A sparse array interferometric SAR three-dimensional imaging method based on elevation position prior
Patent Information
- Application Number
- CN202610961930.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-30
- Publication Date
- 2026-09-11
AI Technical Summary
[0004]第一，观测矩阵性能下降
[0030] Significantly improves imaging quality: By incorporating data-driven prior information on geometric structure into the imaging inversion process, the algorithm is effectively guided to escape the local optimum caused by high sidelobes of non-uniform arrays, greatly suppressing stray points and erroneous estimations, and the reconstructed three-dimensional structure of buildings is more complete and clearer.
Smart Images
Figure CN122731663A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of radar signal processing and remote sensing imaging, and more specifically, to a sparse array interferometric SAR three-dimensional imaging method based on prior elevation position. Background Technology
[0002] Synthetic aperture radar (SAR) 3D imaging technology, especially array interferometric SAR, acquires the 3D spatial information of targets through single-pass imaging, and has significant application value in fields such as urban mapping, terrain modeling, disaster assessment, and forest resource surveys. Compared with traditional 2D SAR imaging, 3D SAR imaging can acquire the elevation information of targets, achieving true 3D spatial reconstruction.
[0003] However, on lightweight platforms such as UAVs, due to limitations in payload capacity, power consumption, and cost, systems typically employ sparse antenna arrays with few channels and short baselines. While this sparse array simplifies hardware, it also leads to two core problems:
[0004] First, the performance of the observation matrix degrades. The autocorrelation function of the observation matrix of a sparse array typically has high sidelobe characteristics. In noisy environments, 3D imaging algorithms based on minimum residual optimization (such as orthogonal matching pursuit OMP, iterative shrinkage thresholding, etc.) are prone to getting trapped in local optima, generating a large number of incorrectly estimated stray points, which seriously pollutes the quality of the 3D point cloud.
[0005] Second, elevation ambiguity is a significant issue. For tall targets (such as high-rise buildings), the system's maximum unambiguous height is often insufficient to cover their true height, leading to phase entanglement. Furthermore, the dihedral reflection structure formed by the ground and building walls introduces secondary scattering false targets, further interfering with the 3D imaging results.
[0006] Existing 3D imaging methods primarily focus on improving signal processing algorithms, such as atomic norm minimization, sparse Bayesian learning, and multi-signal classification algorithms. While these methods improve imaging performance to some extent, the performance improvement is limited when observational information is severely insufficient, and they also have high computational complexity. More importantly, existing methods fail to fully utilize the explicit geometric priors of targets such as urban buildings—e.g., vertical facades and horizontal roofs—which are ubiquitous and easily extracted in urban scenes.
[0007] Therefore, there is an urgent need for a method that can effectively utilize prior information about scene structure, overcome the inherent defects of sparse arrays, and improve the quality of 3D imaging. Summary of the Invention
[0008] To address the aforementioned technical problems in related technologies, this invention provides a sparse array interferometric SAR three-dimensional imaging method based on prior elevation location, which can solve the above problems.
[0009] To achieve the above-mentioned technical objectives, the technical solution of the present invention is implemented as follows:
[0010] A sparse array interferometric SAR three-dimensional imaging method based on prior elevation location includes the following steps:
[0011] Step S1: Obtain single-pass echo data of the target scene from the sparse array interferometric SAR system, perform two-dimensional SAR imaging and registration, and then use the first three-dimensional imaging algorithm to perform elevation inversion at the range-azimuth grid corresponding to each pixel of the two-dimensional SAR image to generate a preliminary three-dimensional point cloud.
[0012] Step S2: Process the preliminary 3D point cloud to extract prior information about the geometric structure of the target scene, specifically including:
[0013] Step S21: Use a dimension-reduced Hough transform to detect the horizontal and vertical planes in the preliminary 3D point cloud to obtain candidate plane parameters;
[0014] Step S22: According to SAR imaging geometry, in the two-dimensional parameter space spanned by the azimuth dimension and the lower view dimension, there is at most one effective scattering center in the spatial resolution cell corresponding to each azimuth-lower view grid cell. Based on the physical constraint that there is at most one effective scattering center in the spatial resolution cell, the candidate plane is screened to remove false planes caused by stray points or incorrect estimations, and the true fitting plane is obtained.
[0015] Step S23: Based on the true fitting plane and the SAR imaging geometric model, calculate the candidate target elevation value corresponding to each two-dimensional SAR image pixel to form a priori elevation set;
[0016] Step S3: Using the prior elevation set as the initial value or solution constraint for the second three-dimensional imaging algorithm, perform three-dimensional imaging on the echo data again to generate the final high-precision three-dimensional point cloud.
[0017] Furthermore, the dimension reduction Hough transform in step S21 specifically involves: for horizontal plane detection, fixing the polar angle parameter to zero and searching only the azimuth and polar radius; for vertical plane detection, fixing the azimuth parameter to zero and searching only the polar angle and polar radius; simultaneously, the RANSAC algorithm is introduced to adjust local voting weights and address the plane missing problem caused by occlusion.
[0018] Furthermore, the specific method for plane selection in step S22 is as follows: establish a coordinate plane with the azimuth direction as the horizontal axis and the downward viewing angle as the vertical axis, project the candidate plane onto the plane, and count the number of point clouds belonging to each plane in each projected grid; if more than a preset proportion of the grids in the point cloud corresponding to a candidate plane have been occupied by the point cloud corresponding to another candidate plane, then the candidate plane is determined to be a false plane and is removed.
[0019] Furthermore, in step S22, for the case where the same two-dimensional SAR image pixel corresponds to multiple true fitting planes, a multi-plane attribution decision strategy is adopted to determine the priority or weight of each candidate prior, specifically based on at least one of the following criteria:
[0020] Scattering intensity criterion: Calculate the average amplitude of the scattering coefficients of the point clouds belonging to each plane in the neighborhood of the pixel. The higher the amplitude, the greater the corresponding prior weight.
[0021] Point cloud density criterion: Calculate the point cloud density of each plane in the neighborhood of the pixel. The higher the density, the greater the prior weight.
[0022] Distance criterion: Prioritize candidate priors that place the elevation values within the maximum unambiguous height range of the system.
[0023] Furthermore, the second 3D imaging algorithm in step S3 is a priori-guided conjugate gradient algorithm. This algorithm models the 3D imaging problem as a meshless parameter estimation problem, using prior elevation as the initial value for iteration and introducing Tikhonov regularization constraints. It iteratively updates the elevation estimate by maximizing the objective function with regularization terms and along the conjugate gradient direction until convergence. The regularization term utilizes prior elevation information to construct constraints, which are expressed as follows: ,in The prior elevation vector, For regularization parameters, This is an estimate of the current elevation.
[0024] Furthermore, the objective function is: ,in For the observed signal vector, To estimate from the current elevation A defined observation matrix for False inverse, satisfying: ,in for The transpose of .
[0025] Furthermore, the first three-dimensional imaging algorithm is a greedy sparse reconstruction algorithm with lower computational complexity than the second three-dimensional imaging algorithm, preferably an orthogonal matching pursuit (OMP) algorithm.
[0026] Furthermore, the sparse array interferometric SAR system is a non-uniform antenna array, which includes nested arrays or coprime arrays; the observation matrix in step S3 It is a virtual uniform linear array observation matrix constructed based on the differential cooperative array of the sparse array, and utilizes the theoretical elevation resolution capability of the virtual aperture extension system generated by the nested array.
[0027] Furthermore, in step S23, the geometric model for calculating the elevation value of the candidate target adopts a spherical wave incident model or a plane wave incident model to adapt to the imaging geometry at different flight altitudes.
[0028] Furthermore, the two-dimensional SAR imaging in step S1 includes range compression, azimuth compression and motion compensation processing, and the registration includes pixel-level or sub-pixel-level registration of the images in each channel.
[0029] The beneficial effects of this invention are:
[0030] Significantly improves imaging quality: By incorporating data-driven prior information on geometric structure into the imaging inversion process, the algorithm is effectively guided to escape the local optimum caused by high sidelobes of non-uniform arrays, greatly suppressing stray points and erroneous estimations, and the reconstructed three-dimensional structure of buildings is more complete and clearer.
[0031] Automated solution to complex problems: Without human intervention, this invention can automatically identify and utilize scene structure to effectively handle phase entanglement caused by insufficient maximum unambiguous height of the system, as well as false targets caused by secondary scattering due to dihedral reflection, thereby enhancing the practicality and robustness of the method.
[0032] Balancing efficiency and accuracy: This method employs a two-stage strategy of "fast coarse imaging + fine optimization." First, a high-efficiency algorithm (such as OMP) is used to obtain prior information, followed by optimization based on that prior information using a high-precision algorithm. Compared to directly using highly complex algorithms (such as ANM), this method significantly reduces overall computational overhead while maintaining final imaging accuracy.
[0033] It has strong adaptability to sparse arrays: It is particularly suitable for small and lightweight UAV-borne SAR systems with few antenna channels and non-uniform baselines, providing an effective technical approach to solve the bottleneck problem of poor 3D imaging quality on such platforms. Attached Figure Description
[0034] 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.
[0035] Figure 1 This is a flowchart of a sparse array interferometric SAR three-dimensional imaging method based on prior elevation position, as described in an embodiment of the present invention.
[0036] Figure 2 This is a coarse imaging point cloud image as described in an embodiment of the present invention;
[0037] Figure 3 This is a schematic diagram of the geometric model for prior position calculation as described in an embodiment of the present invention;
[0038] Figure 4 This is a fine-scale point cloud imaging result diagram as described in an embodiment of the present invention. Detailed Implementation
[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.
[0040] like Figure 1 As shown, this invention discloses a sparse array interferometric SAR three-dimensional imaging method based on prior elevation location, comprising the following steps:
[0041] Step S1: Coarse Imaging and Point Cloud Generation. Acquire single-pass echo data of the target scene from the sparse array interferometric SAR system, perform 2D SAR imaging and image registration, and then use a first-class 3D imaging algorithm with low computational complexity (such as the orthogonal matching pursuit OMP algorithm) to perform elevation inversion for each range-azimuth resolution unit to generate a preliminary 3D point cloud of the target scene.
[0042] Step S2: Prior Information Extraction. The preliminary 3D point cloud is processed to extract prior information about the geometric structure of the target scene. This step specifically includes:
[0043] Step S21: Dimensionally Reduced Hough Transform Plane Detection. Considering the characteristics of urban scenes where buildings are mainly horizontal planes (ground, rooftops) and vertical planes (floors), the standard 3D Hough transform is reduced in dimension. Parameter search spaces are defined for the horizontal plane (polar angle θ=0) and the vertical plane (azimuth angle φ=0), respectively. Voting is performed in the reduced parameter spaces to detect candidate horizontal and vertical plane parameters. Simultaneously, the RANSAC algorithm is introduced to adjust local voting weights and address the plane missingness problem caused by occlusion.
[0044] Step S22: Plane selection based on SAR imaging geometry. Based on the physical constraint in SAR imaging that "there is one and only one true scattering point within each azimuth-downward-view cell," the candidate planes detected in Step S21 are selected. The candidate planes are projected onto the azimuth-downward-view plane, and the number of point clouds belonging to each plane within each cell is counted. If the cell containing most points corresponding to a plane is already occupied by other planes, the plane is determined to be a false plane caused by stray points or incorrect estimation, and it is removed to obtain the true set of fitted planes.
[0045] Step S23: Prior Elevation Calculation. Based on the filtered true fitting planes (including ground, floor, roof, etc.), and combined with the SAR imaging geometric model, the elevation value of the candidate target corresponding to each 2D SAR image pixel is calculated to form a prior elevation set. Specifically, for each pixel, its spatial intersection with each fitting plane is solved, and the vertical height of the intersection is a prior elevation estimate for that pixel.
[0046] Step S3: Refined 3D Imaging. The prior elevation set generated in step S23 is used as the initial value or solution constraint for the second type of 3D imaging algorithm to perform 3D imaging again on the original echo data. Preferably, a prior-guided nonlinear conjugate gradient (NCG) algorithm is used, with the prior elevation as the initial value for iteration. By optimizing a maximum likelihood objective function, the precise elevation and scattering coefficient within each pixel are solved, thereby generating the final high-precision 3D point cloud. The conjugate gradient algorithm models the 3D imaging problem as a meshless sparse parameter estimation problem. Its objective function is constructed based on the maximum likelihood criterion of single snapshot data, and the target elevation is solved iteratively using the conjugate gradient method.
[0047] Example:
[0048] This embodiment is based on a typical UAV-borne sparse array interferometric SAR system, whose antenna array is a non-uniformly sparse configuration, with a nested array of normalized baselines [0,1,3,5]. The target scene is an urban building complex.
[0049] Step 1: Data Acquisition and Coarse Imaging
[0050] A SAR system mounted on an unmanned aerial vehicle (UAV) platform flies in side-looking mode to acquire raw echo data. After range compression, azimuth compression (such as the ω-k algorithm or range-Doppler algorithm), and motion compensation, a multi-channel 2D SAR image sequence is generated. After registration of the images from each channel, pixel sequences at the same spatial location are extracted as the observation vector y. The OMP algorithm is used to solve the sparse linear equation y=Ax, where A is the observation matrix and x is the elevation scattering coefficient distribution to be determined. This process is repeated for each pixel to obtain... Figure 2 The image shows a preliminary 3D point cloud containing numerous stray points and phase entanglement.
[0051] Step 2: Extract prior elevation information
[0052] (2.1) Dimensional reduction Hough transform: Perform dimensional reduction Hough transform on the initial point cloud.
[0053] In the ground distance coordinate system, the three-dimensional coordinates of a point on the point cloud are denoted as [ ]. This represents the projection of the distance from the target to the antenna onto the ground. Indicates azimuth coordinates. This represents the height parameter perpendicular to the ground.
[0054] For horizontal plane detection: only traverse polar angles. (Single value) and azimuth Calculate the distance parameter for each candidate horizontal plane. .right Parameter space voting, with peak values corresponding to ground and rooftop planes.
[0055] For vertical plane detection: only traverse the azimuth angle. (Single value) and polar angle Calculate the distance parameter for each candidate vertical plane. .right Parameter space voting, the peak value corresponds to the wall plane with different orientations.
[0056] Since secondary scattering false targets are usually generated by the dihedral structure formed by the ground and the building wall, their spatial position is not on the horizontal plane (ground, roof) or vertical plane (wall) that we detect through the dimension-reduced Hough transform, but forms an inclined screen, and is therefore naturally and implicitly filtered out during the planar detection process.
[0057] In practical UAV-borne SAR observations, buildings may be obstructed by trees, protruding balconies, or shadows, resulting in incomplete floor and ground point clouds and large-area gaps in the initial 3D point cloud. To address this issue, this embodiment introduces the RANSAC algorithm for local voting weight adjustment during the dimensionality reduction Hough transform. Specifically, in parameter space voting, equal-weighted voting is not used; instead, voting weights are assigned based on the point density and spatial consistency of each local point cloud region. Regions with sparse point clouds or obvious outliers have reduced voting weights, while regions with continuous point clouds and consistent structures have increased voting weights. Simultaneously, the RANSAC algorithm iteratively generates candidate plane models by randomly sampling a minimum point set (e.g., 3 points for a horizontal plane and 2 points for a vertical plane), counts the number of interior points that fit the model, and finally selects the model with the most interior points as the fitting plane. This method can robustly extract the correct building structure plane even when the point cloud has less than 40% missing points or noise.
[0058] (2.2) Plane Filtering and Multi-Plane Assignment Decision: Project the candidate floor, ground, and roof planes onto the azimuth-downward view coordinate plane. Analyze which plane's point cloud dominates within each azimuth-downward view grid cell. If a candidate plane (e.g., an incorrect plane) covers an area where most grid cells are already "occupied" by the point cloud of another more reliable plane (e.g., the ground), discard it. Finally, retain the correct geometric planes.
[0059] In urban scenarios, a single pixel in a 2D SAR image may simultaneously correspond to multiple geometric planes. For example, a building at the end of a road may have its ground pixels geometrically located on both the ground plane and the extension direction of the building's walls; similarly, a corner pixel of a building may belong to two mutually perpendicular walls. For such cases, this embodiment employs the following multi-plane attribution decision strategy: First, for each pixel, calculate its intersections with all true fitted planes (the planes filtered in step S22) to obtain multiple candidate prior elevations. Then, the priority or weight of each candidate prior is determined according to the following criteria:
[0060] Scattering intensity criterion: Calculate the average amplitude of the scattering coefficient of the point cloud belonging to each plane in the neighborhood of the pixel. The higher the amplitude, the greater the prior weight of the plane.
[0061] Point cloud density criterion: Calculate the point cloud density of each plane in the neighborhood of the pixel. The higher the density of the plane, the greater the prior weight.
[0062] Distance criterion: For phase-wound scenarios, priority is given to candidate priors that make the elevation values fall within the maximum unambiguous height range of the system.
[0063] Ultimately, the prior elevation set for each pixel can be a weighted list, or the pixel with the highest weight can be selected as the initial value for iteration. This strategy effectively solves the prior ambiguity problem in complex overlapping areas such as corners and ground-wall boundaries.
[0064] (2.3) Calculate the prior: based on Figure 3 The geometric model shown calculates the intersection points of each pixel coordinate (azimuth a, slant range r) with each fitted plane (Ax+By+Cz+D=0) in a 2D SAR image, taking into account the radar platform height H. The elevation value z0 in the 3D coordinates (x0, y0, z0) of the intersection point is a priori elevation for that pixel. The priori elevations of all pixels are saved and used as initial values for subsequent fine imaging.
[0065] Step 3: Prior-based fine imaging
[0066] (3.1) Regularized Conjugate Gradient Algorithm: A prior-guided nonlinear conjugate gradient (NCG) algorithm is used for fine 3D imaging. To strengthen the prior constraints, a Tikhonov regularization term is introduced to construct an equivalent minimization objective function of the maximum a posteriori estimation method:
[0067]
[0068] The first item The second term represents the projected residuals of the observed data (equivalent to the negative log-likelihood). These are prior constraints. The objective function is about... Smooth and differentiable, the solution is obtained using the nonlinear conjugate gradient method, with the following iterative steps:
[0069] Calculate the current gradient ,in This represents the estimated elevation in the i-th iteration. Let i represent the objective function of the i-th iteration. Represents the gradient operator;
[0070] Update conjugate direction ,in The Polak-Ribière formula is adopted;
[0071] along Perform Armijo line search to determine step size ;
[0072] renew .
[0073] Based on prior elevation as initial value for iteration The process iterates until convergence. The regularization term continuously guides the solution towards the prior value throughout the optimization process, effectively suppressing the local optimum trap caused by the high sidelobes of the non-uniform array.
[0074] (3.2) Virtual Aperture Extension Based on Differential Cooperative Array: Considering the constraint of the limited number of channels on the UAV-borne platform, the sparse array interferometric SAR system of this invention preferably adopts a nested array configuration. Based on this, the observation matrix in the fine imaging step S3... Instead of directly using the original physical array, a virtual uniform linear array observation matrix is constructed based on the differential cooperative array of the sparse array.
[0075] Specifically, for the set of physical array element positions Its differential cooperative array is defined as Taking a two-level nested array (e.g., physical element positions [0, 1, 3, 5]) as an example, its differential cooperative array can generate a continuous set of virtual elements from -5 to 5, effectively expanding the aperture to twice the physical aperture. In this embodiment, the observation matrix corresponding to this virtual uniform linear array is used as the observation model for fine imaging, thereby improving the theoretical elevation resolution of the system by approximately one time while keeping the number of physical channels unchanged.
[0076] However, the expansion of the virtual aperture introduces phase ambiguity. This invention utilizes the prior elevation set extracted in step S23. The true phase of each candidate target in the virtual array observation is uniquely determined, thereby deblurring the image. This collaborative design integrates the degree-of-freedom extension technique of array signal processing with the deep prior geometry of the SAR scene, achieving a "1+1>2" effect.
[0077] Experimental results:
[0078] Due to prior elevation The algorithm has approximated the true value and converges quickly to the global optimum, effectively avoiding the local optimum trap caused by high sidelobes in a non-uniform array. Furthermore, this process does not require a pre-set discrete grid, fundamentally solving the off-grid effect. Finally, the result is as follows: Figure 4 The detailed 3D point cloud shown. (Compared to...) Figure 2 compared to, Figure 4 The number of stray points was significantly reduced, the building structure was clear, and phase entanglement and secondary scattering false targets were effectively suppressed.
[0079] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A sparse array interferometric SAR three-dimensional imaging method based on prior elevation location, characterized in that, Includes the following steps: Step S1: Obtain single-pass echo data of the target scene from the sparse array interferometric SAR system, perform two-dimensional SAR imaging and registration, and then use the first three-dimensional imaging algorithm to perform elevation inversion at the range-azimuth grid corresponding to each pixel of the two-dimensional SAR image to generate a preliminary three-dimensional point cloud. Step S2: Process the preliminary 3D point cloud to extract prior information about the geometric structure of the target scene, specifically including: Step S21: Use a dimension-reduced Hough transform to detect the horizontal and vertical planes in the preliminary 3D point cloud to obtain candidate plane parameters; Step S22: According to SAR imaging geometry, in the two-dimensional parameter space spanned by the azimuth dimension and the lower view dimension, there is at most one effective scattering center in the spatial resolution cell corresponding to each azimuth-lower view grid cell. Based on the physical constraint that there is at most one effective scattering center in the spatial resolution cell, the candidate plane is screened to remove false planes caused by stray points or incorrect estimations, and the true fitting plane is obtained. Step S23: Based on the true fitting plane and the SAR imaging geometric model, calculate the candidate target elevation value corresponding to each two-dimensional SAR image pixel to form a priori elevation set; Step S3: Using the prior elevation set as the initial value or solution constraint for the second three-dimensional imaging algorithm, perform three-dimensional imaging on the echo data again to generate the final high-precision three-dimensional point cloud.
2. The sparse array interferometric SAR three-dimensional imaging method based on prior elevation location as described in claim 1, characterized in that, The dimension reduction Hough transform in step S21 is as follows: for horizontal plane detection, the polar angle parameter is fixed at zero, and only the azimuth and polar radius are searched; for vertical plane detection, the azimuth parameter is fixed at zero, and only the polar angle and polar radius are searched; at the same time, the RANSAC algorithm is introduced to adjust the local voting weights and handle the plane missing problem caused by occlusion.
3. The sparse array interferometric SAR three-dimensional imaging method based on prior elevation location as described in claim 1, characterized in that, The specific method for plane selection in step S22 is as follows: establish a coordinate plane with the azimuth direction as the horizontal axis and the downward view as the vertical axis, project the candidate planes onto the plane, and count the number of point clouds belonging to each plane within each projection grid. If a candidate plane has a point cloud whose grid exceeds a preset proportion that is occupied by the point cloud of another candidate plane, then the candidate plane is determined to be a false plane and is removed.
4. The sparse array interferometric SAR three-dimensional imaging method based on prior elevation location as described in claim 1, characterized in that, In step S22, for the case where the same two-dimensional SAR image pixel corresponds to multiple true fitting planes, a multi-plane assignment decision strategy is adopted to determine the priority or weight of each candidate prior, specifically based on at least one of the following criteria: Scattering intensity criterion: Calculate the average amplitude of the scattering coefficients of the point clouds belonging to each plane in the neighborhood of the pixel. The higher the amplitude, the greater the corresponding prior weight. Point cloud density criterion: Calculate the point cloud density of each plane in the neighborhood of the pixel. The higher the density, the greater the prior weight. Distance criterion: Prioritize candidate priors that place the elevation values within the maximum unambiguous height range of the system.
5. The sparse array interferometric SAR three-dimensional imaging method based on prior elevation location as described in claim 1, characterized in that, The second 3D imaging algorithm in step S3 is a priori-guided conjugate gradient algorithm. This algorithm models the 3D imaging problem as a meshless parameter estimation problem, using prior elevation as the initial value for iteration and introducing Tikhonov regularization constraints. It iteratively updates the elevation estimate by maximizing the objective function with regularization terms and along the conjugate gradient direction until convergence. The regularization term utilizes prior elevation information to construct constraints, which are expressed as follows: ,in The prior elevation vector, For regularization parameters, This is an estimate of the current elevation.
6. The sparse array interferometric SAR three-dimensional imaging method based on prior elevation location as described in claim 5, characterized in that, The objective function is: ,in For the observed signal vector, To estimate from the current elevation A defined observation matrix for False inverse, satisfying: ,in for The transpose of .
7. The sparse array interferometric SAR three-dimensional imaging method based on prior elevation location as described in claim 1, characterized in that, The first three-dimensional imaging algorithm is a greedy sparse reconstruction algorithm with lower computational complexity than the second three-dimensional imaging algorithm.
8. The sparse array interferometric SAR three-dimensional imaging method based on prior elevation location as described in claim 1, characterized in that, The sparse array interferometric SAR system is a non-uniform antenna array, which includes nested arrays or coprime arrays; the observation matrix in step S3 It is a virtual uniform linear array observation matrix constructed based on the differential cooperative array of the sparse array, and utilizes the theoretical elevation resolution capability of the virtual aperture extension system generated by the nested array.
9. A sparse array interferometric SAR three-dimensional imaging method based on prior elevation location as described in claim 1, characterized in that, In step S23, the geometric model for calculating the elevation value of the candidate target adopts either a spherical wave incident model or a plane wave incident model to adapt to the imaging geometry at different flight altitudes.
10. A sparse array interferometric SAR three-dimensional imaging method based on prior elevation location as described in claim 1, characterized in that, The two-dimensional SAR imaging in step S1 includes range compression, azimuth compression and motion compensation processing, and the registration includes pixel-level or sub-pixel-level registration of images in each channel.