Coastal zone land and water terrain three-dimensional modeling method fusing active and passive remote sensing of unmanned aerial vehicles

Abstract

This invention relates to the field of marine surveying and mapping technology, specifically to a method for 3D modeling of coastal land and water topography that integrates active and passive remote sensing from unmanned aerial vehicles (UAVs). The method includes: denoising and interpolating airborne laser bathymetry point clouds to construct a continuous water depth field, registering it with orthorectified images and repairing flares; using the water depth field as a geometric prior, extracting spectral references for land areas, and performing depth-layered color correction for water areas; dividing shallow and deep water areas based on critical depths, performing offset compensation and channel ratio reconstruction respectively, and then performing smooth weighted fusion; extracting high-frequency structural information, constructing segmented detail injection weights to generate seamless land and water orthorectified images; performing weighted mean-centered normal optimization on the point cloud, generating an initial mesh through constrained Poisson reconstruction and alpha shape clipping; using radial basis function interpolation to repair holes, and outputting an integrated 3D model through slope-accurate texture mapping. This invention solves the problem of pseudo-surfaces caused by underwater image color degradation and inconsistency in point cloud normals.

Description

Technical Field

[0001] This invention relates to the field of marine surveying and mapping technology, specifically to a method for three-dimensional modeling of coastal water and land terrain by integrating active and passive remote sensing from unmanned aerial vehicles. Background Technology

[0002] The coastal zone is a key transitional zone for land-sea interaction. High-precision integrated land-sea 3D models are the core foundational data for marine resource development, ecological environment monitoring, and refined management of the coastal zone. With the rapid development of marine engineering construction, coastal wetland protection, and other fields, there is an urgent need for integrated land-sea 3D models with both geometric accuracy and texture detail to fill the data gap in cross-media scenarios.

[0003] Currently, traditional single data sources are insufficient to simultaneously meet the demands of large-scale coverage and high-precision modeling. The fusion of airborne laser bathymetry (active remote sensing) and UAV photogrammetry (passive remote sensing) has become the core technological approach in this field. However, due to the complexity of cross-media measurement environments, the fusion of the two still faces two major challenges: I. Underwater optical images commonly suffer from color distortion and texture degradation due to water absorption and scattering. Existing underwater image enhancement methods (such as multi-scale fusion based on color shift correction and adaptive histogram equalization) are insufficient in adjusting complex terrain textures in shallow water, and are prone to local color oversaturation or harsh water-land boundaries in extremely turbid waters without prior information.

[0004] II. ALB point cloud normal estimation exhibits significant bias in areas where land and water meet and in complex terrain. Existing 3D modeling and normal optimization methods (such as traditional moving least squares MLS, Poisson reconstruction, etc.) are susceptible to high-frequency noise, which can easily lead to directional perturbations, resulting in local abnormal undulations (pseudo-surfaces), geometric holes, and discontinuities between geometry and texture on the reconstructed model surface.

[0005] In summary, developing a 3D modeling method for coastal land and water topography that integrates active and passive UAV remote sensing, which can effectively overcome color degradation in underwater images and eliminate point cloud normal inconsistencies and local hole defects, is an important problem that urgently needs to be solved in the field of marine surveying and remote sensing. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a method for three-dimensional modeling of coastal land and water terrain that integrates active and passive remote sensing from unmanned aerial vehicles (UAVs). This method aims to solve the problems of color degradation and texture blurring in underwater optical images under cross-medium measurement environments, as well as the occurrence of pseudo-surfaces and geometric holes in the reconstructed model due to inconsistencies in the normals of airborne laser bathymetry point clouds.

[0007] To address the above problems, the technical solution of this invention is as follows: a three-dimensional modeling method for coastal water and land terrain integrating active and passive remote sensing from unmanned aerial vehicles, comprising the following steps: S1. Perform denoising and interpolation on the original airborne laser bathymetry point cloud, spatially register it with the UAV orthophoto and construct a continuous water depth field, generate a land-water mask, identify and repair water surface flares in the image, and output standardized input data. S2. Using the continuous water depth field as the geometric prior, the water-land masking branch process is performed, the inherent spectrum of the land area is extracted as the reference benchmark, and the water area is subjected to depth-layered color correction. S3. Based on the critical depth, the shallow water effective area and the deep water reconstruction area are divided, and the physical water depth driven offset compensation and the brightness translation constraint channel ratio reconstruction are performed respectively. The deep and shallow water compensation results are seamlessly integrated through depth-related smoothing weights. S4. Convert the color-corrected image to grayscale to extract high-frequency structural information, construct a segmented detail injection weight function, set a transition zone depth threshold to generate smooth fusion weights, and fuse the detail injection model with the color-compensated image to generate a seamless land and water orthophoto. S5. Perform weighted mean-centered normal optimization on the airborne laser bathymetry point cloud, generate an initial watertight mesh through constrained Poisson reconstruction, and perform patch-level clipping based on the alpha shape boundary model; S6. Implicit surface interpolation using radial basis functions is used to repair local holes, and texture mapping is performed based on the slope of the triangular facets to output an integrated 3D model of land and water.

[0008] Furthermore, in S1, constructing a continuous water depth field includes: The underwater discrete elevation point set is obtained by using natural neighbor interpolation. P lidar Reconstructed into a continuous underwater elevation function: , In the formula, z ( x (pixels) x Underwater elevation, F interp For interpolation operators; Introducing tidal level benchmark Z ref With the maximum water depth in the survey area D max A normalized depth field is constructed to quantify the degree of water body decay, serving as a geometric constraint for subsequent compensation. D ( x ): , In the formula, Ω water and Ω land They represent water and land areas respectively. Pixels after tide level correction xThe actual water depth.

[0009] Furthermore, in S1, the restoration of water surface flares in the image includes: The image is converted to a hue-saturation-brightness space. An initial flare mask is constructed based on low-saturation and high-brightness features. The RGB channel statistical median of the pixels in the neighboring water areas of the flare is extracted to replace the benchmark. The mask is Gaussian smoothed to obtain continuous fusion weights, completing the local soft replacement. , In the formula, I raw ( x () represents the original pixel value. I ′ raw ( x () represents the pixel value after the flare is replaced. The median of the RGB channels of the pixels in the neighborhood of the flare. For continuous fusion weights.

[0010] Furthermore, in S2, performing deep layered color correction includes: The water depth is divided into multiple levels, the RGB mean values ​​of each water layer are statistically analyzed, the RGB three-channel reference mean values ​​of the dry land area are extracted, a depth-related color offset is defined and one-dimensional smoothing is applied to obtain a continuous compensation function: , In the formula, To stratify water depth, k For layer sequence number, Stratification of water depth The corresponding RGB three-channel color compensation offset, The RGB average value of each water layer. This is the reference average value for the three RGB channels in arid land areas.

[0011] Furthermore, in S3, the shallow water effective zone employs offset compensation driven by physical water depth: , In the formula, Image after compensation for shallow water area. I c ( x () represents the pixel values ​​of the image after flare removal. For pixels x The actual water depth at the location To compensate for the offset The actual water depth obtained by one-dimensional smooth interpolation Continuous three-channel color compensation vector; The smooth brightness component of the original image is extracted from the deep-water reconstruction region, and the channel ratio is constrained by the prior of the land spectrum to construct the compensation component: , In the formula, Reconstructed and compensated pixel RGB vectors for deep water areas. Smooth the brightness component of the original image. Critical depth Brightness compensation amount This is the land area reference mean. It is a scalar average of the land area reference mean.

[0012] Furthermore, in S3, the seamless integration of deep and shallow water compensation results is as follows: , In the formula, J c ( x (This refers to the merged color-compensated image.) For depth-dependent Bezier smoothing weight functions, This is an intermediate calculation term for the deep-water compensation component after being weighted by Bessel smoothing.

[0013] Furthermore, in S4, the piecewise detail injection weight function is: , In the formula, k 0 and k 1 is the control factor. η For deep-water structure fidelity magnification factor, This is the critical depth.

[0014] Furthermore, in S5, the weighted mean-centered normal optimization includes: constructing a comprehensive weight function that integrates spatial proximity, normal similarity, and curvature characteristics within the moving least squares framework. : , In the formula, κ i For local curvature, λ For curvature adjustment factor, p i 、p j The first in the airborne laser bathymetry point cloud i , No. j Three-dimensional point p i This is the current area for optimization. p j yes p i neighborhood points, d ij For point p i andp j Spatial distance between σ d The standard deviation of the distance from the weight. n i and n j Let be the initial normal vector. σ n The standard deviation of normal similarity. This is the normal similarity weight term.

[0015] Furthermore, in S6, the implicit surface interpolation of the radial basis function includes: assuming that the sampling point at the boundary of the hole is the center point of the radial basis function. c j The corresponding normal direction is n i Construct implicit functions combining radial basis functions and linear terms. f ( p ): , In the formula, M For the number of RBF center points, For radial basis kernel functions, w j As weight, c j The center point of the radial basis functions, The coefficients of the linear term, b For constant terms, p is the position vector of the three-dimensional point to be evaluated.

[0016] Furthermore, in S6, texture mapping based on the slope of the triangular facet includes, when the slope angle of the triangular facet is greater than a preset slope threshold, texture mapping is performed using a local tangent plane UV parameterization method. The UV calculation formula for texture mapping is: , In the formula, u、v For two-dimensional texture coordinates, u a、 v a Anchor coordinates for the texture corresponding to the centroid. s The scale factor from physical length to texture coordinates. t 1, t 2 is a local tangential basis vector. Let be any point on the surface. c The center of mass of the sheet.

[0017] Compared with the prior art, the present invention has the following beneficial effects: Using airborne laser bathymetry to invert water depth as a global geometric prior constraint, a dual-track color correction system with differences between shallow and deep water was established. First, a continuous color compensation benchmark was established based on layered spectral statistics. Then, two types of correction logics were distinguished: shallow water offset compensation and deep water brightness constraint reconstruction. Bezier smoothing weights were used to achieve seamless color transition between shallow and deep water, and between land and water areas. At the same time, underwater degradation high-frequency textures were dynamically amplified based on water depth segmentation weights. The measured underwater image quality index (UIQM) was significantly improved, the color difference between land and water was greatly reduced, and the visual consistency and detail fidelity of coastal zone images were greatly improved.

[0018] Within the framework of moving least squares, a comprehensive weighting function is constructed by integrating spatial distance, normal similarity, and local curvature features. This function performs weighted mean-centered normal optimization on airborne laser bathymetry point clouds, unifying and regularizing the global normal field of the point clouds. This effectively suppresses normal disturbances in point clouds in complex coastal terrain and water-land interface areas. Combined with global Poisson reconstruction and alpha shape patch boundary clipping, an initial watertight mesh that fits the real shoreline is obtained. Then, radial basis function implicit surface interpolation is used to seamlessly fill local holes in the mesh. Multiple geometric processing methods work together to eliminate the problems of pseudo-surfaces, boundary expansion, and a large number of geometric holes generated by traditional reconstruction. The surface smoothness, topological continuity, and overall geometric accuracy of the model are significantly improved.

[0019] A slope-level texture mapping scheme is adopted. Normalized orthophoto projection is used to quickly match textures in gentle grid areas. Steep slopes exceeding the threshold are constructed separately with local orthogonal tangent planes and independent UV parameterization is performed to solve the problem of texture stretching and distortion on steep slopes. High-fidelity seamless orthophoto textures are accurately fitted to a hole-free continuous 3D mesh, realizing model-level integration of terrain geometry and image texture. The output integrated land and water scene 3D model is geometrically accurate and has natural and distortion-free textures.

[0020] The entire modeling process fully covers the entire chain of point cloud preprocessing, image correction and enhancement, 3D mesh reconstruction, and texture mapping. It can complete automated modeling by integrating only two types of conventional active and passive remote sensing data sources: airborne laser depth sounding and UAV orthophotos. No additional multi-source auxiliary data is required. It is suitable for complex cross-media scenarios such as large-scale coastal zones, tidal flats, and reefs. It can be directly applied to business scenarios such as marine resource surveys, coastal wetland ecological protection, nearshore disaster prevention and mitigation, and refined management of coastal engineering. It has strong engineering practicality. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the three-dimensional modeling method of the present invention; Figure 2 This is a flowchart of the three-dimensional modeling method of the present invention; Figure 3 This is a schematic diagram illustrating the principle of local median soft fusion of water surface flares in this invention; Figure 4This is a schematic diagram illustrating the principle of depth-dependent color attenuation and color shift in this invention; Figure 5 This is a schematic diagram of the high-frequency detail enhancement mechanism based on geometric depth in this invention; Figure 6 This is a schematic diagram illustrating the principle of normal vector calculation based on multi-feature comprehensive weight optimization in this invention; Figure 7 This is a schematic diagram of the implicit curved hole repair principle based on RBF in this invention; Figure 8 This is a schematic diagram illustrating the principle of patch-level precise mapping and texture shading in this invention; Figure 9 The figure shows the three-dimensional model of the integrated land and water system in this invention. In the figure, (a) is a modeling effect of the land-water transition zone, (b) is a modeling effect of the deep water zone, and (c) is a modeling effect of the land area. Detailed Implementation

[0022] This invention provides a method for 3D modeling of coastal water and land terrain based on active and passive remote sensing from unmanned aerial vehicles (UAVs). The method flow is as follows: Figure 1 , Figure 2 As shown, this paper addresses the challenges of color degradation in underwater images under cross-medium environments, inconsistencies in the normals of airborne laser bathymetry point clouds, and localized holes that restrict the accuracy of 3D modeling. Using water depth inverted from airborne laser bathymetry as a geometric prior, a depth-driven statistical color correction and texture enhancement model is constructed to achieve spectral consistency correction between land and water. A weighted mean-centered pseudo-surface suppression and normal consistency optimization algorithm is proposed, combined with topological consistency-constrained Poisson reconstruction and implicit surface interpolation, to construct a continuous watertight mesh. The specific steps are as follows: S1. Perform denoising and interpolation on the original airborne laser bathymetry point cloud, spatially register it with the UAV orthophoto and construct a continuous water depth field, generate a land-water mask, identify and repair water surface flares in the image, and output standardized input data. S2. Using the continuous water depth field as the geometric prior, the water-land masking branch process is performed, the inherent spectrum of the land area is extracted as the reference benchmark, and the water area is subjected to depth-layered color correction. S3. Based on the critical depth, the shallow water effective area and the deep water reconstruction area are divided, and the physical water depth driven offset compensation and the brightness translation constraint channel ratio reconstruction are performed respectively. The deep and shallow water compensation results are seamlessly integrated through depth-related smoothing weights. S4. Convert the color-corrected image to grayscale to extract high-frequency structural information, construct a segmented detail injection weight function, set a transition zone depth threshold to generate smooth fusion weights, and fuse the detail injection model with the color-compensated image to generate a seamless land and water orthophoto. S5. Perform weighted mean-centered normal optimization on the airborne laser bathymetry point cloud, generate an initial watertight mesh through constrained Poisson reconstruction, and perform patch-level clipping based on the alpha shape boundary model; S6. Implicit surface interpolation using radial basis functions is used to repair local holes, and texture mapping is performed based on the slope of the triangular facets to output an integrated 3D model of land and water.

[0023] Specifically, a method for 3D modeling of coastal water and land terrain that integrates active and passive remote sensing from unmanned aerial vehicles includes the following steps: S1. Data Preprocessing: Denoise and interpolate the original airborne laser bathymetry point cloud, spatially register it with the UAV orthophoto and construct a continuous water depth field, generate a land-water mask, identify and repair water surface flares in the image, and output standardized input data.

[0024] The purpose of this step is to provide uniform, clean, and spatially aligned standardized input data for subsequent color correction and 3D reconstruction. This includes the following sub-steps: S101. Construction of Continuous Water Depth Field: For the discrete original ALB point cloud, denoising is first performed to remove outliers caused by waves, suspended objects, and system noise; then, natural neighbor interpolation is used to construct the discrete underwater elevation point set. P lidar Reconstructed as a pixel-by-pixel continuous underwater elevation function: , In the formula, For pixels x Underwater elevation, As an interpolation operator, natural neighbor interpolation can adaptively adjust the interpolation neighborhood based on the local point density, maintaining the continuity and local fidelity of the elevation surface in coastal areas with uneven point cloud density, and avoiding false undulations.

[0025] S102, Construction of normalized depth field; introduction of tidal baseline. Z ref With the maximum water depth in the survey area D max A normalized depth field is constructed to quantify the degree of water body decay, serving as a geometric constraint for subsequent compensation. D ( x ): , In the formula, Ω water and Ω land They represent water and land areas respectively. Pixels after tide level correction xThe normalized depth field serves as a water-land mask: water pixels are assigned continuous depth values ​​between 0 and 1, while land pixels are assigned a value of 0, thus providing a unified geometric prior for subsequent water-land mask branching. At the same time, the continuous depth field is spatially registered with the UAV orthophoto based on a unified geographic coordinate system, so that each image pixel can be indexed to the corresponding depth value.

[0026] S103, Water flare restoration; To avoid confusion between the beach and flares in the RGB color space, the UAV orthophoto image was converted to the Hue / Saturation / Lightness (HSV) space, and an initial flare mask was constructed based on low saturation and high lightness features; after morphological expansion and masking to cover the halo, as shown... Figure 3 As shown, the RGB channel median of pixels in the neighboring water area of ​​the flare is extracted. C med Alternative benchmarks are used to obtain continuous fusion weights by Gaussian smoothing of the mask. W(x) Complete the local soft replacement: , In the formula, These are the original pixel values. The pixel values ​​after the flare is replaced. The median of the RGB channels of the pixels in the neighborhood of the flare. W(x) The fusion weights are continuous; because the fusion weights, after Gaussian smoothing, approach 1 inside the mask, approach 0 outside the mask, and transition continuously at the boundary, the flare region is "soft-replaced" by the statistical median of the neighboring water body. This removes the bright flares on the water surface and avoids the boundary marks caused by hard mask replacement. At this point, the standardized input data is output, including the registered restored image. Normalized depth field D(x) And water and land shelters.

[0027] S2. Using the continuous water depth field as the geometric prior, the water and land masking is processed, the inherent spectrum of the land is extracted as the reference, and the water is subjected to depth-layered color correction.

[0028] This step uses the continuous water depth field output by S1 as the geometric prior, performs branching processing according to the land-water mask, and establishes a depth-driven color attenuation statistical model, specifically including: Deep layered color correction and continuous compensation function construction; in the terrestrial branch, extract the reference mean of the three RGB channels of arid terrestrial land. As an approximation of the unattenuated intrinsic spectrum; in the water branch, the water depth is divided into multiple levels. Statistical analysis of the RGB mean values ​​of each water layer ,like Figure 4 As shown, a depth-dependent color offset is defined, and a one-dimensional smoothing is applied to the offsets at discrete levels to obtain a continuous compensation function: , In the formula, To stratify water depth, k For layer sequence number, Stratification of water depth The corresponding RGB three-channel color compensation offset, The RGB average value of each water layer. The RGB three-channel reference mean is used for dry land areas. This continuous compensation function describes the statistical law of color attenuation with water depth under the effects of water absorption and scattering, providing a depth-by-depth benchmark for subsequent compensation.

[0029] S3. Based on the critical depth, the shallow water effective area and the deep water reconstruction area are divided. Physical water depth driven offset compensation and brightness translation constraint channel ratio reconstruction are performed respectively. The deep and shallow water compensation results are seamlessly integrated through depth-related smoothing weights.

[0030] Considering the strong nonlinear attenuation of light waves in water, and to avoid channel overflow and color shift caused by global offset compensation in deep water areas, this invention proposes a dual-track color correction model, specifically including: S301. Dynamic determination of critical depth: Statistically determine the 95th percentile of pixels in dark water bodies to dynamically determine the critical depth at which offset compensation fails. z crit This critical depth precisely divides the water area into a shallow effective zone and a deep reconstructed zone, avoiding the subjectivity of manually setting the boundary depth.

[0031] S302, Shallow water effective zone offset compensation; for actual water depth d ( x )≤ z crit In the shallow water effective zone, offset compensation driven by physical water depth is used to obtain the shallow water compensation component: , In the formula, Image after compensation for shallow water area. I c ( x () represents the pixel values ​​of the image after flare removal. For pixels x The actual water depth at the location To compensate for the offset The actual water depth obtained by one-dimensional smooth interpolation The continuous three-channel color compensation vector; the water attenuation in shallow water areas is relatively light, and offset compensation can achieve natural color restoration.

[0032] S303, deep-water reconstruction zone channel ratio reconstruction; tailored to actual water depth d (x )≥ z crit Extract the smooth brightness component from the deep-water reconstruction region of the original image. L blur ( x The brightness is shifted to the critical depth level, and the channel ratio is constrained a priori by the terrestrial spectrum to construct a compensation component: , In the formula, Reconstructed and compensated pixel RGB vectors for deep water areas. Smooth the brightness component of the original image. Critical depth Brightness compensation amount This is the land area reference mean. The scalar average of the land reference mean is used. This reconstruction method no longer performs channel-by-channel offset compensation for the deep water area. Instead, it restores the overall brightness by shifting the brightness and restores the color balance by using the a priori channel ratio of the land spectrum. This fundamentally avoids channel overflow and color shift caused by offset compensation in the deep water area.

[0033] S304, seamless integration of shallow and deep water compensation results; introduction of depth-related Bezier smoothing weights. W(d(x)) This achieves seamless fusion of shallow and deep water compensation results, resulting in a color-compensated image. , In the formula, J c ( x (This refers to the merged color-compensated image.) W(d(x)) For depth-dependent Bezier smoothing weight functions, G(x) This is an intermediate calculation term for the deep-water compensation component after being weighted by Bessel smoothing. The Bessel smoothing weight continuously and monotonically increases from 0 to 1 near the critical depth, so that the shallow-water offset compensation and the deep-water ratio reconstruction are smoothly weighted and connected in the transition zone, eliminating the color abrupt change at the boundary.

[0034] S4. Convert the color-corrected image to grayscale to extract high-frequency structural information, construct a segmented detail injection weight function, set a transition zone depth threshold to generate smooth fusion weights, and fuse the detail injection model with the color-compensated image to generate a seamless land and water orthophoto.

[0035] This step, building upon color correction, further restores high-frequency texture details degraded underwater, specifically including: S401. High-frequency structure information extraction: Convert the color-corrected image to grayscale and extract high-frequency structure information using Gaussian filtering. I ( xThe high-frequency component, which is the difference between the grayscale image and its Gaussian smoothed result, carries the texture and structural details of underwater features.

[0036] S402. Construction of a segmented detail injection weight function; A segmented detail injection weight function is constructed to address the texture degradation characteristics of different depth domains. Ω (d(x)): , In the formula, k 0 and k 1 is the control factor. η For deep-water structure fidelity magnification factor, The critical depth is reached; texture degradation is less severe in shallow water areas, so a smaller base injection intensity is used. Texture attenuation is severe in deep water areas, and the injection weight increases with depth. η Magnify, thereby dynamically amplifying the high-frequency textures that are degraded underwater.

[0037] S403, Transition band smoothing and blending weight generation and detail injection; setting transition band depth threshold. z shore Generate 0≤ using a smooth step function d ( x )≤ z shore Fusion weights increasing from 0 to 1 within the range W shore ( x This process smoothly transitions the land area (weight 0, preserving the original texture) to the water area (weight 1, injecting enhanced texture); the detail-injected model is then fused with the color-compensated image to obtain the final seamless enhanced orthophoto of land and water. , In the formula, To achieve the final seamless augmented orthophoto of land and water, For color compensation images, For smooth transition band fusion weights, Inject weight functions into the piecewise details. For high-frequency structural information; such as Figure 5 As shown, this mechanism ensures a smooth and seamless transition between the tones and textures of the water and land areas.

[0038] S5. Perform weighted mean-centered normal optimization on the airborne laser sounding point cloud, generate an initial watertight mesh through constrained Poisson reconstruction, and perform patch-level clipping based on the alpha shape boundary model.

[0039] This step addresses the pseudo-surface problem caused by inconsistent normals in ALB point clouds, and constructs an initial watertight mesh that fits the real shoreline. Specifically, it includes: S501, Construction of Multi-Feature Comprehensive Weight Function; such as Figure 6 The principle diagram of point cloud normal optimization shows that a comprehensive weight function integrating spatial proximity, normal similarity, and curvature features is constructed within the moving least squares (MLS) framework. : , In the formula, κ i For local curvature, λ For curvature adjustment factor, p i 、p j The first in the airborne laser bathymetry point cloud i , No. j Three-dimensional point p i This is the current area for optimization. p j yes p i neighborhood points, d ij For point p i and p j Spatial distance between σ d The standard deviation of the distance from the weight. n i and n j Let be the initial normal vector. σ n The standard deviation of normal similarity. The first term, Gaussian factor, ensures the dominant contribution of spatially neighboring points; the second term, Gaussian factor, suppresses the participation of neighboring points with excessively large normal differences in estimation, preventing the normals in areas with abrupt changes in normals, such as the boundary between land and water, from being erroneously averaged; the third term, curvature factor, enhances the weight of high curvature feature areas, preserving terrain details.

[0040] S502, Weighted mean centering normal optimization; weighted mean centering smoothing of point cloud based on comprehensive weights: using comprehensive weights We perform weighted mean centering on the neighborhood points, re-estimate the local tangent plane of each sampling point and update the normal, and iteratively obtain a uniform and regular optimized normal field, which effectively suppresses the normal perturbation of point clouds in complex coastal topography and water-land interface areas.

[0041] S503, Constrained Poisson Reconstruction: A global Poisson surface reconstruction method is adopted, which generates a continuous surface by solving the Poisson equation between the normal field gradient and the indicator function, and introduces a unified correction method for the patch orientation of the global reference direction. Avoid reverse normal direction.

[0042] S504. Based on the AlphaShape boundary model, perform patch-level trimming on the initial watertight mesh: construct an alpha shape boundary for the point cloud with a set alpha radius, delete pseudo surface patches located outside the boundary and generated by over-extension of Poisson reconstruction, and obtain a trimmed watertight mesh that fits the actual shoreline range.

[0043] S6. Implicit surface interpolation using radial basis functions is used to repair local holes, and texture mapping is performed based on the slope of the triangular facets to output an integrated 3D model of land and water.

[0044] This step completes hole repair and texture mapping, outputting the final model, specifically including: S601, RBF implicit curved hole repair; detecting local hole boundaries in the watertight mesh after clipping, targeting local model hole areas, such as... Figure 7 As shown, the implicit surface interpolation method using radial basis functions (RBF) is employed; the sampling points at the hole boundary are assumed to be the center points of the radial basis functions. c j The corresponding normal direction is n i Construct a combination of radial basis functions and linear terms that satisfy the constraints. f ( p The implicit function repairs the hole areas in the local model: , In the formula, M For the number of RBF center points, For radial basis kernel functions, w j As weight, c j The center point of RBF, The coefficients of the linear term, b For constant terms, p Let be the position vector of the three-dimensional point to be evaluated; the coefficients are determined by solving a system of linear equations consisting of the position constraints and normal constraints of the hole boundary points, and then... f(p) = 0 The generation of zero isosurfaces and the smooth, continuous repair patches of the surrounding curved surfaces enable seamless filling of holes, greatly improving the accuracy of watertight mesh closure and topological continuity.

[0045] S602, precise texture mapping for slope discrimination; such as Figure 8 As shown, to achieve high-precision mapping between enhanced orthophotos and reconstructed meshes, a normalized orthophoto projection relationship is established based on a unified geographic coordinate system in flat areas; to suppress texture distortion in steep slope areas, a normalized orthophoto projection relationship is established based on the normal vector of triangular patches. n f With global unit vector ez The slope is determined by the included angle, specifically the angle between the slope of the triangular facet and the horizontal plane. When the slope exceeds the set value At that time, texture mapping is performed by constructing orthogonal tangent plane basis vectors locally on the surface and using a surface-level independent UV parameterization method: In the formula, u、v For two-dimensional texture coordinates, u a , v a Anchor coordinates for the texture corresponding to the centroid. s The scale factor from physical length to texture coordinates. t 1, t 2 is a local tangential basis vector. Let be any point on the surface. c Using the centroid of the surface, this independent UV parameterization at the surface level ensures that the texture sampling of steep surfaces is no longer affected by orthographic compression, thus solving the problem of texture stretching and distortion on steep slopes.

[0046] S603, Model Output: The seamless enhanced orthophoto of land and water generated in S4 is used as the texture source and is applied to the repaired continuous watertight mesh surface according to the above mapping relationship. Finally, a land and water integrated real-world 3D model with highly integrated geometry and texture is output.

[0047] Experiment and Effect Analysis To verify the performance of the method of this invention, airborne LiDAR bathymetry data along the coast of location A and synchronously acquired UAV orthophotos were used for verification. The experiment employed the Mapper 20KU ALB system mounted on a multi-rotor UAV. Figure 9 The image shown is a 3D model of the integrated land and water system in this invention, where (a) is a modeling effect of the land-water transition zone, (b) is a modeling effect of the deep water zone, and (c) is a modeling effect of the land area. (1) In terms of image enhancement quality: the proposed depth-driven enhancement algorithm effectively removes the bright flares on the water surface; after applying the method of this embodiment, the UIQM index of the enhanced underwater image is improved to 4.86, the water-land color difference ΔE is reduced from 14.97 to 5.94 (a reduction of 60%), the underwater color restoration is natural, and the water-land transition is smooth.

[0048] (2) Regarding the accuracy of 3D reconstruction: The proposed normal optimization and RBF repair algorithms effectively improve the defects of boundary expansion, pseudo-surface and hole in traditional Poisson reconstruction. The overall fitting accuracy C2M RMSE is reduced from 0.23 m to 0.18 m (an improvement of about 21.7%), the surface roughness is reduced from 0.41 rad to 0.21 rad, and the number of holes is reduced from 584 in the traditional method to 45.

[0049] The above data fully demonstrate that the method achieves a significant improvement in geometric accuracy and topological integrity in complex coastal environments.

[0050] This invention also provides a 3D modeling system for coastal water and land topography that integrates active and passive remote sensing from unmanned aerial vehicles (UAVs). The system includes: (1) Data preprocessing module, used to interpolate ALB point cloud to construct a continuous water depth field and register it with the image, and to identify and repair water surface flares in the image; (2) Active and passive fusion color correction enhancement module, which is used to perform depth layering and dual-track color correction with water depth as geometric prior, and to inject high-frequency details of texture with weights; (3) The three-dimensional model construction module is used to perform weighted mean centering normal optimization and constrained Poisson reconstruction on the point cloud, use RBF interpolation to repair holes, and accurately map the enhanced orthophoto onto the three-dimensional mesh surface.

[0051] This invention integrates prior knowledge of water depth from airborne laser bathymetry to construct a depth-driven statistical color correction and high-frequency detail enhancement model, effectively solving the problems of color degradation in underwater images and interference from bright flares on the water surface in cross-medium environments. By optimizing the point cloud normal vector through multi-feature comprehensive weighting, it significantly reduces pseudo-surface defects caused by inconsistencies in point cloud normals and local sparsity. Combined with radial basis function interpolation, it effectively repairs small-scale geometric holes, greatly improving the accuracy and topological continuity of watertight meshes under complex terrain. Through slope-constrained patch-level accurate texture mapping, it achieves a high degree of geometric and textural fusion between enhanced orthophotos and reconstructed meshes, obtaining a high-fidelity, seamless, and continuous integrated land-water 3D model that can directly meet the urgent needs of key areas such as refined marine resource surveys, coastal ecological protection, and nearshore disaster prevention and mitigation.

[0052] The above specific embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to examples, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for integrated modeling of coastal water and land topography combining active and passive UAV remote sensing, characterized in that, Includes the following steps: S1. Perform denoising and interpolation on the original airborne laser bathymetry point cloud, spatially register it with the UAV orthophoto and construct a continuous water depth field, generate a land-water mask, identify and repair water surface flares in the image, and output standardized input data. S2. Using the continuous water depth field as the geometric prior, the water-land masking branch process is performed, the inherent spectrum of the land area is extracted as the reference benchmark, and the water area is subjected to depth-layered color correction. S3. Based on the critical depth, the shallow water effective area and the deep water reconstruction area are divided, and the physical water depth driven offset compensation and the brightness translation constraint channel ratio reconstruction are performed respectively. The deep and shallow water compensation results are seamlessly integrated through depth-related smoothing weights. S4. Convert the color-corrected image to grayscale to extract high-frequency structural information, construct a segmented detail injection weight function, set a transition zone depth threshold to generate smooth fusion weights, and fuse the detail injection model with the color-compensated image to generate a seamless land and water orthophoto. S5. Perform weighted mean-centered normal optimization on the airborne laser bathymetry point cloud, generate an initial watertight mesh through constrained Poisson reconstruction, and perform patch-level clipping based on the alpha shape boundary model; S6. Local holes are repaired by implicit surface interpolation using radial basis functions, and texture mapping is performed based on the slope of the triangular facets to output a three-dimensional model of integrated land and water scenery. In S3, the shallow water effective zone uses physical water depth-driven offset compensation: ； In the formula, Image after compensation for shallow water area. I c ( x () represents the pixel values ​​of the image after flare removal. For pixels x The actual water depth at the location To compensate for the offset The actual water depth obtained by one-dimensional smooth interpolation Continuous three-channel color compensation vector; The smooth brightness component of the original image is extracted from the deep-water reconstruction region, and the channel ratio is constrained by the prior of the land spectrum to construct the compensation component: ； In the formula, Reconstructed and compensated pixel RGB vectors for deep water areas. Smooth the brightness component of the original image. Critical depth Brightness compensation amount This is the land area reference mean. The scalar average of the land reference mean; In S3, the seamless integration of deep and shallow water compensation results is as follows: ； In the formula, J c ( x (This refers to the merged color-compensated image.) For depth-dependent Bezier smoothing weight functions, This is an intermediate calculation term for the deep-water compensation component after Bessel smoothing and weighting. In S4, the piecewise detail injection weight function is: ； In the formula, k 0 and k 1 is the control factor. η For deep-water structure fidelity magnification factor, This is the critical depth.

2. The method according to claim 1, characterized in that: In S1, constructing a continuous depth field includes: The underwater discrete elevation point set is obtained by using natural neighbor interpolation. P lidar Reconstructed into a continuous underwater elevation function: ； In the formula, z ( x (pixels) x Underwater elevation, F interp For interpolation operators; Introducing tidal level benchmark Z ref With the maximum water depth in the survey area D max A normalized depth field is constructed to quantify the degree of water body decay, serving as a geometric constraint for subsequent compensation. D ( x ): ； In the formula, Ω water and Ω land They represent water and land areas respectively. Pixels after tide level correction x The actual water depth.

3. The method according to claim 2, characterized in that: In S1, the restoration of water surface flares in the image includes: The image is converted to a hue-saturation-brightness space. An initial flare mask is constructed based on low-saturation and high-brightness features. The RGB channel statistical median of the pixels in the neighboring water areas of the flare is extracted to replace the benchmark. The mask is Gaussian smoothed to obtain continuous fusion weights, completing the local soft replacement. ； In the formula, I raw ( x () represents the original pixel value. I ′ raw ( x () represents the pixel value after the flare is replaced. The median of the RGB channels of the pixels in the neighborhood of the flare. For continuous fusion weights.

4. The method according to claim 3, characterized in that: In S2, performing depth-layered color correction includes: The water depth is divided into multiple levels, the RGB mean values ​​of each water layer are statistically analyzed, the RGB three-channel reference mean values ​​of the dry land area are extracted, a depth-related color offset is defined and one-dimensional smoothing is applied to obtain a continuous compensation function: ； In the formula, To stratify water depth, k For layer sequence number, Stratification of water depth The corresponding RGB three-channel color compensation offset, The RGB average value of each water layer. This is the reference average value for the three RGB channels in arid land areas.

5. The method according to claim 4, characterized in that: In S5, the weighted mean-centered normal optimization includes: constructing a comprehensive weight function that integrates spatial proximity, normal similarity, and curvature characteristics within the moving least squares framework. : ； In the formula, κ i For local curvature, λ For curvature adjustment factor, p i 、p j The first in the airborne laser bathymetry point cloud i , No. j Three-dimensional point p i This is the current area for optimization. p j yes p i neighborhood points, d ij For point p i and p j Spatial distance between σ d The standard deviation of the distance from the weight. n i and n j Let be the initial normal vector. σ n The standard deviation of normal similarity. This is the normal similarity weight term.

6. The method according to claim 5, characterized in that: In S6, implicit surface interpolation of radial basis functions includes: assuming that the sampling point at the hole boundary is the center point of the radial basis function. c j The corresponding normal direction is n i Construct implicit functions combining radial basis functions and linear terms. f ( p ): ； In the formula, M For the number of RBF center points, For radial basis kernel functions, w j As weight, c j The center point of the radial basis functions, The coefficients of the linear term, b For constant terms, p is the position vector of the three-dimensional point to be evaluated.

7. The method according to claim 6, characterized in that: In S6, texture mapping based on the slope of the triangular facet includes: when the slope angle of the triangular facet is greater than a preset slope threshold, texture mapping is performed using a local tangent plane UV parameterization method. The UV calculation formula for texture mapping is: ； In the formula, u、v For two-dimensional texture coordinates, u a、 v a Anchor coordinates for the texture corresponding to the centroid. s The scale factor from physical length to texture coordinates. t 1, t 2 is a local tangential basis vector. Let be any point on the surface. c The center of mass of the sheet.

Images