uses linear least squares fitting or nonlinear least squares fitting to obtain a piecewise singular vector. Specifically, the endpoints of the piecewise fitting of the singular vector are identified by searching for peaks and valleys, and linear least squares fitting is used between adjacent peaks and valleys to obtain the piecewise singular vector; or, based on the singular vector, nonlinear least squares fitting is used to obtain the piecewise singular vector.

[0012] Preferably, the target reconstruction module is further used to determine whether the polyline singular vector is related to the target by using the area of ​​the peak-valley triangle of the polyline singular vector as a threshold. Specifically, the mean and standard deviation of the area of ​​the peak-valley triangle of all polyline singular vectors are statistically analyzed, and the mean plus the standard deviation is used as the threshold.

[0013] The present invention discloses the following technical effects: This invention captures both the energy and structural features of the target, enabling a more comprehensive reflection of target information. By fitting singular vectors using piecewise linear least squares, it can simultaneously remove noise and retain the target. Target extraction is achieved by reconstructing the target using target-related singular vectors, allowing for more accurate utilization of effective information. It is applicable to the target extraction field in SAR image processing.

[0014] This invention proposes to extract SAR images by using singular vectors based on singular value decomposition. It can simultaneously achieve target extraction and noise removal without involving conventional processing methods such as filtering and edge detection. This avoids the problems of blurred image edges caused by filtering and noise removal and edge detection being greatly affected by noise. There is no sliding window search process, which improves the execution efficiency of the algorithm. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 These are comparative illustrations showing the effects of image extraction as described in this invention; Figure 2 This is a schematic diagram of the method described in this invention, where i represents the ordinal number of the left singular vector and j represents the ordinal number of the right singular vector; Figure 3 This is a schematic diagram of the peak-valley triangle of the singular vector of the broken line described in this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0018] like Figures 1-3As shown, the present invention provides a SAR image target extraction method, comprising the following steps: 1. Singular Vector Set Generation: The SAR image is treated as a matrix. After singular value decomposition, a left singular vector matrix and a right singular vector matrix are obtained. All the singular vectors contained therein constitute the singular vector set, which is used as the object for singular vector selection.

[0019] 2. Piecewise linear least squares fitting of singular vectors: In the singular vectors, the target corresponds to the large-scale change part, and the noise corresponds to the small-scale change part. By using piecewise linear least squares fitting to remove noise and retain the target, a piecewise singular vector is obtained.

[0020] 3. Target-related polyline singular vector screening: Polyline singular vector features belong to structurally related features. When a target exists, the polyline singular vector shows a significant peak spike. By setting a peak-valley triangle area threshold, target-related polyline singular vectors are screened from the polyline singular vector set.

[0021] 4. Target Reconstruction and Target Extraction: In the set of singular vectors related to the target, those at the top indicate stronger energy features, and the appearance of consecutive sub-sub-vectors indicates concentrated structural features. By comprehensively considering energy and structural features, the first few consecutive target-related singular vectors are selected to reconstruct the target and achieve target extraction.

[0022] Specifically, the present invention includes the following design in the generation of singular vector sets: Let the SAR image matrix be X, and the singular value decomposition form of X be:

[0023] Where U is the left singular matrix, S is the singular value matrix, V is the right singular matrix, u is the left singular vector, s is the singular value, v is the right singular vector, and T denotes the matrix transpose.

[0024] The left singular vector u comes from the Gram matrix XX T The orthogonalization of eigenvectors expresses the horizontal structural features of the target; the right singular vector v comes from the Gram matrix X. T The orthogonalization of the X eigenvectors represents the target's structural features in the vertical direction. Both the left singular vector u and the right singular vector v contain target features, and together they form the singular vector set C. ; Singular vector feature analysis: In SAR images, the horizontal features of a target are contained in the row vectors of the image matrix X, and the Gram matrix XX T The elements are the inner product of the X row vectors. When the target's horizontal features are significant, they will appear in the Gram matrix XX. TLocal maxima are formed on the main diagonal. The vertical features of the target are contained in the column vectors of the image matrix X, and the Gram matrix X... T The elements of X are the inner products of the column vectors of X. When the target's vertical features are significant, they will appear in the Gram matrix X. T A local maximum is formed on the main diagonal of X.

[0025] Using eigenvalue decomposition, the Gram matrix XX T It is decomposed into the sum of m sub-matrices. If the target exists and its horizontal features are prominent, the Gram matrix XX T The main diagonal will form local maxima, and the main diagonals of several submatrices will also form local maxima. The elements of the corresponding singular vector u will also exhibit local maxima. If the target does not exist or the horizontal features are not obvious, the elements of all submatrices will exhibit an irregular random distribution, and the elements of the corresponding singular vector u will also exhibit an irregular random distribution.

[0026] Using eigenvalue decomposition, the Gram matrix X is... T X is decomposed into the sum of n sub-matrices. If the target exists and its vertical features are prominent, the main diagonal of the Gram matrix XTX will form a local maximum, and the main diagonals of several sub-matrices will also form local maxima. The elements of the corresponding singular vector v will also exhibit local maxima. If the target does not exist or its vertical features are not obvious, the elements of all sub-matrices will exhibit an irregular random distribution, and the elements of the corresponding singular vector v will also exhibit an irregular random distribution.

[0027] Singular vector piecewise linear least squares fitting: Targets correspond to spikes in singular vectors, while noise corresponds to random distributions in singular vectors. Since the presence or structure of a target is unknown beforehand, any spikes in the singular vectors could be target information and must be retained before the target information is confirmed. This necessitates searching for all spike features in the singular vectors. Considering the need to retain spikes, remove noise, and simplify computation, piecewise linear least squares fitting of the singular vectors is chosen. The endpoints of the piecewise fitting of the singular vectors are determined by searching for peaks and valleys. Linear least squares fitting is then applied between adjacent peaks and valleys, and the resulting broken line replaces the original singular vector, serving as the object for filtering target-related singular vectors.

[0028] Target-related singular vector filtering: After piecewise linear least squares fitting of the singular vectors, the original set of singular vectors is transformed into a set of piecewise singular vectors. Some piecewise singular vectors clearly express the target information, while others do not. The mean and standard deviation of the areas of the peak-valley triangles of all piecewise singular vectors are calculated. Using the mean plus the standard deviation as a threshold, piecewise singular vectors exceeding the threshold are selected as target-related singular vectors. Target-related singular vectors may be left or right singular vectors; the union of their indices is used to reconstruct the target singular vector index set.

[0029] The specific algorithm steps are as follows: (1) Search the set of left singular vectors {u f1 ,u f2 ,…,u fn}, m>n, count the singular vectors whose spikes exceed the detection threshold, and record their indices to generate a set. ; Among them, u f Let represent the left singular vector after piecewise linear least squares fitting, and p represent the total number of singular vectors exceeding the threshold.

[0030] (2) Search the set of right singular vectors {v f1 ,v f2 ,…,v fn}, m>n, count the singular vectors whose spikes exceed the detection threshold, and record their indices to generate a set. ; Among them, v f Let represent the right singular vector after piecewise linear least squares fitting, and q represent the total number of singular vectors exceeding the threshold.

[0031] (3) Find the union of sets M and N, and generate the set of all indices.

[0032] Where r represents the total number of indexes.

[0033] The singular vectors uf and vf, whose indices are elements of set R, are the target reconstructed singular vectors.

[0034] Target Reconstruction and Target Extraction: If the index set R is non-empty, select the first consecutive indices to form a subset P, and then use the polyline singular vector u according to the index subset P. f and v f When reconstructing the target, only singular vector information is needed, not singular value information.

[0035] ; If the index set R is empty, it means that the SAR image has no target, and the SAR image is discarded.

[0036] For targets in the reconstructed SAR image, target extraction is achieved by setting a threshold, which can be selected as the standard deviation of the reconstructed image. The simulation results of SAR image target extraction are as follows: Figure 1 As shown.

[0037] This invention optimizes the target feature selection strategy: singular values ​​represent energy-related features, and singular vectors represent structure-related features. Using either one alone is insufficient to express the target information. Using both singular value features and singular vector features simultaneously can more comprehensively describe the target information, which is a more optimized target feature selection strategy.

[0038] This invention avoids the contradiction between noise removal and target extraction: target extraction is achieved through singular vector fitting, filtering, and reconstruction, and noise removal is achieved during the fitting stage. It does not involve filtering and edge detection processes, thus avoiding the contradiction between the two.

[0039] This invention improves the efficiency of the algorithm: it utilizes the decomposition and reconstruction of SAR images to achieve target extraction, eliminating the need for a sliding window search process and thus improving the efficiency of the algorithm.

[0040] In summary, this invention obtains a set of singular vectors through singular value decomposition, uses piecewise linear least squares to fit the singular vectors to remove noise and retain the target, selects target-related piecewise linear singular vectors based on the peak-valley triangle area characteristics of the piecewise linear singular vectors, and reconstructs the target by taking the first few consecutive target-related piecewise linear singular vectors with the highest index to achieve target extraction.

[0041] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0042] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0043] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for target extraction from SAR images, characterized in that, Includes the following steps: Singular value decomposition is performed on SAR images to obtain singular vectors that represent the structural features of the target. Using the singular vectors, a piecewise singular vector is obtained through fitting. By utilizing the peak-valley triangle area feature of the polyline singular vector, target-related polyline singular vectors are screened for target reconstruction.

2. The SAR image target extraction method according to claim 1, characterized in that: When obtaining singular vectors, the horizontal structural features of the target are expressed by the left singular vector, and the vertical structural features are expressed by the right singular vector. The left singular vector is derived from the Gram matrix XX. T Orthogonalization of eigenvectors, with right singular vectors derived from Gram matrix X. T The orthogonalization of the X feature vectors gives the SAR image matrix as X.

3. The SAR image target extraction method according to claim 1, characterized in that: When obtaining the piecewise singular vector, the endpoints of the piecewise fitting of the singular vector are identified by searching for peaks and valleys. Linear least squares fitting is then used between adjacent peaks and valleys to obtain the piecewise singular vector.

4. The SAR image target extraction method according to claim 1, characterized in that: When obtaining the polyline singular vector, a nonlinear least squares fitting is used based on the singular vector to obtain the polyline singular vector.

5. The SAR image target extraction method according to claim 3, characterized in that: When filtering target-related singular vectors, the area of ​​the peak-valley triangle of the singular vector is used as a threshold to determine whether the singular vector is related to the target.

6. The SAR image target extraction method according to claim 5, characterized in that: When setting the threshold, the mean and standard deviation of the area of ​​the peak-valley triangle of all singular vectors in the broken line are calculated, and the mean plus the standard deviation is used as the threshold.

7. A SAR image target extraction system, characterized in that, include: The singular value decomposition module is used to perform singular value decomposition based on SAR images to obtain singular vectors that represent the structural features of the target. The singular vector fitting module is used to fit and obtain a piecewise singular vector using the singular vector; The target reconstruction module is used to filter target-related singular vectors by utilizing the peak-valley triangle area feature of the polyline singular vectors and then reconstruct the target.

8. The SAR image target extraction system according to claim 7, characterized in that: The singular value decomposition module is further configured to express the horizontal structural features of the target using a left singular vector, and to express the vertical structural features of the target using a right singular vector, wherein the left singular vector is derived from the Gram matrix XX. T Orthogonalization of eigenvectors, with right singular vectors derived from Gram matrix X. T The orthogonalization of the X feature vectors gives the SAR image matrix as X.

9. The SAR image target extraction system according to claim 8, characterized in that: The singular vector fitting module uses linear least squares fitting or nonlinear least squares fitting to obtain a piecewise singular vector. Specifically, the endpoints of the piecewise fitting of the singular vector are identified by searching for peaks and valleys, and linear least squares fitting is used between adjacent peaks and valleys to obtain the piecewise singular vector; or, based on the singular vector, nonlinear least squares fitting is used to obtain the piecewise singular vector.

10. The SAR image target extraction system according to claim 9, characterized in that: The target reconstruction module is also used to determine whether the polyline singular vector is related to the target by using the area of ​​the peak-valley triangle of the polyline singular vector as a threshold. Specifically, the mean and standard deviation of the area of ​​the peak-valley triangle of all polyline singular vectors are statistically analyzed, and the mean plus the standard deviation is used as the threshold.

Images