Method, device and equipment for denoising three-dimensional point cloud of array interferometric synthetic aperture radar

Abstract

The application discloses a method, device and equipment for denoising of three-dimensional point cloud of array interferometric synthetic aperture radar (SAR), and belongs to the technical field of SAR data processing and application. The method comprises the following steps: acquiring array interferometric SAR image data and multi-pass SAR point cloud; screening point cloud to be processed; calculating point cloud dispersion coefficient of k neighborhood and eliminating outlier noise points; performing K-means clustering, eliminating outlier noise clusters by using a threshold segmentation algorithm; acquiring a surface normal vector of each sampling point by using a principal component analysis method; estimating curvature of any sampling point in the array interferometric SAR point cloud; removing near-ground object noise by using an improved bilateral filtering algorithm; performing surface fitting, and extracting points on the surface of the surface as fitted ground points; rasterizing the fitted ground points, and removing mirror image noise by performing grid coordinate correspondence between the grid and points in the SAR point cloud. The application performs multi-element denoising on the array interferometric SAR point cloud, and improves the precision.

Description

Technical Field

[0001] This invention relates to a method, apparatus, and equipment for denoising three-dimensional point clouds in array interferometric synthetic aperture radar, belonging to the field of data processing and application technology of array interferometric synthetic aperture radar (SAR). Background Technology

[0002] Compared to synthetic aperture radar (SAR) two-dimensional imaging, SAR three-dimensional imaging can eliminate target overlay caused by radar side-looking imaging, significantly improving target resolution and identification capabilities, and is one of the important directions for the development of SAR technology. Array interferometric SAR (ASAR) technology utilizes a cross-heading array antenna for side-looking imaging. Based on the acquired multi-angle observation data, it achieves three-dimensional reconstruction of complex scenes through ASAR three-dimensional imaging, obtaining a regional three-dimensional point cloud. However, due to the influence of weather, environment, equipment, and imaging methods, ASAR point clouds may contain some noise and erroneous points, affecting further applications of ASAR point clouds.

[0003] To obtain higher quality array interferometric SAR point clouds, all point clouds need to be denoised before further processing. Common point cloud denoising methods include: (1) a single-level filtering method for all noise. This denoising method takes into account all noise and achieves good denoising effect visually. However, it may also lead to over-processing of some areas, resulting in the loss of point cloud detail features. (2) a hierarchical multi-level denoising method that processes different noises separately. This method considers the distribution patterns and characteristics of different noises and uses corresponding filtering algorithms in a targeted manner, so that detail features are well preserved. In addition, there is a denoising method that divides the region, that is, the point cloud data to be denoised is divided according to the scene, and different denoising algorithms are used for different scenes, which also improves the denoising effect to a certain extent.

[0004] Unlike other point cloud acquisition methods, the point cloud data acquired by array interferometric SAR 3D imaging technology is mainly obtained by recalculating after registration with SAR images. Furthermore, because SAR is a coherent imaging method, array interferometric SAR point clouds contain numerous cluttered points and image noise due to multipath effects. Therefore, array interferometric SAR point clouds have a wide variety of noise points, resulting in poor performance of traditional point cloud denoising algorithms, and a large number of noise points still remain in the denoised point cloud. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a method, apparatus, and device for denoising three-dimensional point clouds in array interferometric synthetic aperture radar (SAR). This method solves the problem that existing point cloud denoising techniques cannot be directly applied to point clouds in array interferometric SAR, effectively improving the accuracy of point clouds.

[0006] The technical solution adopted by this invention to solve its technical problem is as follows:

[0007] In a first aspect, the present invention provides a method for denoising three-dimensional point clouds of array interferometric synthetic aperture radar, comprising the following steps:

[0008] Acquire array interferometric SAR image data and use array interferometric SAR three-dimensional imaging technology to obtain multi-strip array interferometric SAR point clouds;

[0009] Noisy array interferometric SAR point clouds are selected as the point clouds to be processed;

[0010] Determine the value of k for the neighborhood size, calculate the point cloud dispersion coefficient of the k-neighborhood of each sampling point in the point cloud to be processed, and use an adaptive threshold segmentation algorithm to eliminate outlier noise points;

[0011] K-means clustering was performed on the array interferometric SAR point cloud to eliminate outlier noise points. The final clustering result was obtained by setting the minimum inter-class distance (MinDistance), and an adaptive threshold segmentation algorithm was used to eliminate outlier noise clusters.

[0012] The surface normal vector of each sampling point is obtained by principal component analysis, and the normal vector is corrected by a Gaussian weighted average filter.

[0013] Principal component analysis is used to estimate the curvature at any sampling point in an array interferometric SAR point cloud.

[0014] The bilateral filtering factor is improved by utilizing the curvature of the sampling points and the corrected normal vector, and the improved bilateral filtering algorithm is used to remove near-ground noise.

[0015] The least squares method is used to fit the surface of the array interferometric SAR point cloud after eliminating outlier noise and near-ground object noise, and the points on the surface are extracted as the ground points for fitting.

[0016] The fitted ground points are rasterized, and the grid coordinates are mapped to the points in the array interferometric SAR point cloud. Mirror noise is removed based on the elevation values ​​of the raster.

[0017] As one possible implementation of this embodiment, the point cloud dispersion coefficient of the k-neighborhood of the sampling point is the average distance between the current sampling point and all points within its k-nearest neighbor range, calculated using the following formula:

[0018]

[0019] Where PCDC is the point cloud dispersion coefficient of the current sampling point, k is the neighborhood size, and qi∈X is the set of points in the k-neighborhood. For sampling point p and any neighboring point q i The Euclidean distance between them.

[0020] As one possible implementation of this embodiment, after performing K-means clustering on the array interferometric SAR point cloud with outlier noise points eliminated, the outlier noise clusters and ground feature points of the original point cloud are segmented and evaluated using formula (2):

[0021] M = Q1(b1-b) G ) 2 +Q2(b2-b G ) 2 (2)

[0022] in, This represents the total proportion of in-class point values ​​for each ground feature point cloud category. This represents the total proportion of in-class points for outlier noise clusters. This represents the average number of points within each class in the point cloud of ground features. This represents the average number of within-class points for outlier noise clusters.

[0023] As one possible implementation of this embodiment, the surface normal vector of each sampling point is:

[0024]

[0025] in, For point p in the k-neighborhood of the sampling point i The normal vector, For sampling point p and points p in its k-neighborhood i The Euclidean distance between them, σ n The influence factor of the average distance between points in the k-neighborhood on the normal vector.

[0026] As one possible implementation of this embodiment, the curvature at the sampling point is:

[0027]

[0028] ω i For sampling point p i The curvature at point p, λ1, λ2, λ3 are the eigenvalues ​​obtained by eigenvalue decomposition of the covariance matrix of the k-neighbor points of the sampling point, λ1 < λ2 < λ3, and the eigenvector corresponding to the smallest eigenvalue λ1 is the eigenvector of the sampling point p. i The normal vector.

[0029] As one possible implementation of this embodiment, the step of improving the bilateral filter factor using the sampling point curvature and the corrected normal vector, and removing near-ground noise using the improved bilateral filter algorithm, includes:

[0030] The bilateral filter factor α is improved by using the curvature of the sampling points and the optimized normal vector of the sampling points as parameters. i :

[0031]

[0032] Among them, W C W S Both are Gaussian kernel functions, representing the nearest neighbor pair of sampling points p. i Influence weight, For sampling point p i Smooth weights of distances to neighboring points, x = ||p i -p j ||, For sampling point p i Feature domain weights for distances to neighboring points, y = || <n i ,n j >-1||+ω i ;σ c Indicates sampling point p i The influence factor of the distance to each nearest neighbor on the point is generally taken as the radius of the neighborhood; σ s This indicates that each nearest neighbor point is at sampling point p. i The influence factor of the projected distance on the normal vector at a point on that point is generally taken as the standard deviation of the nearest neighbor points; when σ c Once determined, the smooth distance of the point cloud in the normal direction and σ s Proportional; n i ,n j Sampling point p i And the optimized normal vector at sampling point pj;

[0033] Based on the improved bilateral filter factor α i Calculate sampling point p i Geometric location after removing near-ground noise:

[0034] p′i=p i +α i ·n i (8).

[0035] As one possible implementation of this embodiment, the formula for surface fitting of the array interferometric SAR point cloud based on the least squares method to eliminate outlier noise and near-ground object noise is as follows:

[0036] z(x,y)=ax 2 +bxy+cy 2 +dx+ey+f (9)

[0037] Where a, b, c, d, e, and f are surface fitting coefficients.

[0038] Secondly, an embodiment of the present invention provides a device for denoising three-dimensional point clouds of array interferometric synthetic aperture radar, comprising:

[0039] The data acquisition module is used to acquire array interferometric SAR image data and obtain multi-strip array interferometric SAR point clouds using array interferometric SAR three-dimensional imaging technology;

[0040] The point cloud filtering module is used to filter noisy array interferometric SAR point clouds as point clouds to be processed.

[0041] The outlier noise point elimination module is used to determine the value of k for the neighborhood size, calculate the point cloud dispersion coefficient of the k-neighborhood of each sampling point in the point cloud to be processed, and eliminate outlier noise points using an adaptive threshold segmentation algorithm.

[0042] The outlier noise cluster elimination module is used to perform K-means clustering on the array interferometric SAR point cloud to eliminate outlier noise points. The final clustering result is obtained by setting the minimum inter-class distance (MinDistance), and the outlier noise clusters are eliminated using an adaptive threshold segmentation algorithm.

[0043] The normal vector acquisition module is used to obtain the surface normal vector of each sampling point through principal component analysis and to correct the normal vector using a Gaussian weighted average filter.

[0044] The curvature calculation module is used to estimate the curvature at any sampling point in an array interferometric SAR point cloud using principal component analysis.

[0045] The near-ground object noise removal module is used to improve the bilateral filter factor by utilizing the curvature of the sampling points and the corrected normal vector, and to remove near-ground object noise using the improved bilateral filter algorithm;

[0046] The surface fitting module is used to perform surface fitting on array interferometric SAR point clouds that have eliminated outlier noise and near-ground object noise based on the least squares method, and extract the points on the surface as the fitted ground points.

[0047] The mirror noise removal module is used to rasterize the fitted ground points, and to match the grid coordinates with the points in the array interferometric SAR point cloud, removing mirror noise based on the elevation values ​​of the grid.

[0048] Thirdly, an embodiment of the present invention provides a computer device comprising: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to implement the three-dimensional point cloud denoising method of array interferometric synthetic aperture radar as described above.

[0049] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program for executing the three-dimensional point cloud denoising method for array interferometric synthetic aperture radar as described above.

[0050] The beneficial effects of the technical solutions of the embodiments of the present invention are as follows:

[0051] This invention provides a method for denoising three-dimensional point clouds in interferometric synthetic aperture radar (ISA). It utilizes an airborne platform equipped with an ISA antenna to acquire ISA images, performs image registration and pixel-by-pixel resolution to obtain a multi-strip three-dimensional point cloud. The method analyzes the noise distribution in the ISA point cloud, classifying it into three categories: outlier noise, near-ground object noise, and image noise. A hierarchical noise removal method based on noise classification is constructed, employing different denoising algorithms for each of the three noise types to remove noise points from the ISA point cloud. This invention is applicable to solving the problem of difficult noise removal in ISA point clouds, eliminating gross errors and erroneous points, improving the geometric accuracy of ISA point cloud data, and facilitating the application of ISA products.

[0052] Considering that tomographic SAR technology is affected by the inherent characteristics of SAR technology, and shares similarities with array interferometric SAR technology, the hierarchical array interferometric SAR point cloud denoising method based on noise classification proposed in this invention is also applicable to solving the denoising problem of tomographic SAR point clouds. This invention uses hierarchical filtering based on noise classification, rather than applying a single filtering algorithm to all noise types, effectively avoiding the loss of surface details and sharp features caused by past noise in the denoised array interferometric SAR point cloud. The hierarchical filtering based on noise classification in this invention can effectively remove noise points in the point cloud while better preserving the feature information. This invention effectively eliminates coarse and erroneous points in array interferometric SAR point clouds, improves the accuracy of array interferometric SAR point clouds, and can further promote the application of array interferometric SAR technology in topographic mapping. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating a method for denoising three-dimensional point clouds in an array interferometric synthetic aperture radar according to an exemplary embodiment;

[0054] Figure 2 This is a schematic diagram illustrating point cloud data obtained based on array interferometric SAR technology according to an exemplary embodiment;

[0055] Figure 3 This is a structural block diagram illustrating a point cloud data acquisition method based on array interferometric SAR technology, according to an exemplary embodiment.

[0056] Figure 4 This is a schematic diagram illustrating a surface fitting-based mirror noise removal according to an exemplary embodiment;

[0057] Figures 5(a) and 5(b) are comparative images of the effect of applying the present invention to denoise three-dimensional point clouds of array interferometric synthetic aperture radar according to an exemplary embodiment (wherein, Figure 5(a) is a schematic diagram of the point cloud before denoising, and Figure 5(b) is a schematic diagram of the result after denoising by the algorithm of the present invention).

[0058] The present invention will be further described below with reference to the accompanying drawings and embodiments: Detailed Implementation

[0059] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings. Many different embodiments or examples are disclosed below to implement different structures of the invention. To simplify the disclosure of the invention, components and arrangements of specific examples are described below. Furthermore, reference numerals and / or letters may be repeated in different examples. This repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. It should be noted that the components illustrated in the drawings are not necessarily drawn to scale. Descriptions of well-known components and processing techniques and processes are omitted in this invention to avoid unnecessarily limiting the invention.

[0060] The purpose of this invention is to solve the problem that existing point cloud denoising techniques cannot be directly applied to array interferometric SAR point clouds. This invention fully analyzes the noise characteristics of array interferometric SAR point clouds and removes point cloud noise by applying different filtering algorithms to different types of noise. This effectively solves the problem of denoising array interferometric SAR point clouds, removes noise points from array interferometric SAR point clouds, and effectively improves the accuracy of array interferometric SAR point clouds.

[0061] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for denoising three-dimensional point clouds of array interferometric synthetic aperture radar, which includes the following steps:

[0062] Acquire array interferometric SAR image data and use array interferometric SAR three-dimensional imaging technology to obtain multi-strip array interferometric SAR point clouds, such as... Figure 2 As shown;

[0063] Noisy array interferometric SAR point clouds are selected as the point clouds to be processed;

[0064] Determine the value of k for the neighborhood size, calculate the point cloud dispersion coefficient of the k-neighborhood of each sampling point in the point cloud to be processed, and use an adaptive threshold segmentation algorithm to eliminate outlier noise points;

[0065] K-means clustering was performed on the array interferometric SAR point cloud to eliminate outlier noise points. The final clustering result was obtained by setting the minimum inter-class distance (MinDistance), and an adaptive threshold segmentation algorithm was used to eliminate outlier noise clusters.

[0066] The surface normal vector of each sampling point is obtained by principal component analysis, and the normal vector is corrected by a Gaussian weighted average filter.

[0067] Principal component analysis is used to estimate the curvature at any sampling point in an array interferometric SAR point cloud.

[0068] The bilateral filtering factor is improved by utilizing the curvature of the sampling points and the corrected normal vector, and the improved bilateral filtering algorithm is used to remove near-ground noise.

[0069] The least squares method is used to fit the surface of the array interferometric SAR point cloud after eliminating outlier noise and near-ground object noise, and the points on the surface are extracted as the ground points for fitting.

[0070] The fitted ground points are rasterized, and the grid coordinates are mapped to the points in the array interferometric SAR point cloud. Mirror noise is removed based on the elevation values ​​of the raster.

[0071] As one possible implementation of this embodiment, the point cloud dispersion coefficient of the k-neighborhood of the sampling point is the average distance between the current sampling point and all points within its k-nearest neighbor range, calculated using the following formula:

[0072]

[0073] Where PCDC is the point cloud dispersion coefficient of the current sampling point, k is the neighborhood size, and q i Let X be the set of points in the k-neighborhood. For sampling point p and any neighboring point q i The Euclidean distance between them.

[0074] As one possible implementation of this embodiment, after performing K-means clustering on the array interferometric SAR point cloud with outlier noise points eliminated, the outlier noise clusters and ground feature points of the original point cloud are segmented and evaluated using formula (2):

[0075] M = Q1(b1-b) G ) 2 +Q2(b2-b G ) 2 (2)

[0076] in, This represents the total proportion of in-class point values ​​for each ground feature point cloud category. This represents the total proportion of in-class points for outlier noise clusters. This represents the average number of points within each class in the point cloud of ground features. This represents the average number of within-class points for outlier noise clusters.

[0077] As one possible implementation of this embodiment, the surface normal vector of each sampling point is:

[0078]

[0079] in, For point p in the k-neighborhood of the sampling point i The normal vector, For sampling point p and points p in its k-neighborhood i The Euclidean distance between them, σ n The influence factor of the average distance between points in the k-neighborhood on the normal vector.

[0080] As one possible implementation of this embodiment, the curvature at the sampling point is:

[0081]

[0082] ω i For sampling point p i The curvature at point p, λ1, λ2, λ3 are the eigenvalues ​​obtained by eigenvalue decomposition of the covariance matrix of the k-neighbor points of the sampling point, λ1 < λ2 < λ3, and the eigenvector corresponding to the smallest eigenvalue λ1 is the eigenvector of the sampling point p. i The normal vector.

[0083] As one possible implementation of this embodiment, the step of improving the bilateral filter factor using the sampling point curvature and the corrected normal vector, and removing near-ground noise using the improved bilateral filter algorithm, includes:

[0084] The bilateral filter factor α is improved by using the curvature of the sampling points and the optimized normal vector of the sampling points as parameters. i :

[0085]

[0086] Among them, W C W S Both are Gaussian kernel functions, representing the nearest neighbor pair of sampling points p. i Influence weight, For sampling point p i Smooth weights of distances to neighboring points, x = ||p i -p j ||, For sampling point p i Feature domain weights for distances to neighboring points, y = || <ni ,n j >-1||+ω i ;σ c Indicates sampling point p i The influence factor of the distance to each nearest neighbor on the point is generally taken as the radius of the neighborhood; σ s This indicates that each nearest neighbor point is at sampling point p. i The influence factor of the projected distance on the normal vector at a point on that point is generally taken as the standard deviation of the nearest neighbor points; when σ c Once determined, the smooth distance of the point cloud in the normal direction and σ s Proportional; n i ,n j Sampling point p i and sampling point p j Optimized normal vector at the location;

[0087] Based on the improved bilateral filter factor α i Calculate sampling point p i Geometric location after removing near-ground noise:

[0088] p i '=p i +α i ·n i (8).

[0089] As one possible implementation of this embodiment, the formula for surface fitting of the array interferometric SAR point cloud based on the least squares method to eliminate outlier noise and near-ground object noise is as follows:

[0090] z(x,y)=ax 2 +bxy+cy 2 +dx+ey+f (9)

[0091] Where a, b, c, d, e, and f are surface fitting coefficients.

[0092] This invention provides an apparatus for denoising three-dimensional point clouds in array interferometric synthetic aperture radar, comprising:

[0093] The data acquisition module is used to acquire array interferometric SAR image data and obtain multi-strip array interferometric SAR point clouds using array interferometric SAR three-dimensional imaging technology;

[0094] The point cloud filtering module is used to filter noisy array interferometric SAR point clouds as point clouds to be processed.

[0095] The outlier noise point elimination module is used to determine the value of k for the neighborhood size, calculate the point cloud dispersion coefficient of the k-neighborhood of each sampling point in the point cloud to be processed, and eliminate outlier noise points using an adaptive threshold segmentation algorithm.

[0096] The outlier noise cluster elimination module is used to perform K-means clustering on the array interferometric SAR point cloud to eliminate outlier noise points. The final clustering result is obtained by setting the minimum inter-class distance (MinDistance), and the outlier noise clusters are eliminated using an adaptive threshold segmentation algorithm.

[0097] The normal vector acquisition module is used to obtain the surface normal vector of each sampling point through principal component analysis and to correct the normal vector using a Gaussian weighted average filter.

[0098] The curvature calculation module is used to estimate the curvature at any sampling point in an array interferometric SAR point cloud using principal component analysis.

[0099] The near-ground object noise removal module is used to improve the bilateral filter factor by utilizing the curvature of the sampling points and the corrected normal vector, and to remove near-ground object noise using the improved bilateral filter algorithm;

[0100] The surface fitting module is used to perform surface fitting on array interferometric SAR point clouds that have eliminated outlier noise and near-ground object noise based on the least squares method, and extract the points on the surface as the fitted ground points.

[0101] The mirror noise removal module is used to rasterize the fitted ground points, and to match the grid coordinates with the points in the array interferometric SAR point cloud, removing mirror noise based on the elevation values ​​of the grid.

[0102] This invention denoises point clouds by classifying noise in array interferometric SAR point clouds and combining algorithms such as kd-tree, K-means clustering, adaptive thresholding, normal vector correction, improved bilateral filtering, and surface fitting. The overall denoising process is as follows: Figure 1 As shown:

[0103] Step S1: Acquire array interferometric SAR images using an array interferometric SAR antenna mounted on an airborne platform, and obtain multi-strip array interferometric SAR point clouds using array interferometric SAR three-dimensional imaging technology, such as... Figure 2 As shown.

[0104] Step S2: Select noisy array interferometric SAR point clouds as the point clouds to be processed, and classify the noise types into three categories: outlier noise, near-ground object noise, and image noise, such as... Figure 3 As shown.

[0105] Step S3: First, determine the size of the k-neighborhood. Then, take the average distance between the current sampling point and all points within its k-nearest neighbor range as the point cloud dispersion coefficient (PCDC) of the current sampling point. The calculation method is shown in formula (1):

[0106]

[0107] wherein k is the neighborhood size, q i ∈X is the point set within the k-neighborhood, is the Euclidean distance between the sampling point and any neighborhood point.

[0108] Step S4: define the intra-cluster point number sequence after point cloud clustering as P={P i |i=1,2,...,n}. Let P t =th, where t is the serial number index corresponding to the intra-cluster point number, then t can divide i into two parts: 1<i≤t and t<i≤n. Wherein, when 1<i≤t, P i represents the intra-cluster point number of the feature point cloud, and when t<i≤n, P i represents the intra-cluster point number of the outlier noise cluster. The proportion of each intra-cluster point number in the intra-cluster point number sequence of all categories is defined as then the total proportion of the intra-cluster point numbers of the feature point cloud categories is the total proportion of the intra-cluster point numbers of the outlier noise cluster categories is

[0109] the average intra-cluster point number of the feature point cloud is the average intra-cluster point number of the outlier noise cluster is the outlier noise clusters and feature points of the original point cloud are segmented and evaluated by formula (2):

[0110] M=Q1(b1-b G ) 2 +Q2(b2-b G ) 2 (2)

[0111] Step S5: calculate the normal vector by principal component analysis (PCA). For a given point set P={p i ,i∈1,2,...,n}, the k-neighborhood point set of p i ∈P is N(p i )={p i ,i∈1,2,...,k}, obtain the covariance matrix C of point p i :

[0112]

[0113] wherein, perform eigenvalue decomposition on matrix C to obtain eigenvalues λ1, λ2, λ3. If λ1<λ2<λ3, the eigenvector corresponding to the minimum eigenvalue λ1 is the normal vector n of point p i as the normal vector n i .

[0114] Based on the normal vector at the sampling point The normal vector of the centroid in its neighborhood The dot product adjusts the direction of the normal vector, as shown in the following formula:

[0115]

[0116] If the product of the centroid normal vectors of the normal vector neighborhood at the sampling point is greater than 0, then the normal vector at that point remains unchanged; if the product is less than 0, then the normal vector at the sampling point is in the opposite direction of the original normal vector.

[0117] To ensure continuous change of the normal vector within the k-neighborhood, a weighted Gaussian average filter is used as the normal vector correction model to adjust the normal vector of the array interferometric SAR point cloud after removing outlier noise clusters, as shown in Equation (5):

[0118]

[0119] in, Let be the normal vector of the points in the k-neighborhood of the sampling point. σ is the Euclidean distance between a sampling point and points in its k-neighborhood. n The influence factor of the average distance between points in the k-neighborhood on the normal vector.

[0120] Step S6: Principal component analysis can be used to estimate the curvature at any sampling point in the array interferometric SAR point cloud, point p. i curvature ω at i As shown in formula (6):

[0121]

[0122] Step S7, set the curvature ω of the sampling points i and the optimized sampling point normal vector n i As a parameter, the bilateral filter factor α is improved. i As shown in formula (7):

[0123]

[0124] Among them, W C W S Both are Gaussian kernel functions, representing the nearest neighbor pair p. i Influence weight, For point p i Smooth weights of distances to neighboring points, x = ||p i -p j ||, For point p i Feature domain weights for distances to neighboring points, y = || <n i ,nj >-1||+ω i ;σ c p i The influence factor of the distance to each nearest neighbor on the point is generally taken as the radius of the neighborhood; σ s Indicates that each nearest neighbor point is at p i The influence factor of the projected distance on the normal vector at a point on that point is generally taken as the standard deviation of the nearest neighbor points; when σ c Once determined, the smooth distance of the point cloud in the normal direction and σ s Proportional.

[0125] Improve the bilateral filter factor α i Substituting into formula (8), we can calculate point p. i Smoothed geometric position:

[0126] p i '=p i +α i ·n i (8).

[0127] Step S8, assuming the terrain surface is a spatial curved surface, this surface can be fitted using formula (9):

[0128] z(x,y)=ax 2 +bxy+cy 2 +dx+ey+f (9)

[0129] Where a, b, c, d, e, and f are surface fitting coefficients.

[0130] After fitting the local quadratic surface using the least squares method, the objective function is established as shown in formula (10):

[0131]

[0132] By combining equations (9) and (10) and solving the system of linear equations, we obtain the quadratic fitted surface equation shown in equation (11):

[0133]

[0134] Among them, z i Let z(x) be a point on the surface to be fitted. i ,y i ) for sampling point p i Points on the surface after quadratic fitting.

[0135] Step S9: For the raster data (digital elevation model) obtained through quadratic surface fitting, the point cloud within the lower vertical region of the raster surface is defined by the row and column numbers corresponding to each grid cell. Figure 4As shown, all points within the region are traversed. If the elevation value of a point is less than the elevation value of the raster data, the point is identified as a mirror noise point and removed.

[0136] This invention is applicable to solving the problem of noise reduction in three-dimensional point clouds of array interferometric SAR, improving the geometric accuracy of the obtained array interferometric SAR point clouds, and facilitating the mapping application of array interferometric SAR technology.

[0137] This invention effectively avoids the problem of some areas of noise not being removed or feature information being lost when using a single noise filtering method. Based on the data characteristics of array interferometric SAR point clouds, this invention classifies noise types, making it possible to use targeted denoising methods for different noise types, effectively removing noise while better preserving detailed feature information in the point cloud. This invention effectively improves the geometric accuracy of array interferometric SAR point clouds, further promoting the application of array interferometric SAR technology in surveying and mapping. The hierarchical three-dimensional point cloud denoising method based on noise classification described in this invention is applicable to point clouds generated by array interferometric SAR and tomographic SAR three-dimensional imaging techniques.

[0138] An experiment on point cloud denoising using the present invention was conducted, and the results are shown in Figure 5. The upper figure is a side view of the point cloud before denoising, and the lower figure is a side view of the point cloud after applying the algorithm of the present invention. Comparing the two figures in Figure 5, it can be seen that after applying the present invention, noise points in the point cloud were removed, and the quality of the point cloud was greatly improved.

[0139] An embodiment of the present invention provides a computer device, comprising: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to implement the three-dimensional point cloud denoising method of array interferometric synthetic aperture radar as described above.

[0140] The aforementioned memory and processor are all general-purpose memory and processor, and no specific limitations are made here. When the processor runs the computer program stored in the memory, it can execute the above-mentioned method for denoising three-dimensional point clouds of array interferometric synthetic aperture radar.

[0141] Those skilled in the art will understand that the structure of a computer device does not constitute a limitation on the computer device, and may include more or fewer components than shown in the figure, or combine some components, or split some components, or have different component arrangements.

[0142] In some embodiments, the computer device may further include a touchscreen for displaying a graphical user interface (such as an application launch screen) and receiving user operations on the graphical user interface (such as launching an application). The touchscreen may include a display panel and a touch panel. The display panel may be configured as an LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode), or similar technology. The touch panel can collect user touch or non-touch operations on or near it and generate pre-set operation commands, such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near the touch panel. Additionally, the touch panel may include a touch detection device and a touch controller. The touch detection device detects the user's touch position and posture, and detects the signals generated by the touch operation, transmitting the signals to the touch controller. The touch controller receives touch information from the touch detection device, converts it into information that the processor can process, sends it to the processor, and can also receive and execute commands from the processor. Furthermore, the touch panel can be implemented using various types of technologies, such as resistive, capacitive, infrared, and surface acoustic wave, or any future-developed technology. Furthermore, the touch panel can cover the display panel. Users can operate on or near the touch panel, which covers the display panel, based on the graphical user interface displayed on the display panel. After detecting the operation on or near the touch panel, the touch panel transmits it to the processor to determine the user input. The processor then responds to the user input by providing corresponding visual output on the display panel. Additionally, the touch panel and display panel can be implemented as two separate components or integrated together.

[0143] Corresponding to the above application startup method, this embodiment of the invention also provides a computer-readable storage medium storing a computer program for executing the array interferometric synthetic aperture radar three-dimensional point cloud denoising method as described above.

[0144] The application launch device provided in this application embodiment can be specific hardware on the device or software or firmware installed on the device. The device provided in this application embodiment has the same implementation principle and technical effects as the foregoing method embodiments. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the foregoing method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0145] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0146] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0147] The modules described as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0148] In addition, the functional modules in the embodiments provided in this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0149] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0150] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.

[0151] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for denoising three-dimensional point clouds in array interferometric synthetic aperture radar, characterized in that, Includes the following steps: Acquire array interferometric SAR image data and use array interferometric SAR three-dimensional imaging technology to obtain multi-strip array interferometric SAR point clouds; Array interferometric SAR point clouds containing outlier noise, near-ground object noise, and image noise were selected as point clouds to be processed. Determine the value of k for the neighborhood size, calculate the point cloud dispersion coefficient of the k-neighborhood of each sampling point in the point cloud to be processed, and use an adaptive threshold segmentation algorithm to eliminate outlier noise points; K-means clustering was performed on the array interferometric SAR point cloud to eliminate outlier noise points. The final clustering result was obtained by setting the minimum inter-class distance, and an adaptive threshold segmentation algorithm was used to eliminate outlier noise clusters. The surface normal vector of each sampling point is obtained by principal component analysis, and the normal vector is corrected by a Gaussian weighted average filter. Principal component analysis is used to estimate the curvature at any sampling point in an array interferometric SAR point cloud. The bilateral filtering factor is improved by utilizing the curvature of the sampling points and the corrected normal vector, and the improved bilateral filtering algorithm is used to remove near-ground noise. The least squares method is used to fit the surface of the array interferometric SAR point cloud after eliminating outlier noise and near-ground object noise, and the points on the surface are extracted as the ground points for fitting. The fitted ground points are rasterized, and the grid coordinates are mapped to the points in the array interferometric SAR point cloud. Mirror noise is removed based on the elevation values ​​of the raster. The surface normal vector of each sampling point is: （5） in, For sampling points k Points within the neighborhood The normal vector, Sampling points p With k Points within the neighborhood The Euclidean distance between them for k The influence factor of the average distance between points in the neighborhood on the normal vector; The curvature at the sampling point is: （6） Sampling points curvature at that point , , For sampling points k The eigenvalues ​​obtained by eigenvalue decomposition of the neighborhood point covariance matrix. Minimum eigenvalue The corresponding feature vector is the sampling point. The normal vector; The improvement of the bilateral filter factor using the curvature of the sampling points and the corrected normal vector, and the removal of near-ground noise using the improved bilateral filter algorithm, includes: The bilateral filter factor is improved by using the curvature of the sampling points and the optimized normal vector of the sampling points as parameters. : （7） in, , Both are Gaussian kernel functions, representing the nearest neighbor pairs of sampling points. Influence weight, Sampling points Smooth weights of distances to neighboring points. , Sampling points Feature domain weights based on distance to neighboring points ; Indicates the sampling point to The influence factor of the distance of each nearest neighbor on the point; This indicates that each nearest neighbor point is at the sampling point. The influence factor of the projected distance on the normal vector at a point on that point; when Once determined, the smooth distance of the point cloud in the normal direction and Proportional; Sampling points and sampling points Optimized normal vector at the location; Based on the improved bilateral filter factor Calculate sampling points Geometric location after removing near-ground noise: （8）。 2. The method for denoising three-dimensional point clouds of array interferometric synthetic aperture radar according to claim 1, characterized in that, The sampling points k The point cloud dispersion coefficient of the neighborhood is the ratio of the current sampling point to its neighboring points. k The average distance between all points within the nearest neighbor range is calculated using the following formula: （1） Where PCDC is the point cloud dispersion coefficient of the current sampling point. k For the size of the neighborhood, for k The set of points in the neighborhood, Sampling points p With any neighboring point The Euclidean distance between them.

3. The method for denoising three-dimensional point clouds of array interferometric synthetic aperture radar according to claim 1, characterized in that, After performing K-means clustering on the array interferometric SAR point cloud with outlier noise points removed, the outlier noise clusters and ground feature points of the original point cloud are segmented and evaluated using formula (2): （2） in, This represents the total proportion of in-class point values ​​for each ground feature point cloud category. This represents the total proportion of in-class points for outlier noise clusters. This represents the average number of points within each class in the point cloud of ground features. This represents the average number of within-class points for outlier noise clusters.

4. The method for denoising three-dimensional point clouds of array interferometric synthetic aperture radar according to claim 1, characterized in that, The formula for surface fitting of array interferometric SAR point clouds that eliminates outlier noise and near-ground object noise based on the least squares method is as follows: （9） in, These are the surface fitting coefficients.

5. An apparatus for denoising three-dimensional point clouds of interferometric synthetic aperture radar (ISSAR), used to implement the method for denoising three-dimensional point clouds of ISAR as described in any one of claims 1 to 4, characterized in that, include: The data acquisition module is used to acquire array interferometric SAR image data and obtain multi-strip array interferometric SAR point clouds using array interferometric SAR three-dimensional imaging technology; The point cloud filtering module is used to filter noisy array interferometric SAR point clouds as point clouds to be processed. The outlier noise point elimination module is used to determine the value of k for the neighborhood size, calculate the point cloud dispersion coefficient of the k-neighborhood of each sampling point in the point cloud to be processed, and eliminate outlier noise points using an adaptive threshold segmentation algorithm. The outlier noise cluster elimination module is used to perform K-means clustering on the array interferometric SAR point cloud to eliminate outlier noise points. The final clustering result is obtained by setting the minimum inter-class distance, and the outlier noise clusters are eliminated using an adaptive threshold segmentation algorithm. The normal vector acquisition module is used to obtain the surface normal vector of each sampling point through principal component analysis and to correct the normal vector using a Gaussian weighted average filter. The curvature calculation module is used to estimate the curvature at any sampling point in an array interferometric SAR point cloud using principal component analysis. The near-ground object noise removal module is used to improve the bilateral filter factor by utilizing the curvature of the sampling points and the corrected normal vector, and to remove near-ground object noise using the improved bilateral filter algorithm; The surface fitting module is used to perform surface fitting on array interferometric SAR point clouds that have eliminated outlier noise and near-ground object noise based on the least squares method, and extract the points on the surface as the fitted ground points. The mirror noise removal module is used to rasterize the fitted ground points, and to match the grid coordinates with the points in the array interferometric SAR point cloud, removing mirror noise based on the elevation values ​​of the grid.

6. A computer device, characterized in that, include: The computer device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to implement the method for denoising three-dimensional point clouds of array interferometric synthetic aperture radar as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program for executing the method for denoising three-dimensional point clouds of array interferometric synthetic aperture radar as described in any one of claims 1-4.

Images