SAR moving target extraction method based on maxtree pre-segmentation method
By using the Maxtree pre-segmentation method to process SAR moving target images, the problem of extracting peak values using Radon transform is solved, and the moving targets can be extracted one by one in the case of multi-target aliasing, thus improving the accuracy and clarity of the extraction.
Patent Information
- Application Number
- CN202410532706.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-30
- Publication Date
- 2026-01-02
- Estimated Expiration
- 2044-04-30
AI Technical Summary
In existing technologies, moving target images in the range compression-azimuth time domain are difficult to extract peak values from Radon transform due to noise and aliasing of multiple target signals, making it impossible to extract moving targets one by one.
The Maxtree-based pre-segmentation method is adopted. By performing a Decirp operation on the range-compression-azimuth time domain signal, a Maxtree is constructed and filtered to obtain the pre-segmentation result. Then, it is converted to the range-compression-azimuth time domain, and Radon transform is used to extract individual moving targets and perform clustering and superposition.
In cases of multiple targets overlapping, it can extract moving targets one by one, improving the accuracy and clarity of extraction and reducing noise interference.
Smart Images
Figure CN118465715B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of moving target extraction method, in particular to a SAR moving target extraction method based on Maxtree pre-segmentation method. BACKGROUND
[0002] SAR moving target detection has been widely used in military and civilian fields, and the main challenge of detection is how to accurately extract moving targets from ground clutter. At present, the common methods are mainly divided into single-channel SAR moving target detection and multi-channel SAR moving target detection.
[0003] Single-channel SAR moving target detection algorithm is mainly divided into two categories: echo data domain-based method and image domain-based method. Among them, the detection method based on echo data domain mainly uses the phase change or Doppler shift of echo signal caused by target motion for detection. While the detection method based on image domain is to detect moving targets by observing the change of pixel intensity of moving targets in different sub-images of SAR image.
[0004] Multi-channel SAR moving target detection algorithm mainly includes the following: offset phase center antenna (DPCA), space-time adaptive processing (STAP), along-track interferometry (ATI) and detection method based on moving target shadow, etc.
[0005] Disadvantages of prior art
[0006] Single-channel SAR moving target detection method is easily disturbed by clutter and noise, and its clutter suppression effect has limitations, which leads to poor detection performance for slow and weak targets. In contrast, although multi-channel SAR moving target detection method improves performance by increasing the number of receiving channels, it also increases the system cost and needs to solve the problem of azimuth-time calibration and phase compensation between channels. In addition, in the case of target signal aliasing, both methods have limitations in extracting individual moving targets.
[0007] The present application aims to solve the problem that the Radon transform extraction peak is difficult due to noise and multi-target signal aliasing in the distance compressed-azimuth time domain moving target image, and then individual moving targets cannot be extracted. SUMMARY
[0008] In order to solve the problems existing in the prior art, the present application provides a SAR moving target extraction method based on Maxtree pre-segmentation method, which solves the problem that the Radon transform extraction peak is difficult due to noise and multi-target signal aliasing in the distance compressed-azimuth time domain moving target image, and then individual moving targets cannot be extracted.
[0009] The technical scheme of the present application is as follows:
[0010] The SAR moving target extraction method based on the Maxtree pre-segmentation method comprises the following steps:
[0011] Step S1: performing Dechirp operation on the range-compression-azimuth time domain signal to obtain a moving target defocusing track graph with relatively high signal-to-noise ratio;
[0012] Step S2: performing Maxtree moving target pre-segmentation on the moving target defocusing track graph to obtain a pre-segmentation result;
[0013] Step S3: converting the pre-segmentation result to the range-compression-azimuth time domain, and then extracting a single moving target by using Radon transform;
[0014] Step S4: clustering and superimposing the single moving target extracted by using Radon transform to obtain an extraction result of all moving targets.
[0015] Preferably, step S1 is specifically as follows:
[0016] After performing Fourier transform on the range-compression-azimuth time domain signal in the azimuth direction and performing gray value remapping, a moving target defocusing track graph with relatively high signal-to-noise ratio is obtained.
[0017] Preferably, step S2 comprises the following sub-steps:
[0018] Sub-step S21: constructing a Maxtree;
[0019] Sub-step S22: filtering the Maxtree by using an extinction filter to obtain a filtered Maxtree;
[0020] Sub-step S23: filtering the filtered Maxtree by using an attribute filter to obtain a Maxtree filtered by the attribute filter;
[0021] Sub-step S24: after the Maxtree filtered by the attribute filter is filtered by the extinction filter, the gray value attribute filter and the area attribute filter, outputting a region represented by a leaf node of the Maxtree, so that a pre-segmentation result of the moving target is obtained.
[0022] Preferably, step S3 comprises the following sub-steps:
[0023] Sub-step S31: performing inverse Fourier transform on the pre-segmentation result to the range-compression-azimuth time domain;
[0024] Sub-step S32: extracting a single target by using Radon transform.
[0025] Preferably, the single target in step S32 comprises a Radon transform graph and a moving target extraction graph.
[0026] The beneficial effects of the moving target extraction method based on the Maxtree pre-segmentation method of the present application are as follows:
[0027] 1. The present application introduces Maxtree into SAR moving target detection, providing a novel idea.
[0028] 2. Compared with the traditional single-channel or multi-channel SAR moving target detection algorithm, the method can extract moving targets one by one in the case of multi-target aliasing. BRIEF DESCRIPTION OF DRAWINGS
[0029] Figure 1 The overall flowchart of the present application.
[0030] Figure 2 The range-compressed-azimuth time domain moving target amplitude graph of the present application.
[0031] Figure 3 The Radon transform result graph of the moving target of the present application.
[0032] Figure 4 The moving target defocusing trajectory graph of the imaging domain of the present application.
[0033] Figure 5 is the NA / NI representation method and the Maxtree representation of the gray scale image of the present application; Figure 5(a) shows an example image (pixel position: gray scale level); Figure 5(b) is the NA / NI Maxtree representation; Figure 5(c) is the node-oriented Maxtree graph (node sequence number: gray scale level [area value]).
[0034] Figure 6 is a filter process schematic diagram of the extinction filter of the present application; Figure 6(a) is an example graph of the original Maxtree; Figure 6(b) is the Maxtree with marked reserved nodes; Figure 6(c) is the Maxtree with unmarked nodes pruned.
[0035] Figure 7 is an area attribute filtering schematic diagram of the present application (the node inner label is represented as: node sequence number: gray scale level [area value]); Figure 7(a) is an example graph of the original Maxtree; Figure 7(b) is the Maxtree after area attribute filtering.
[0036] Figure 8 is a region graph represented by leaf nodes of the present application; Figure 8(a) represents the region represented by leaf node 1; Figure 8(b) represents the region represented by leaf node 2; Figure 8(c) represents the region represented by leaf node 3; Figure 8(d) represents the region represented by leaf node 4.
[0037] Figure 9 The cumulative graph of all the regions represented by the leaf nodes of the present application.
[0038] Fig. 10 is a result diagram of the pre-segmentation result of the present application after inverse Fourier transform; Fig. 10(a) is a result diagram of inverse Fourier transform of Fig. 8(a) to the range-compressed-azimuth time domain; Fig. 10(b) is a result diagram of inverse Fourier transform of Fig. 8(b) to the range-compressed-azimuth time domain; Fig. 10(c) is a result diagram of inverse Fourier transform of Fig. 8(c) to the range-compressed-azimuth time domain; Fig. 10(d) is a result diagram of inverse Fourier transform of Fig. 8(d) to the range-compressed-azimuth time domain.
[0039] Fig. 11 is a result diagram of Radon transform of Fig. 10 of the present application; Fig. 11(a) is a result diagram of Radon transform of Fig. 10(a); Fig. 11(b) is a result diagram of Radon transform of Fig. 10(b); Fig. 11(c) is a result diagram of Radon transform of Fig. 10(c); Fig. 11(d) is a result diagram of Radon transform of Fig. 10(d).
[0040] Fig. 12 is a result diagram of moving target extraction corresponding to the leaf node representing area of the present application; Fig. 12(a) is a result diagram of moving target extraction corresponding to Fig. 8(a); Fig. 12(b) is a result diagram of moving target extraction corresponding to Fig. 8(b); Fig. 12(c) is a result diagram of moving target extraction corresponding to Fig. 8(c); Fig. 12(d) is a result diagram of moving target extraction corresponding to Fig. 8(d).
[0041] Figure 13 Fig. 13 is a result diagram of moving target extraction of the present application based on the Maxtree pre-segmentation method. DETAILED DESCRIPTION
[0042] The embodiments of the present application will be described in detail below with reference to the accompanying drawings.
[0043] The specific embodiments of the present application will be described below to facilitate the understanding of the present application for those skilled in the art, but it should be clear that the present application is not limited to the scope of the specific embodiments, and for those skilled in the art, any changes within the spirit and scope of the present application as defined and determined by the appended claims are obvious, and all the inventions utilizing the concept of the present application are within the scope of protection.
[0044] The SAR moving target extraction method based on the Maxtree pre-segmentation method comprises the following steps:
[0045] Step S1: performing Dechirp operation on the range-compressed-azimuth time domain signal to obtain a moving target defocusing trajectory diagram with relatively high signal-to-noise ratio;
[0046] Step S2: performing Maxtree moving target pre-segmentation on the moving target defocusing trajectory diagram to obtain a pre-segmentation result;
[0047] Step S3: converting the pre-segmentation result to the range-compression-azimuth time domain, and extracting single moving targets by using Radon transform;
[0048] Step S4: clustering and superimposing the single moving targets extracted by using Radon transform to obtain the extraction result of all moving targets.
[0049] Step S1 of the embodiment is specifically:
[0050] After the range-compression-azimuth time domain signal is subjected to Fourier transform in the azimuth direction and gray value remapping, a moving target defocus track diagram with relatively high signal-to-noise ratio is obtained.
[0051] Step S2 of the embodiment includes the following sub-steps:
[0052] Sub-step S21: constructing a Maxtree;
[0053] Sub-step S22: filtering the Maxtree by using an extinction filter to obtain a filtered Maxtree;
[0054] Sub-step S23: filtering the filtered Maxtree by using an attribute filter to obtain a Maxtree filtered by the attribute filter;
[0055] Sub-step S24: outputting a region represented by a leaf node of the Maxtree after the Maxtree filtered by the attribute filter is subjected to the extinction filter, a gray value attribute filter and an area attribute filter, so that a pre-segmentation result of moving targets is obtained.
[0056] Step S3 of the embodiment includes the following sub-steps:
[0057] Sub-step S31: performing inverse Fourier transform on the pre-segmentation result to the range-compression-azimuth time domain;
[0058] Sub-step S32: extracting single targets by using Radon transform.
[0059] The single targets in step S32 of the embodiment include a Radon transform diagram and a moving target extraction diagram.
[0060] In the implementation of the embodiment, the measured SAR data used by the application is obtained by an airborne SAR working in an X-band and having a three-channel SAR-GMTI mode, and a highway with many moving vehicles is selected as the data acquisition site. The obtained original echo data is subjected to clutter cancellation by using DPCA technology, and the result of moving targets in the range-compression-azimuth time domain after the clutter is eliminated is as shown in Figure 2 The result after Radon transform is as shown in Figure 3As shown, it is found that it is relatively difficult to separate the peak values, and the peak values exist ambiguity and weakening, which is because the moving target image of the domain exists noise and the aliasing of multi-target signals, so the peak values directly through the Radon transform cannot extract the moving targets one by one.
[0061] Therefore, in view of the problem that the moving target image of the range-compression-azimuth time domain cannot be Radon transformed to extract the moving targets one by one due to the noise and the aliasing of multi-target signals, a "divide and conquer" strategy is adopted, that is, firstly, coarse imaging is performed to obtain an image with relatively high signal-to-noise ratio, and then the Maxtree is used to separate the aliasing targets, and after the separation into small blocks, the Radon transform is used to extract each moving target. The specific implementation steps are as follows:
[0062] Step one: Dechirp operation is performed on the range-compression-azimuth time domain signal.
[0063] The Figure 2 The moving target signal before azimuth compression shown in FIG. 4 is subjected to Fourier transform in the azimuth direction, and after gray value remapping, an image with relatively high signal-to-noise ratio of the moving target defocusing track is obtained as shown in FIG. 5. Figure 4
[0064] Step two: Maxtree moving target pre-segmentation.
[0065] (1) Based on Figure 4 8 connectivity is used to construct the Maxtree. FIG. 5 shows the creation result of the Maxtree, wherein FIG. 5(a) is an example image, FIG. 5(b) is a NA / NI maximum tree representation, and FIG. 5(c) is a maximum tree graph facing the node.
[0066] (2) The Maxtree is filtered by using an extinction filter.
[0067] For the defocusing track image of the SAR moving target, the moving target usually has a large backscattering coefficient, so its gray value in the image is relatively bright and concentrated, but there is serious coherent speckle and noise interference. In order to remove noise and irrelevant areas such as small changes, the extinction filter is used to realize the simplification and feature extraction of the image based on the constructed Maxtree structure, so as to better highlight the important features and structures in the image. The filtering process of the extinction filter is shown in FIG. 6, wherein FIG. 6(a) is the original maximum tree, the yellow nodes in the figure are the three nodes with the highest extinction value selected, the green nodes in FIG. 6(b) represent the nodes to be retained in the path from the three selected leaf nodes to the root node, and FIG. 6(c) is the maximum tree after pruning the unmarked nodes.
[0068] (3) The Maxtree is filtered again by using an attribute filter.
[0069] Attribute filter is a connected filter, as shown in Fig. 7, used to remove connected regions that do not satisfy the threshold condition. A single attribute or a set of attributes can be used to determine which connected regions should be removed in order to simplify the connected regions in the image and better extract the relevant features in the image. In this experiment, the nodes with area and gray value less than the selected threshold are contracted to their parent nodes in order to enhance the connected features of the image.
[0070] (4) Pre-segmentation result.
[0071] After the Maxtree built by the graph is filtered by the extinction filter, the gray value attribute filter and the area attribute filter, the region represented by the leaf node of the Maxtree is output, and the pre-segmentation result of the moving target can be obtained. Fig. 8 shows part of the pre-segmentation result. In order to judge whether all the regions represented by the leaf nodes cover all the moving targets, all the regions represented by the leaf nodes are accumulated to obtain Figure 9 . By comparing Figure 4 with Figure 9 , it is found that all the leaf nodes basically cover the moving targets in Figure 4 , and the contours and positions of the segmented moving targets are basically consistent with the corresponding moving targets in the defocus image. However, there is still a small amount of multi-target aliasing in the moving targets extracted by the pre-segmentation result. In summary, Maxtree can better achieve pre-segmentation of moving targets.
[0072] Step three: convert the pre-segmentation result of Maxtree to the range-compression-azimuth-time domain, and then use Radon transform to extract individual moving targets.
[0073] Since there is still a small amount of multi-target aliasing problem in the pre-segmentation result of Maxtree moving targets, it is still not possible to accurately extract individual moving targets. Therefore, the moving targets extracted in the imaging domain need to be inverse Fourier transformed to the range-compression-azimuth-time domain, and then Radon transform is used to achieve the extraction of individual moving targets. Specifically, the pre-segmentation result is inverse Fourier transformed to the range-compression-azimuth-time domain to obtain Fig. 10. It can be found that the noise of the moving target image is greatly reduced at this time. Then, Radon transform is used to extract individual targets, and the Radon transform graph is shown in Fig. 11. It is found that the peak value of the Radon transform graph is clear and easy to extract. The extraction result of the moving target is shown in Fig. 12, in which each green line represents a moving target. From the extraction result, compared with the Radon transform directly extracting the moving target, the accuracy of the moving target extracted by this method has been greatly improved.
[0074] Step four: cluster and stack all the individual moving targets extracted by Radon transform to obtain the extraction result of all the moving targets.
[0075] The final extraction result of all the moving targets is shown inFigure 13 as shown.
Claims
1. A SAR moving target extraction method based on the Maxtree pre-segmentation method, characterized in that, Includes the following steps: Step S1: Perform Decirp operation on the range compression-azimuth time domain signal to obtain a moving target defocus trajectory map with a relatively high signal-to-noise ratio; Step S2: Perform Maxtree pre-segmentation on the defocused trajectory map of the moving target to obtain the pre-segmentation result; Step S3: Convert the pre-segmentation results to the range compression-azimuth time domain, and then use Radon transform to extract individual moving targets; Step S4: Cluster and overlay the individual moving targets extracted using Radon transform to obtain the extraction results of all moving targets.
2. The SAR moving target extraction method based on the Maxtree pre-segmentation method according to claim 1, characterized in that, Step S1 specifically involves: By performing a Fourier transform on the range compression-azimuth time domain signal in the azimuth direction and then remapping the gray values, a defocused trajectory map of the moving target with a relatively high signal-to-noise ratio is obtained.
3. The SAR moving target extraction method based on the Maxtree pre-segmentation method according to claim 1, characterized in that, Step S2 includes the following sub-steps: Sub-step S21: Construct the Maxtree; Sub-step S22: Filter the Maxtree using an extinction filter to obtain the filtered Maxtree; Sub-step S23: Filter the filtered Maxtree using an attribute filter to obtain the Maxtree filtered by the attribute filter. Sub-step S24: After filtering the Maxtree by the attribute filter, pass it through the extinction filter, gray value attribute filter and area attribute filter, and output the region represented by the leaf node of the largest tree, so as to obtain the pre-segmentation result of the moving target.
4. The SAR moving target extraction method based on the Maxtree pre-segmentation method according to claim 1, characterized in that, Step S3 includes the following sub-steps: Sub-step S31: Perform an inverse Fourier transform on the pre-segmentation result to the range compression-azimuth time domain; Sub-step S32: Use Radon transform to extract individual targets.
5. The SAR moving target extraction method based on the Maxtree pre-segmentation method according to claim 4, characterized in that, The individual targets in step S32 include: the Radon transform map and the moving target extraction map.