SAR Tomography Method Based on Single-Look Iterative MUSIC and Self-Cancelling Sequential GLRT

Abstract

This application relates to a SAR tomography method based on single-look iterative MUSIC and self-cancelling sequential GLRT. The method includes: selecting candidate PS points from the tomographic dataset based on the amplitude deviation index; dividing the dataset into sub-regions according to the number of PS candidate points; generating subnets for each sub-region using ant colony search and PS candidate points; given a search starting point, connecting the starting point to all PS candidate points within a set distance range; performing 3D imaging and scatterer detection on each connection edge; retaining only connection edges containing one scatterer based on the 3D imaging and scatterer detection results; after parameter estimation of all connection edges in all sub-regions, performing deduplication and maximum connection network search on all connection edges; and using weighted least squares to perform network adjustment estimation on the maximum connection network to obtain the absolute height of each remaining PS point. This method offers high resolution, low sidelobes, and accurate detection of overlapping scatterers.

Description

Technical Field

[0001] This application relates to the field of satellite remote sensing imaging technology, and in particular to a SAR tomography method based on single-view iterative MUSIC and self-cancelling sequential GLRT. Background Technology

[0002] Three-dimensional information of urban buildings is of great significance for urban planning and management, environmental impact assessment, and smart city construction. Traditional field measurements are time-consuming, labor-intensive, and extremely inefficient. While optical photogrammetry can obtain relatively realistic 3D models, its sensors are generally mounted on UAVs or airborne platforms, making it difficult to acquire building information on a large scale, and it is also costly. Spaceborne SAR technology has gained favor among researchers in recent years due to its advantages of wide coverage, revisitability, all-day and all-weather operation. Among them, multi-temporal and multi-passage InSAR technology, represented by PSI, is the most classic and mature. However, because PSI technology assumes that there is at most one strong scattering element target within a single pixel, it can only identify single-scattering element targets, ignoring dual / multi-scattering elements superimposed within the same pixel.

[0003] To address this problem, TomoSAR technology was developed. It provides a third-dimensional resolution beyond the azimuth-range two-dimensional plane by forming a second synthetic aperture in the height direction. Compared to traditional PSI technology, SAR tomography, with its height resolution, can identify overlapping PS points, thus significantly increasing the spatial density of PS points and retrieving finer target details. Furthermore, because SAR tomography utilizes both amplitude and phase information, PS point estimation accuracy is higher.

[0004] After more than 20 years of continuous development, TomoSAR technology has achieved remarkable research results in urban areas. Building tomography inversion includes two key steps: 3D imaging and model selection. First, its imaging methods can be broadly classified into three categories: non-parametric spectrum estimation methods: these do not require manual input parameters, have high computational efficiency, but poor resolution; classic methods include Beamforming and Capon. Parametric spectrum estimation methods: these require manual input parameters (such as the number of scatterers), represented by the MUSIC method. Sparse spectrum estimation methods: based on the sparsity of the signal, compressed sensing technology is introduced into tomography, achieving ultra-high resolution, but with high computational complexity and time consumption. Second, scatterer detection mainly focuses on PS points. The PS point detection problem is actually a model selection (MS) problem, aiming to determine the number of scatterers and the height position of each scatterer. To this end, numerous scholars have conducted extensive research, proposing various PS point detectors. The two most common types are: Generalized Likelihood Ratio Test (GLRT) detectors and Information Theoretic Criteria (ITC) detectors. Among GLRT detectors, SGLRTC is the most representative. Because its judgment process implicitly includes the beamforming method, this method is also called BF-SGLRTC. Although this method is simple to operate and highly efficient, the BF method has low resolution and is susceptible to sidelobes, resulting in energy leakage. Therefore, it is essential to develop a tomographic method with high resolution, low sidelobes, and accurate detection of stacked scatterers. Summary of the Invention

[0005] Therefore, it is necessary to provide a SAR tomography method based on single-view iterative MUSIC and self-cancelling sequential GLRT that has high resolution, low sidelobes, and can accurately detect overlapping scatterers, in order to address the above-mentioned technical problems.

[0006] A SAR tomography method based on single-look iterative MUSIC and self-cancelling sequential GLRT, the method comprising:

[0007] Obtain multi-temporal datasets; preprocess the multi-temporal datasets to obtain tomographic datasets;

[0008] Based on the amplitude deviation index, candidate PS points are selected from the tomographic dataset. Sub-regions are divided according to the number of candidate PS points. Ant colony search and candidate PS points are used to generate subnets for each sub-region. Given a search starting point, the starting point is connected to all candidate PS points within a set distance range. 3D imaging and scattering object detection are performed on each connection edge. Based on the 3D imaging and scattering object detection results, only connection edges containing one scattering object are retained. At the same time, the imaging quality of the connection edge is judged. If the imaging quality is higher than the set threshold, the connection edge is retained. The end point of the effective connection edge is used as the starting point of the next ant colony search. Candidate point connection, 3D imaging and scattering object detection are continued. After the parameter estimation of the connection edges of all sub-regions is completed, all connection edges are deduplicated and the maximum connection network is searched to remove duplicate and isolated connection edges, ensuring that the maximum connection network is connected and non-repeating.

[0009] We use weighted least squares to perform network adjustment on the maximum connected network, select a PS point as the height reference point, and then estimate the absolute height of each remaining PS point by the relative height difference of each connecting edge.

[0010] The aforementioned SAR tomography method based on single-view iterative MUSIC and self-cancelling sequential GLRT utilizes a quadtree segmentation method to divide the target area into several sub-regions and simultaneously construct a network. It then employs a tomography method based on single-view iterative MUSIC and sequential self-cancelling GLRT to verify the number of scatterers and estimate their height. Noise signal energy is eliminated through iteration, and the characteristics of SGLRTC sequential estimation are utilized to estimate the information of one scatterer at a time. This results in fewer sidelobe information in the obtained tomographic spectrum and also allows for better differentiation of nearby overlapping scatterers. This provides an effective guarantee for high-precision inversion of urban building height information and improves the accuracy of SAR tomography. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating a SAR tomography method based on single-view iterative MUSIC and self-cancelling sequential GLRT in one embodiment.

[0012] Figure 2 This is a block diagram of a SAR tomography method based on single-look iterative MUSIC and self-cancelling sequential GLRT in one embodiment;

[0013] Figure 3 Here is an example diagram of quadtree segmentation in one embodiment;

[0014] Figure 4 This is a schematic diagram of the quadtree segmentation result using measured data in another embodiment;

[0015] Figure 5This is a diagram showing the network construction result in one embodiment;

[0016] Figure 6 This is a height inversion result map of all PS points in a region in one embodiment. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] In one embodiment, such as Figure 1 and Figure 2 As shown, a SAR tomography method based on single-look iterative MUSIC and self-cancelling sequential GLRT is provided, including the following steps:

[0019] Step 102: Obtain the multi-temporal dataset; preprocess the multi-temporal dataset to obtain the tomographic dataset.

[0020] The multi-temporal dataset is preprocessed. This dataset was obtained by multiple observations of the same target at different altitudes with similar incident angles by a multi-pass spaceborne SAR satellite. The resulting dataset requires preprocessing operations before tomographic imaging, including registration and de-skewing. The main purpose of registration is to ensure that the same pixel coordinates in different images correspond to the same target feature. The registration accuracy needs to reach the sub-pixel level, and this operation can be performed using different software. De-skewing is to facilitate focusing in the altitude direction. The angular frequency support domain in the altitude direction needs to be corrected to near zero frequency. Slant range de-skewing or topographic reference de-skewing can be used. After completing the above two steps, a tomographic dataset that can be used for tomographic processing is obtained. .

[0021] Step 104: Select PS candidate points from the tomographic dataset based on the amplitude deviation index. Divide the data into sub-regions according to the number of PS candidate points. Use ant colony search and PS candidate points to generate subnets for each sub-region. Given a search starting point, connect the starting point to all PS candidate points within a set distance range. Perform 3D imaging and scattering object detection on each connection edge. Based on the 3D imaging and scattering object detection results, retain only the connection edge containing one scattering object. At the same time, the imaging quality of the connection edge is judged. If the imaging quality is higher than the set threshold, the connection edge is retained. Use the end point of the effective connection edge as the starting point of the next ant colony search. Continue to connect candidate points, perform 3D imaging and scattering object detection. After the parameter estimation of the connection edges of all sub-regions is completed, perform deduplication and maximum connection network search on all connection edges to remove duplicate and isolated connection edges, ensuring that the maximum connection network is connected and non-repeating.

[0022] In a specific embodiment, based on the 3D imaging and scatterer detection results, only the connecting edges containing one scatterer are retained. Simultaneously, the imaging quality of the connecting edges is judged; if the imaging quality is higher than a set threshold, the connecting edge is retained; otherwise, it is discarded. The end point of the valid connecting edge is used as the starting point for the next ant colony search, and the above search, 3D imaging, and scatterer detection operations continue. Based on the sub-network constructed for each sub-region, all sub-networks are deduplicated in terms of connecting edges, ultimately obtaining the SPS network and the scatterer estimation parameters for each SPS edge.

[0023] Based on the amplitude deviation index, PS candidate points are selected, and sub-regions are divided according to their quantity. This method can adaptively and reasonably subdivide complex target regions according to data characteristics, allowing subsequent processing to be more targeted within each sub-region. This avoids the complexity and inaccuracy that may occur when processing the entire large region uniformly, thus improving processing efficiency and accuracy. Ant colony search and PS candidate points are used to generate subnets, and the process of connecting edges is carried out from a given starting point. The ant colony search algorithm has strong global search capabilities and adaptability, and can quickly find optimal paths or connection methods in complex solution spaces. Combined with PS candidate points, a network structure reflecting the characteristics of the target region can be efficiently constructed, providing a reasonable framework for subsequent imaging and detection. Moreover, continuous optimization is performed during the connection edge processing, gradually eliminating unsuitable connections, further improving the accuracy of data processing.

[0024] Each connecting edge undergoes 3D imaging and scatterer detection. Based on the results, only connecting edges containing one scatterer are retained. This step effectively filters out valid information that truly represents the building target, removing false scatterer information that may be generated by noise or other interference factors. This allows subsequent processing based on these valid connecting edges to more accurately reflect the actual situation of the building. Simultaneously, judging the imaging quality of the connecting edges ensures the high reliability of the data used, avoiding errors caused by low-quality imaging data.

[0025] By employing a tomographic imaging method based on single-view iterative MUSIC and sequential self-cancellation GLRT for scatterer count verification and height estimation, this method offers the advantage of iteratively eliminating noise signal energy and reducing noise interference in scatterer detection and height estimation. Utilizing the sequential estimation characteristic of SGLRTC, information for only one scatterer is estimated at a time. This progressively refined estimation method can more accurately distinguish the characteristics of each scatterer, resulting in fewer sidelobe information in the obtained tomographic spectrum. This improves the resolution of scatterers, especially those overlapping scatterers at close range, thus providing strong support for high-precision inversion of building height information.

[0026] Step 106: Perform network adjustment on the maximum connected network using weighted least squares, select a PS point as the height reference point, and then estimate the absolute height of each remaining PS point by the relative height difference of each connecting edge.

[0027] After parameter estimation of all sub-regions' connecting edges is completed, deduplication and a maximum connectivity search are performed on all connecting edges to remove duplicate and isolated edges, ensuring that the maximum connectivity network is connected and non-repeating. Weighted least squares are then used to adjust the maximum connectivity network. First, a PS point is selected as the height reference point, and then the absolute height of each remaining PS point is estimated using the relative height differences between the connecting edges.

[0028] The target area is divided into several sub-regions using a quadtree segmentation method, and network construction is performed simultaneously. Quadtree segmentation can efficiently divide spatial regions hierarchically, facilitating data management and processing at different levels. Simultaneous network construction better integrates information from each sub-region, forming a complete and accurate network structure that reflects the overall picture of the target area. Finally, weighted least squares is used to adjust the maximum connectivity network, and a PS point is selected as the height reference point. The absolute height of each remaining PS point is estimated by the relative height difference of each connecting edge. This adjustment method effectively eliminates systematic and random errors generated during the measurement process, further improving the accuracy of height information inversion and making the final urban building height information more accurate and reliable.

[0029] The aforementioned SAR tomography method based on single-view iterative MUSIC and self-cancelling sequential GLRT utilizes a quadtree segmentation method to divide the target area into several sub-regions and simultaneously construct a network. It then employs a tomography method based on single-view iterative MUSIC and sequential self-cancelling GLRT to verify the number of scatterers and estimate their height. Noise signal energy is eliminated through iteration, and the characteristics of SGLRTC sequential estimation are utilized to estimate the information of one scatterer at a time. This results in fewer sidelobe information in the obtained tomographic spectrum and also allows for better differentiation of nearby overlapping scatterers. This provides an effective guarantee for high-precision inversion of urban building height information and improves the accuracy of SAR tomography.

[0030] In one embodiment, sub-regions are divided according to the number of PS candidate points, including:

[0031] The quadtree is split according to the upper limit threshold of the number of PS candidate points in the sub-region. If the total number of PS candidate points in the current sub-region is higher than the threshold, the sub-region is split again until the total number of PS candidate points in each sub-region is lower than the threshold. The final range of each sub-region is expanded by a certain width from the original region to the surrounding effective region.

[0032] In a specific embodiment, for global PS candidate points, quadtree segmentation is used to divide the region into sub-regions. For each target region (initial region or segmented sub-region), the number of PS points within the region is determined:

[0033]

[0034] When the target area PS quantity Less than the threshold When (set to 5000 here), the operation is performed in this area. : Stop splitting, otherwise execute the operation. Continue segmentation. For each sub-region obtained, although their spatial size varies, the number of PS points within each region is below the set threshold. During ant colony network construction, each sub-network extends outwards by a distance... . This is the upper limit threshold for the distance between connected edges, which is set to 80m here.

[0035] In one embodiment, three-dimensional imaging and scatterer detection are performed on each connection edge, including:

[0036] For each connecting edge, the initial observation signals of the two endpoints are differentially divided. First, SLI-MUSIC three-dimensional imaging is performed, and the Beamforming method is used to obtain a preliminary backscattering energy spectrum estimate. Then, the covariance matrix is ​​estimated, and the covariance matrix is ​​decomposed into eigenvalues ​​to obtain eigenvalues ​​and eigenvectors.

[0037] Backscatter profile estimation and noise energy estimation are performed using the eigenvalues ​​and eigenvectors to obtain the backscatter profile and noise energy; then an iteration threshold judgment is performed. If the number of iterations is less than a given threshold, the process returns to the covariance matrix estimation. If the number of iterations is not less than a given threshold, the loop is exited and the height estimate of the first scatterer is obtained using the backscatter profile estimation obtained in the last iteration.

[0038] SLI-MUSIC 3D imaging requires two steps. The first step is to obtain the height estimate of the first scatterer using the initial observation signal. The height estimate of the first scatterer is then used to remove the signal of that scatterer from the observations, and the remaining observation signal is used to perform SLI-MUSIC 3D imaging again to obtain the height estimate of the second scatterer.

[0039] For scatterer detection, a two-scatterer hypothesis test is first performed. If the height is greater than the set threshold, the hypothesis is accepted, and the height estimates of the two scatterers are obtained. Otherwise, a single-scatterer hypothesis test is performed. If the height is greater than the set threshold, the hypothesis is accepted, and the estimated height of the main scatterer is obtained. Otherwise, it indicates that the differential signal of the connection edge does not contain a scatterer.

[0040] In one embodiment, a preliminary backscattering energy spectrum estimate is obtained using the Beamforming method, including:

[0041] A preliminary estimate of the backscattering energy spectrum is obtained using the Beamforming method:

[0042]

[0043] in, For the guiding matrix, For the observation vector, This represents the initial backscattering profile.

[0044] In one embodiment, the subsequent process of estimating the covariance matrix includes:

[0045] Perform covariance matrix estimation:

[0046]

[0047] in, Indicates the first The covariance matrix obtained from the second estimation For the first The backscattering profile obtained from the second estimation. For the first The estimated noise energy. express The identity matrix, This refers to the number of images.

[0048] In one embodiment, the covariance matrix is ​​decomposed into eigenvalues ​​to obtain eigenvalues ​​and eigenvectors, including:

[0049] Eigenvalue decomposition of the covariance matrix yields the eigenvalues ​​and eigenvectors as follows:

[0050]

[0051] in, , , Representing the eigenvector The corresponding number One eigenvalue; Represents the eigenvectors in the signal subspace; This represents the eigenvectors in the noise subspace.

[0052] In one embodiment, backscatter profile estimation and noise energy estimation are performed using the eigenvalues ​​and eigenvectors to obtain the backscatter profile and noise energy, including:

[0053] set up That is, only one scatterer signal is detected from the observations each time to estimate the backscattering profile:

[0054]

[0055] in, express The Middle One element, for The Middle column vectors, This represents the feature vector in the noise subspace. For the first The backscattering profile obtained from the second estimation.

[0056] In one embodiment, the noise energy is estimated:

[0057]

[0058] in, Find the trace of a matrix. Indicates the first The noise energy obtained from the second estimation. The sample covariance matrix, Indicates the first The covariance matrix obtained from the second estimation.

[0059] In a specific embodiment, The sample covariance matrix is:

[0060]

[0061] in, To represent multiple views, we set it here. That is, only the current pixel is selected for estimation of the sample covariance matrix.

[0062] Then, the number of iterations is checked; if it is less than or equal to the iteration threshold... If the condition is met, the above operation is repeated; otherwise, the loop is exited, and the parameters to be estimated are obtained using the backscatter profile obtained in the last iteration. Here, the height position corresponding to the maximum wave peak in the backscatter profile is selected as the height estimation result.

[0063] The remaining observed signal is obtained by removing the signal of the first scatterer from the observations using the estimation result of the first scatterer. And utilize SLI-MUSIC 3D imaging was performed again to obtain the estimated height of the second scatterer.

[0064] In one embodiment, for scatterer detection, a two-scatterer hypothesis test is first performed. If the height is greater than a set threshold, the hypothesis is accepted, and the height estimates of the two scatterers are obtained. Otherwise, a single-scatterer hypothesis test is performed. If the height is greater than a set threshold, the hypothesis is accepted, and the estimated height of the main scatterer is obtained. Otherwise, it indicates that the differential signal of the connection edge does not contain a scatterer, including:

[0065] For scatterer detection, the first step is to make an assumption. : Contains two scatterers - Does not contain two scatterers; Judgment:

[0066]

[0067] in, and These represent the projections of the observed signals onto... and The energy corresponding to the component above, ;

[0068] If satisfied If we assume that the observed signal contains two scatterers, the heights of which are the positions estimated in the previous two 3D imagings;

[0069] Next, we make assumptions. (Contains one scatterer - does not contain a scatterer);

[0070] judge:

[0071]

[0072] in, Indicates the projection of the observed signal onto The energy corresponding to the component above, ;

[0073] If satisfied If this is assumed, it means that the observed signal contains a scatterer, and the corresponding height is the position estimated in the first three-dimensional imaging; if this condition is met... If we assume that the observed signal does not contain a scatterer, then it means that the observed signal does not contain a scatterer.

[0074] In a specific embodiment, such as Figure 3 The image shows an example of quadtree partitioning when the sub-region is inside the initial region (e.g., ...). Figure 3 (blue area), the sub-region needs to expand its search range in all directions; when the sub-region is at the edge of the initial region (e.g., ... Figure 3 (Purple area) Sub-regions expand only inwards in adjacent directions to other sub-regions. By widening the search area, connections between each sub-region can be guaranteed. Intersecting sub-regions (such as...) Figure 3The red dashed box indicates that duplicate connection edges will be removed after the parallel network construction of each sub-region is completed. Because the above operation ensures sufficient redundancy, connections between sub-regions can still be achieved after the duplicate edges are removed.

[0075] like Figure 4 The image shown is the result of quadtree segmentation using measured data, which consists of 15 images of city B obtained from observations by satellite A.

[0076] like Figure 5 As shown, this is the network construction result. It can be seen that the network covers most of the area, and the redundancy of the connection edges of each PS point is also guaranteed.

[0077] like Figure 6 As shown, this figure represents the height inversion results of all PS points in the area. The figure is a three-dimensional representation, and it can be clearly seen that the inversion results are consistent with objective reality, and the detailed information of most buildings can be estimated well.

[0078] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0079] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0080] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A SAR tomography method based on single-view iterative MUSIC and self-cancelling sequential GLRT, characterized in that, The method includes: Obtain a multi-temporal dataset; preprocess the multi-temporal dataset to obtain a tomographic dataset; Based on the amplitude deviation index, PS candidate points are selected from the tomographic dataset. Sub-regions are divided according to the number of PS candidate points. Ant colony search and the PS candidate points are used to generate subnets for each sub-region. Given a search starting point, the starting point is connected to all PS candidate points within a set distance range. 3D imaging and scatterer detection are performed on each connection edge. Based on the 3D imaging and scatterer detection results, only connection edges containing one scatterer are retained. At the same time, the imaging quality of the connection edge is judged. If the imaging quality is higher than the set threshold, the connection edge is retained. The end point of the effective connection edge is used as the starting point of the next ant colony search. Candidate point connection, 3D imaging and scatterer detection are continued. After the parameter estimation of the connection edges of all sub-regions is completed, all connection edges are deduplicated and the maximum connection network is searched to remove duplicate and isolated connection edges, ensuring that the maximum connection network is connected and non-repeating. We use weighted least squares to perform network adjustment on the maximum connected network, select a PS point as the height reference point, and then estimate the absolute height of each remaining PS point by the relative height difference of each connecting edge. Perform 3D imaging and scatterer detection on each connecting edge, including: For each connecting edge, the initial observation signals of the two endpoints are differentially divided. First, SLI-MUSIC three-dimensional imaging is performed, including obtaining a preliminary backscattering energy spectrum estimate using the Beamforming method. Then, the covariance matrix is ​​estimated, and then the covariance matrix is ​​decomposed into eigenvalues ​​and eigenvectors. Backscatter profile estimation and noise energy estimation are performed using the eigenvalues ​​and eigenvectors to obtain the backscatter profile and noise energy; then an iteration threshold judgment is performed. If the number of iterations is less than a given threshold, the process returns to the covariance matrix estimation. If the number of iterations is not less than a given threshold, the loop is exited and the height estimate of the first scatterer is obtained using the backscatter profile estimation obtained in the last iteration. SLI-MUSIC 3D imaging requires two steps. The first step is to obtain the height estimate of the first scatterer using the initial observation signal. The signal of the first scatterer is then removed from the observations using the preliminary height estimate of the first scatterer, and the remaining observation signal is used to perform SLI-MUSIC 3D imaging again to obtain the height estimate of the second scatterer. The detection of scatterers includes first performing a two-scatterer hypothesis test. If the height is greater than a set threshold, the hypothesis is accepted, and the height estimates of the two scatterers are obtained. Otherwise, a single-scatterer hypothesis test is performed. If the height is greater than a set threshold, the hypothesis is accepted, and the estimated height of the main scatterer is obtained. Otherwise, it indicates that the differential signal of the connection edge does not contain a scatterer.

2. The method according to claim 1, characterized in that, The region is divided into sub-regions based on the number of PS candidate points, including: The quadtree is split according to the upper limit threshold of the number of PS candidate points in the sub-region. If the total number of PS candidate points in the current sub-region is higher than the threshold, the sub-region is split again until the total number of PS candidate points in each sub-region is lower than the threshold. The final range of each sub-region is expanded by a certain width from the original region to the surrounding effective region.

3. The method according to any one of claims 1 to 2, characterized in that, A preliminary estimate of the backscattering energy spectrum is obtained using the Beamforming method, including: A preliminary estimate of the backscattering energy spectrum is obtained using the Beamforming method: P.S 0 diag(A -1 g) Where A is the steering matrix, g represents the observation vector, and P 0 This represents the initial backscattering profile.

4. The method according to claim 3, characterized in that, The next step, covariance matrix estimation, includes: Perform covariance matrix estimation: R i =AP i-1 A H +σ i-1 I N Among them, R i Let P represent the covariance matrix obtained from the i-th estimation. i-1 For the backscattering profile obtained from the (i-1)th estimation, σ i-1 Let I be the noise energy estimated in the (i-1)th time. N H represents an N×N identity matrix, where N is the number of images and H represents the transpose operation.

5. The method according to claim 4, characterized in that, Eigenvalue decomposition of the covariance matrix yields eigenvalues ​​and eigenvectors, including: Eigenvalue decomposition of the covariance matrix yields the eigenvalues ​​and eigenvectors as follows: [U,V]=eig(R i ) Among them, U=[E s E n ], Represents the eigenvectors in the signal subspace; This represents the eigenvectors in the noise subspace.

6. The method according to claim 3, characterized in that, Backscattering profile estimation and noise energy estimation are performed using the eigenvalues ​​and eigenvectors, resulting in the backscattering profile and noise energy, including: Set N s =1, meaning that only one scatterer signal is detected from the observations each time for backscattering profile estimation: in, P represents i The m-th element in Let m be the m-th column vector in A. P represents the eigenvectors in the noise subspace. i Let H be the backscattering profile obtained from the i-th estimation, and let H denote the transpose operation.

7. The method according to claim 6, characterized in that, The method further includes: Estimating the noise energy: Where trace represents finding the trace of a matrix, σ i This represents the noise energy obtained from the i-th estimation. Let R be the sample covariance matrix. i Let represent the covariance matrix obtained from the i-th estimation.

8. The method according to claim 1, characterized in that, For scatterer detection, the two-scatterer hypothesis test is first performed. If the height is greater than the set threshold, the hypothesis is accepted, and the height estimates of the two scatterers are obtained. Otherwise, the single-scatterer hypothesis test is performed. If the height is greater than the set threshold, the hypothesis is accepted, and the estimated height of the main scatterer is obtained. Conversely, it indicates that the differential signal of the connection edge does not contain scatterers, including: For scatterer detection, the first step is to make an assumption. Contains two scatterers - does not contain two scatterers; Determine: Where α2 and α r These represent the observed signals projected onto u2 and u2, respectively. The energy corresponding to the component on the surface, T2∈[0,1]; If the H2 hypothesis is satisfied, it means that the observed signal contains two scatterers, and their corresponding heights are the positions estimated by the previous two 3D imagings. Next, we make the following assumptions: H1-H0: contains one scatterer - does not contain a scatterer; judge: Where α1 represents the energy corresponding to the component of the observed signal projected onto u1, and T1∈[0,1]; If the H1 hypothesis is satisfied, it means that the observed signal contains a scatterer, and the corresponding height is the position estimated by the first three-dimensional imaging. If the H0 hypothesis is satisfied, it means that the observed signal does not contain a scatterer.

Images