Wharf shoreline terrain reconstruction method based on remote sensing data

Abstract

The invention discloses a wharf shoreline terrain reconstruction method based on remote sensing data, and belongs to the technical field of three-dimensional terrain reconstruction. The method comprises the following steps: acquiring a multi-temporal multispectral remote sensing image, a synthetic aperture radar image, a laser radar point cloud and synchronous tidal water level data of a target wharf area; processing the multispectral image to generate an initial land and water segmentation map; carrying out tide correction by utilizing the tide water level data, fusing the multi-polarization characteristics of the synthetic aperture radar image, and correcting the ground feature type to obtain an accurate land and water boundary diagram; processing the laser radar point cloud to generate a digital surface model, extracting a shoreline contour with sub-pixel-level precision from the digital surface model, and endowing the shoreline contour with an elevation value; and superposing the shoreline contour with the elevation with the multispectral image to generate a three-dimensional terrain model. According to the method, the problems of low terrain reconstruction precision, large tide influence and insufficient automation degree of a traditional method in a complex wharf environment are solved, and high-precision and high-efficiency wharf shoreline three-dimensional automatic reconstruction is realized.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional terrain reconstruction technology, and in particular to a method for reconstructing wharf shoreline terrain based on remote sensing data. Background Technology

[0002] As vital shipping hubs, ports and wharves require precise knowledge of their shoreline topography for planning, design, construction, safe navigation, and shoreline resource management. Traditional underwater topographic surveying and shoreline mapping rely primarily on manual on-site investigation or shipborne sonar measurements. These methods are not only labor-intensive, time-consuming, and costly, but also susceptible to environmental factors such as weather and sea conditions, making frequent and large-scale operations difficult. Especially in areas with complex topography and limited accessibility for vessels, data acquisition suffers from blind spots, hindering the provision of comprehensive and continuous topographic information.

[0003] With the development of remote sensing technology, non-contact topographic mapping using technologies such as optical satellite imagery, synthetic aperture radar (SAR), and lidar (LiDAR) has become an important technical means. Among these, optical imagery, especially multispectral imagery, provides rich spectral information and is often used for land-water separation and preliminary shoreline contour extraction using methods such as water body indices. However, this method faces several significant problems in practical applications. First, the shoreline location is dynamically affected by tidal fluctuations; the same location may exhibit different characteristics of water or land at different tidal levels. A single-phase image cannot determine the true and stable shoreline location, leading to inaccuracies in the extracted boundaries. Second, the terrain around wharves is complex, including not only natural mudflats and reefs but also numerous man-made structures such as breakwaters, jetties, and wharf platforms. These man-made structures are easily confused with water bodies in water body index calculations, and their shadows can cause misclassification, significantly reducing the accuracy of automatically extracted land-water boundaries in these areas. Furthermore, optical imagery itself is susceptible to weather conditions such as clouds, rain, and fog, leading to data loss or quality degradation.

[0004] Synthetic Aperture Radar (SAR) imagery, with its all-weather, day-and-night operation, can acquire data even in adverse weather conditions, offering a partial solution to the aforementioned problems. Its backscattering characteristics are sensitive to surface roughness and geometry, theoretically allowing for the differentiation of different feature types. However, effectively fusing SAR features with optical imagery for accurate identification and differentiation of specific features such as reefs and artificial structures in dock areas remains a challenge. Existing methods often have simplistic processing workflows and fail to fully leverage the advantages of multi-source data.

[0005] On the other hand, LiDAR technology can actively acquire high-precision 3D point cloud data, making it an important data source for generating high-precision digital elevation models. However, the automation and integration of existing technologies still need to be improved in order to effectively combine LiDAR elevation information with precise 2D shoreline vectors extracted from imagery and automatically generate 3D models that have both accurate geometric shapes and realistic texture features.

[0006] Therefore, there is an urgent need in this field for a method that can comprehensively utilize the advantages of multi-source remote sensing data, effectively overcome the influence of tides, automatically and accurately distinguish complex land features, and achieve high-precision and high-efficiency three-dimensional terrain reconstruction of wharf shorelines. Summary of the Invention

[0007] This invention overcomes the problems of low terrain reconstruction accuracy, large susceptibility to tides, and insufficient automation in existing technologies in complex dock environments, and achieves high-precision and high-efficiency three-dimensional automated reconstruction of dock shorelines.

[0008] To achieve the above objectives, the present invention adopts the following solution: The method for reconstructing wharf shoreline topography based on remote sensing data includes the following steps: S1: Acquire multi-temporal multispectral remote sensing images, synthetic aperture radar images, lidar point cloud data, and corresponding tidal water level data of the target wharf area. S2: Perform atmospheric and geometric corrections on multispectral remote sensing images, calculate the improved normalized water index, and generate an initial land-water separation map. S3: Based on tidal water level data, the initial land-water separation map is tidal corrected, and the VV polarization and VH polarization features of synthetic aperture radar imagery are used to identify land features in the corrected land-water separation map. By calculating the difference in backscattering intensity and texture features between the synthetic aperture radar imagery in the reef area and the artificial structure area of ​​the wharf, a distinction model is established. Based on the distinction model, the land feature type is corrected in the land-water separation map to generate an accurate land-water boundary map. S4: Process lidar point cloud data to generate a digital surface model. Use the Canny edge detection operator to extract the wharf shoreline contour with sub-pixel accuracy from the accurate land-water boundary map, and combine it with the digital surface model to assign elevation values ​​to the shoreline. S5: Spatially overlay the wharf shoreline outline with elevation values ​​onto multispectral remote sensing images, and generate a three-dimensional terrain model of the wharf shoreline using a triangular mesh construction algorithm.

[0009] Preferably, step S1 specifically includes the following steps: The system acquires Sentinel-2 multispectral imagery and Sentinel-1 synthetic aperture radar imagery of the target wharf area using a satellite remote sensing platform. The multispectral remote sensing images are acquired over at least one tidal cycle, including image data from both high and low tide phases. Lidar point cloud data of the target wharf area is acquired using an airborne lidar measurement system, with a point cloud density of at least 16 points per square meter. Tidal level data is obtained from tidal observation records from marine monitoring stations near the target wharf area, including instantaneous tide height values ​​precisely corresponding to the acquisition time of each multispectral remote sensing and synthetic aperture radar image. The acquisition time of the lidar point cloud data is matched with the transit time of the multispectral remote sensing images to calculate the time difference. When the time difference exceeds a set threshold, the lidar point cloud data is corrected for tidal difference. All remote sensing data are uniformly converted to the same coordinate system, and the spatial resolution is resampled to a consistent scale. The data is also cropped according to the boundary of the target wharf area to form a standardized input dataset.

[0010] Preferably, step S5 specifically includes the following steps: A topological check was performed on the wharf shoreline contour data with elevation values ​​to examine the continuity and closure of the contour lines. A constrained Delaunay triangulation algorithm was used to construct a triangular mesh, constrained by the wharf shoreline contour. During the triangular mesh generation process, the maximum area of ​​the triangles was limited to 50 square meters, and the minimum angle was limited to 25 degrees. True-color textures were generated based on the red, green, and blue bands of multispectral remote sensing imagery, and the textures were mapped onto the surface of the triangular mesh through image registration. For buildings and artificial structures in the wharf area, a feature-preserving mesh optimization algorithm was used to adaptively subdivide and smooth the triangular mesh. A multi-level detail model was established, and triangular mesh models of different resolutions were generated through a mesh simplification algorithm. A 3D terrain model including complete geometric and texture information was output, along with a model quality report recording the number of meshes, texture resolution, and model accuracy.

[0011] Preferably, step S2 specifically includes the following steps: Atmospheric correction was performed on the multispectral remote sensing image using dark target subtraction. Geometric fine correction was then performed on the multispectral remote sensing image based on ground control points until the positioning error was less than one pixel. Based on the corrected multispectral remote sensing image, the improved Normalized Difference Water Index (MNDWI) was calculated using the green band and shortwave infrared band. The calculation formula is MNDWI=(Green-SWIR) / (Green+SWIR), where Green is the reflectance of the green band and SWIR is the reflectance of the shortwave infrared band. An adaptive thresholding method was used to segment the distribution image of the improved MNDWI. The segmentation threshold was determined by analyzing the histogram features of the improved MNDWI. Morphological processing was performed on the segmented binary image. Opening operations were used to eliminate small-area noise, and closing operations were used to fill holes to generate an initial land-water segmentation map. The threshold parameters and morphological processing parameters used in the segmentation process were recorded.

[0012] As a preferred option, after generating the initial land-water separation map, quality verification and optimization are performed, including the following steps: The initial land-water segmentation map was validated using synthetic aperture radar (SAR) imagery acquired concurrently. By comparing the consistency between the improved normalized water index segmentation results and the backscattering characteristics of the SAR imagery, potentially misclassified areas were identified. For the unique breakwater structures and breakwater shadow areas of the wharf area, a misclassified area detection model based on multi-feature fusion was established. This model comprehensively utilizes the improved normalized water index value, texture features, and spatial context information. The detected misclassified areas were manually corrected, and the corrected land-water segmentation map was used as input for subsequent tidal correction. Simultaneously, a quality evaluation report was established, recording the number, area ratio, and spatial distribution characteristics of misclassified areas.

[0013] Preferably, step S3 specifically includes: A tide level-shoreline positional relationship model was established based on tidal level data. The instantaneous tide level influence value of each pixel in the tide level-shoreline positional relationship model was calculated using linear interpolation. Each pixel in the initial land-water segmentation map was reclassified according to its corresponding elevation value and instantaneous tide level value to generate a tidal-corrected land-water segmentation map. The synthetic aperture radar image was preprocessed, including radiometric calibration, noise filtering, and terrain correction, and the VV polarization and VH polarization backscattering coefficient features were extracted. The texture feature parameters of the synthetic aperture radar image were calculated, including contrast, correlation, and entropy features based on the gray-level co-occurrence matrix. A backscattering intensity difference model between the reef area and the wharf artificial structure area was established. By analyzing the statistical difference in the ratio of VV polarization and VH polarization backscattering coefficients between the reef area and the wharf artificial structure area, the distinction threshold range was determined.

[0014] As a preferred approach, the method for establishing the differentiation model is as follows: A land cover classifier was constructed using machine learning classification algorithms. The input features of the land cover classifier included VV polarization backscattering coefficient, VH polarization backscattering coefficient, VV / VH polarization ratio, and texture feature parameters. Training samples of known types were collected, including typical reef area samples and dock artificial structure area samples, to train the classifier. During training, cross-validation was used to optimize the classifier parameters, and the classification accuracy of the classifier was trained to over 85%. The trained classifier was used to predict the land cover type for each pixel of the entire land-water boundary map. For uncertain areas in the land cover type prediction results, spatial context analysis was used to further distinguish them by combining the distribution characteristics of land cover types around the pixel. Finally, an accurate land-water boundary map with land cover type labels was generated, which distinguishes between reef areas and dock artificial structure areas.

[0015] As a preferred method, the land cover type correction method based on the differentiation model for land-water separation maps includes: The predicted land cover types output by the classifier are post-processed, and morphological filtering is used to eliminate isolated misclassified pixels. A spatial distribution rule library for land cover types is established, including the natural distribution characteristics of reef areas and the regular geometric shape features of artificial structures such as wharves. Areas that violate the spatial distribution rules are manually and interactively corrected, and the original multispectral remote sensing images and synthetic aperture radar images are compared and verified during the correction process. The corrected land cover type data is fused with the tidal-corrected land-water boundary map to output an accurate land-water boundary map with complete topological information. Each boundary line segment in the accurate land-water boundary map is labeled with a land cover type attribute. At the same time, a land cover classification accuracy report is generated, recording the classification accuracy indicators and correction statistics for various land cover types.

[0016] Preferably, step S4 specifically includes: Gaussian filtering was applied to the accurate land-water boundary map, and convolution was performed using a Gaussian kernel with a standard deviation of 0.5. The image gradient magnitude and direction of the accurate land-water boundary map were calculated, and convolution was performed in the horizontal and vertical directions using a 3×3 Sobel operator. The gradient magnitude was refined using non-maximum suppression, retaining local maxima. Dual threshold detection and edge connection were used, with the high threshold set to 0.7 times the maximum gradient magnitude and the low threshold set to 0.4 times the high threshold. Broken edge segments were connected using an edge tracking algorithm. The extracted edges were refined to sub-pixel level using B-spline curve fitting, achieving an edge localization accuracy of 0.1 pixels. The refined edges were precisely registered with the digital surface model data, and the nearest neighbor method was used to assign a corresponding elevation value to each edge point. The elevation assignment results were quality controlled, and outlier elevation values ​​exceeding the normal range were removed.

[0017] As a preferred option, after assigning elevation values ​​to the shoreline, the elevation data is optimized, including the following steps: A height anomaly detection model was established to identify potential height anomalies by analyzing the height change rate and spatial distribution characteristics of adjacent points. A moving window filtering method was used to smooth the height data, with a window size of 5×5 pixels, and a median filtering algorithm was used to eliminate noise in the height data. A topological consistency check was performed on the height data of the wharf shoreline contour to verify whether the height changes of adjacent points met the requirements of terrain continuity. Special processing rules were established for the unique vertical shoreline structure of the wharf area, and a linear interpolation method was used to supplement missing height data caused by occlusion. Finally, complete wharf shoreline contour data with height attributes was generated, and a height data quality report was output, recording the number of height anomalies and their processing status.

[0018] The present invention has at least the following beneficial effects: (1) By integrating multispectral remote sensing images, synthetic aperture radar images, lidar point clouds and tidal level data, it comprehensively utilizes the advantages of various types of data, effectively overcomes the limitations of a single data source, and significantly improves the overall accuracy and reliability of the results from two-dimensional boundary extraction to three-dimensional model construction; (2) By introducing synchronized tidal level data and establishing a tidal level-shoreline position relationship model, the initial land-water segmentation results are dynamically corrected, so that the finally extracted shoreline can reflect the real topographic boundary, rather than the apparent boundary under the instantaneous water level; (3) By utilizing the multipolarization features and texture features of synthetic aperture radar images, combined with machine learning, the invention achieves significant improvements in the overall accuracy and reliability of the results from two-dimensional boundary extraction to three-dimensional model construction. The learning algorithm constructs a land feature classifier, which can effectively distinguish land features with similar spectral features but different materials and structures, reducing misclassification and improving the accuracy of land feature attributes in the land-water boundary map; (4) Using edge detection and optimization algorithms, a fine shoreline contour is obtained, and by registering it with a digital surface model derived from a high-precision lidar point cloud, an accurate elevation value is automatically assigned to the shoreline, improving the geometric accuracy of the three-dimensional reconstruction; (5) Through constrained triangulation, texture mapping and feature-preserving mesh optimization algorithms, a three-dimensional model with realistic texture, good structural details and correct topological relationships is automatically generated, reducing manual intervention and improving modeling efficiency. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the principle of one method of the present invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0021] like Figure 1 As shown, the method for reconstructing wharf shoreline topography based on remote sensing data provided by the present invention includes the following steps: S1: Acquire multi-temporal multispectral remote sensing images, synthetic aperture radar images, lidar point cloud data, and corresponding tidal water level data of the target wharf area. S2: Perform atmospheric and geometric corrections on multispectral remote sensing images, calculate the improved normalized water index, and generate an initial land-water separation map. S3: Based on tidal water level data, the initial land-water separation map is tidal corrected, and the VV polarization and VH polarization features of synthetic aperture radar imagery are used to identify land features in the corrected land-water separation map. By calculating the difference in backscattering intensity and texture features between the synthetic aperture radar imagery in the reef area and the artificial structure area of ​​the wharf, a distinction model is established. Based on the distinction model, the land feature type is corrected in the land-water separation map to generate an accurate land-water boundary map. S4: Process lidar point cloud data to generate a digital surface model. Use the Canny edge detection operator to extract the wharf shoreline contour with sub-pixel accuracy from the accurate land-water boundary map, and combine it with the digital surface model to assign elevation values ​​to the shoreline. S5: Spatially overlay the wharf shoreline outline with elevation values ​​onto multispectral remote sensing images, and generate a three-dimensional terrain model of the wharf shoreline using a triangular mesh construction algorithm.

[0022] When reconstructing the wharf shoreline topography, multi-source remote sensing data and tidal level information are comprehensively utilized. Through a series of image processing, feature extraction, and model building steps, a refined 3D reconstruction of the wharf shoreline topography is achieved. First, multi-temporal multispectral remote sensing images, synthetic aperture radar (SAR) images, lidar point cloud data, and corresponding temporal tidal level data of the target wharf area are acquired. Multispectral remote sensing images provide rich spectral information, which helps distinguish the land-water boundary; SAR images have all-weather imaging capabilities, especially advantageous in identifying ground feature structures and textures; lidar point cloud data provides high-precision elevation information; and tidal level data is used for subsequent tidal correction to eliminate shoreline position deviations caused by tidal changes. These data must maintain consistency in time and space, typically achieved through time matching and coordinate system alignment to ensure the accuracy of subsequent processing.

[0023] Atmospheric and geometric corrections were performed on the multispectral remote sensing imagery to eliminate errors caused by atmospheric scattering and sensor attitude, thereby improving image quality. Subsequently, an improved Normalized Difference Water Index (MNDWI) was calculated. This index, derived from the reflectance of the green and shortwave infrared bands, effectively enhances the contrast between water and land, thus generating an initial water-land segmentation map. Image segmentation techniques were then used to divide the imagery into water and land portions for subsequent tidal correction and feature identification.

[0024] Tidal correction was performed on the initial land-water boundary map using tidal level data. By establishing a model relating tidal level to shoreline location, each pixel was reclassified to reflect the true land cover. Simultaneously, by combining VV and VH polarization features from synthetic aperture radar imagery, backscattering intensity and texture features were extracted to build a distinguishing model to identify reef areas and artificial structure areas like wharves. This model can be constructed using machine learning methods, taking inputs including polarization features and texture parameters. Through training on known samples, it achieves automatic identification and correction of land cover types, ultimately generating an accurate land-water boundary map containing land cover type attribute information.

[0025] The process involves processing lidar point cloud data to generate a digital surface model (DSM), which provides elevation information of the land surface. Subpixel-level precision wharf shoreline contours are then extracted from a precise land-water boundary map using the Canny edge detection algorithm. This algorithm includes sub-steps such as gradient calculation, non-maximum suppression, and double-threshold detection to ensure edge continuity and accuracy. The extracted edges are then registered with the digital surface model, and an elevation value is assigned to each shoreline point based on spatial correspondence, resulting in shoreline contour data with elevation attributes.

[0026] The wharf shoreline outline with elevation values ​​is spatially overlaid with multispectral remote sensing imagery, and a 3D terrain model is generated using a triangulation algorithm (such as constrained Delaunay triangulation). During this process, maximum area and minimum angle limits for the triangular mesh can be set to ensure the geometric quality of the model. Simultaneously, true-color textures are generated from the multispectral imagery and mapped onto the surface of the triangular mesh to enhance the model's realism. For buildings and artificial structures present in the wharf area, feature-preserving mesh optimization methods can be used to process them, ultimately outputting a 3D terrain model with both geometric and textural information.

[0027] This method achieves automated and high-precision reconstruction of wharf shoreline topography by organically combining multi-source data fusion, tidal correction, machine learning-assisted feature recognition, high-precision edge extraction, and 3D modeling. Compared with traditional single remote sensing methods or field measurement methods, this method has higher efficiency and accuracy, effectively addressing the needs of shoreline change monitoring and terrain modeling in complex environments, and is suitable for practical applications such as port planning and shoreline management.

[0028] In another technical solution, step S1 specifically includes the following steps: The system acquires Sentinel-2 multispectral imagery and Sentinel-1 synthetic aperture radar imagery of the target wharf area using a satellite remote sensing platform. The multispectral remote sensing images are acquired over at least one tidal cycle, including image data from both high and low tide phases. Lidar point cloud data of the target wharf area is acquired using an airborne lidar measurement system, with a point cloud density of at least 16 points per square meter. Tidal level data is obtained from tidal observation records from marine monitoring stations near the target wharf area, including instantaneous tide height values ​​precisely corresponding to the acquisition time of each multispectral remote sensing and synthetic aperture radar image. The acquisition time of the lidar point cloud data is matched with the transit time of the multispectral remote sensing images to calculate the time difference. When the time difference exceeds a set threshold, the lidar point cloud data is corrected for tidal difference. All remote sensing data are uniformly converted to the same coordinate system, and the spatial resolution is resampled to a consistent scale. The data is also cropped according to the boundary of the target wharf area to form a standardized input dataset.

[0029] The data acquisition and preprocessing stages ensure the temporal, spatial, and accuracy consistency of multi-source remote sensing data, providing high-quality input data. Sentinel-2 multispectral imagery and Sentinel-1 synthetic aperture radar imagery are acquired through a satellite remote sensing platform. The acquisition time of the multispectral imagery must cover at least one complete tidal cycle; for example, images from high and low tide times within a consecutive month may be selected to ensure the significant impact of tidal changes on shoreline positions is captured. Synthetic aperture radar imagery utilizes its microwave penetration capability to compensate for the limitations of optical imagery under cloudy and rainy weather conditions. LiDAR point cloud data is collected through an airborne measurement system, and the point cloud density must meet certain requirements, such as no less than 16 points per square meter, to ensure sufficient topographic detail when generating digital surface models. Tidal level data is derived from actual records from marine monitoring stations and is precisely matched with the acquisition time of each image. Time synchronization accuracy is typically required to be controlled within a few minutes to accurately reflect the water level conditions at the moment of imaging.

[0030] During the data preprocessing stage, the time difference between the lidar point cloud and the optical image is checked and corrected. Since tide levels change constantly, if the difference between the point cloud acquisition time and the image transit time is significant, such as exceeding 1 to 2 hours, a tidal level difference correction needs to be applied to the point cloud elevation based on the tidal level change curve to eliminate elevation deviations caused by time asynchrony. All data also needs to be uniformly transformed to the same coordinate system, such as the UTM projected coordinate system, and spatial resolution is kept consistent through resampling. Commonly used resolutions include 0.5 meters, 1 meter, or 2 meters, depending on the application requirements. Finally, the dataset is cropped according to the actual area of ​​the target wharf, removing irrelevant regions to form standardized input data.

[0031] By acquiring high-precision data from multiple sources and undergoing rigorous preprocessing, this method effectively overcomes the limitations of a single data source, significantly improves the spatiotemporal consistency of data, and provides reliable data support for shoreline topography reconstruction. Compared with traditional methods, this process has stronger systematicity and adaptability, and can maintain stable processing results over a large time span and under complex environmental conditions.

[0032] In another technical solution, step S5 specifically includes the following steps: A topological check was performed on the wharf shoreline contour data with elevation values ​​to examine the continuity and closure of the contour lines. A constrained Delaunay triangulation algorithm was used to construct a triangular mesh, constrained by the wharf shoreline contour. During the triangular mesh generation process, the maximum area of ​​the triangles was limited to 50 square meters, and the minimum angle was limited to 25 degrees. True-color textures were generated based on the red, green, and blue bands of multispectral remote sensing imagery, and the textures were mapped onto the surface of the triangular mesh through image registration. For buildings and artificial structures in the wharf area, a feature-preserving mesh optimization algorithm was used to adaptively subdivide and smooth the triangular mesh. A multi-level detail model was established, and triangular mesh models of different resolutions were generated through a mesh simplification algorithm. A 3D terrain model including complete geometric and texture information was output, along with a model quality report recording the number of meshes, texture resolution, and model accuracy.

[0033] The construction and optimization process of the 3D terrain model ensures the quality of the model in terms of geometric accuracy, topological structure, and visual representation. This stage first performs a topological check on the wharf shoreline contour data with elevation attributes to ensure that the contour lines are continuous, closed, and free of self-intersections or breaks. Then, using the shoreline contour as a constraint, a constrained Delaunay triangulation algorithm is used to construct a triangular mesh. This algorithm can optimize the triangle shape while maintaining boundary integrity. During the process, a maximum triangle area limit can be set, for example, not exceeding 50 square meters, to avoid excessively large triangular faces; simultaneously, a minimum angle limit can be set, for example, not less than 25 degrees, to avoid sharp triangles, thereby improving the numerical stability and visualization effect of the mesh.

[0034] Based on mesh generation, true-color textures are synthesized using the red, green, and blue bands of multispectral remote sensing imagery. High-precision registration maps these textures onto the triangular mesh surface, giving the model a realistic appearance. For common artificial structures in the dock area, such as buildings and revetments, feature-preserving mesh optimization algorithms are employed. These include adaptive subdivision of structure edges and smoothing of flat areas, thus improving overall mesh quality while retaining detailed features. To further enhance model applicability, multi-level detail models can be constructed, generating versions at different resolutions using mesh simplification algorithms to meet diverse application needs, from macro-level planning to fine-grained analysis.

[0035] The final output 3D terrain model contains complete geometric and texture information, and includes a model quality report recording information such as the number of meshes, texture resolution, and accuracy metrics. This modeling process, by integrating methods such as constraint meshing, texture mapping, and adaptive optimization, significantly improves the realism and usability of the model. Compared to traditional single modeling methods, it can more comprehensively and accurately reflect the 3D terrain features of the wharf shoreline.

[0036] In another technical solution, step S2 specifically includes the following steps: Atmospheric correction was performed on the multispectral remote sensing image using dark target subtraction. Geometric fine correction was then performed on the multispectral remote sensing image based on ground control points until the positioning error was less than one pixel. Based on the corrected multispectral remote sensing image, the improved Normalized Difference Water Index (MNDWI) was calculated using the green band and shortwave infrared band. The calculation formula is MNDWI=(Green-SWIR) / (Green+SWIR), where Green is the reflectance of the green band and SWIR is the reflectance of the shortwave infrared band. An adaptive thresholding method was used to segment the distribution image of the improved MNDWI. The segmentation threshold was determined by analyzing the histogram features of the improved MNDWI. Morphological processing was performed on the segmented binary image. Opening operations were used to eliminate small-area noise, and closing operations were used to fill holes to generate an initial land-water segmentation map. The threshold parameters and morphological processing parameters used in the segmentation process were recorded.

[0037] The generation of the land-water separation map involves a series of image preprocessing and index calculation steps to effectively distinguish between water bodies and land. Atmospheric correction is performed on the multispectral remote sensing image using methods such as dark target subtraction to eliminate the influence of atmospheric scattering and absorption on the reflectance of ground features, thereby restoring the true spectral characteristics of the features. Subsequently, geometric fine correction is performed, using ground control points to control the image positioning error to a low level, such as no more than one pixel, ensuring high spatial accuracy of the image.

[0038] Based on image correction, an improved Normalized Difference Water Index (MNDWI) is calculated. This index, constructed by combining reflectance in the green and shortwave infrared bands, effectively enhances the spectral difference between water and land, making it particularly suitable for complex environments like docks. After index calculation, an adaptive thresholding method is used to segment the MNDWI image. By analyzing its histogram features, such as troughs or bimodal characteristics, a suitable segmentation threshold is automatically determined, typically ranging from 0 to 0.4, with the specific value adjusted based on the actual image characteristics. The resulting binary image undergoes further morphological processing, such as opening operations to eliminate small noise points (using 3×3 or 5×5 structuring elements) and closing operations to fill internal holes, thereby generating a continuous and complete initial water-land segmentation map. The threshold and processing parameters used are recorded for subsequent quality assessment.

[0039] This method significantly improves the automation and accuracy of land-water boundary extraction, reducing missegmentation caused by differences in lighting, atmospheric conditions, or sensors. Compared to traditional single-threshold segmentation methods, this process is more adaptable and robust, maintaining good segmentation results under various environmental conditions.

[0040] After generating the initial land-water separation map, quality verification and optimization are performed, including the following steps: The initial land-water segmentation map was validated using synthetic aperture radar (SAR) imagery acquired concurrently. By comparing the consistency between the improved normalized water index segmentation results and the backscattering characteristics of the SAR imagery, potentially misclassified areas were identified. For the unique breakwater structures and breakwater shadow areas of the wharf area, a misclassified area detection model based on multi-feature fusion was established. This model comprehensively utilizes the improved normalized water index value, texture features, and spatial context information. The detected misclassified areas were manually corrected, and the corrected land-water segmentation map was used as input for subsequent tidal correction. Simultaneously, a quality evaluation report was established, recording the number, area ratio, and spatial distribution characteristics of misclassified areas.

[0041] The initial land-water segmentation map was validated and optimized. Through multi-source data comparison and feature fusion, misclassified regions were identified and corrected, further improving the reliability of the segmentation results. The initial segmentation results were validated using concurrently acquired synthetic aperture radar (SAP) imagery. By comparing the backscattering characteristics of the MNDWI segmentation results with those of the SAR imagery (e.g., water bodies typically exhibit low backscattering while land exhibits medium to high backscattering), potentially inconsistent regions were identified, such as misclassifications caused by shadows or specular reflections.

[0042] For the unique structures such as breakwaters and seawalls in the wharf area, and their shadowed areas, a misclassification detection model based on multi-feature fusion is constructed. This model comprehensively utilizes MNDWI values, SAR texture features (such as contrast and entropy based on the gray-level co-occurrence matrix), and spatial context information (such as shape, size, and proximity relationships). By setting appropriate feature weights and decision rules, such as logistic regression or simple thresholding rules, it identifies anomalous regions in the initial segmentation. The detected misclassified regions are then manually corrected. Operators can interactively edit the original multispectral and SAR images, and the corrected results are used as input for subsequent tidal correction.

[0043] Upon completion of the entire process, a quality evaluation report is generated, recording the number, area ratio, and distribution characteristics of misclassified areas, providing a basis for process optimization. This verification and optimization mechanism significantly improves the accuracy of land-water segmentation maps in complex dock environments, reduces typical errors in automated processing, and compared with traditional segmentation methods that rely solely on optical imagery, it can better handle special features and shadow interference, ensuring the reliability of subsequent terrain reconstruction processes.

[0044] In another technical solution, step S3 specifically includes: A tide level-shoreline positional relationship model was established based on tidal level data. The instantaneous tide level influence value of each pixel in the tide level-shoreline positional relationship model was calculated using linear interpolation. Each pixel in the initial land-water segmentation map was reclassified according to its corresponding elevation value and instantaneous tide level value to generate a tidal-corrected land-water segmentation map. The synthetic aperture radar image was preprocessed, including radiometric calibration, noise filtering, and terrain correction, and the VV polarization and VH polarization backscattering coefficient features were extracted. The texture feature parameters of the synthetic aperture radar image were calculated, including contrast, correlation, and entropy features based on the gray-level co-occurrence matrix. A backscattering intensity difference model between the reef area and the wharf artificial structure area was established. By analyzing the statistical difference in the ratio of VV polarization and VH polarization backscattering coefficients between the reef area and the wharf artificial structure area, the distinction threshold range was determined.

[0045] Tidal correction and synthetic aperture radar (SAR) feature extraction are performed to improve the accuracy of land-water boundary and feature identification by fusing multidimensional features from tidal levels and SAR imagery. A tidal level-shoreline positional relationship model is constructed based on tidal level data. This model calculates the instantaneous tidal level impact value for each pixel location using linear interpolation, thus reflecting the dynamic influence of tidal rise and fall on shoreline position. The essence of tidal correction lies in reclassifying the initial land-water segmentation results based on the pixel's corresponding elevation and instantaneous tidal level value. For example, during high tide, some low-lying land may be submerged and misclassified as water; correction can restore its true category, generating a more accurate land-water distribution map.

[0046] Simultaneously with tidal correction, SAR imagery undergoes systematic preprocessing to extract effective features. Preprocessing includes radiometric calibration, converting raw digital quantization values ​​into backscattering coefficients to ensure physical comparability of the data; noise filtering, such as using Lee filtering or Gamma MAP filtering to suppress speckle noise, with window sizes potentially ranging from 3×3 to 7×7; and terrain correction, eliminating radiometric distortion caused by terrain in areas with significant undulations. Subsequently, backscattering coefficients from the VV and VH polarization channels are extracted, and texture feature parameters, such as contrast, correlation, and entropy based on the gray-level co-occurrence matrix, are further calculated. Window sizes can be selected as 5×5 or 7×7 to capture the spatial structure information of ground features.

[0047] A distinguishing model was established by analyzing the statistical differences in backscattering coefficient ratios (e.g., VV / VH) and texture features between reef and artificial structure areas (such as wharf areas). This model sets threshold ranges, for example, where the VV / VH ratio may be significantly higher or lower in the reef area than in the artificial structure area, thus providing a basis for subsequent land cover classification. This series of processing steps effectively integrates tidal dynamics and SAR response characteristics, significantly enhancing the ability to distinguish different land cover types in complex coastal environments. Compared with traditional single optical image analysis methods, it exhibits better stability and accuracy.

[0048] The method for establishing the differentiation model is as follows: A land cover classifier was constructed using machine learning classification algorithms. The input features of the land cover classifier included VV polarization backscattering coefficient, VH polarization backscattering coefficient, VV / VH polarization ratio, and texture feature parameters. Training samples of known types were collected, including typical reef area samples and dock artificial structure area samples, to train the classifier. During training, cross-validation was used to optimize the classifier parameters, and the classification accuracy of the classifier was trained to over 85%. The trained classifier was used to predict the land cover type for each pixel of the entire land-water boundary map. For uncertain areas in the land cover type prediction results, spatial context analysis was used to further distinguish them by combining the distribution characteristics of land cover types around the pixel. Finally, an accurate land-water boundary map with land cover type labels was generated, which distinguishes between reef areas and dock artificial structure areas.

[0049] In the process of building and applying a machine learning-based land cover classifier, multi-feature combination and statistical learning algorithms are used to achieve automated and high-precision identification of artificial structures such as reefs and wharves. The input feature set of the land cover classifier is carefully designed, including not only the VV and VH polarization backscattering coefficients of SAR imagery, but also the ratio feature of the two (VV / VH). This ratio is sensitive to differences in surface roughness and dielectric constant, and can effectively enhance the separation between categories. At the same time, texture feature parameters based on the gray-level co-occurrence matrix, such as contrast, correlation, and entropy, are fused to comprehensively describe the attributes of land covers from both spectral response and spatial structure dimensions.

[0050] The classifier is built using training samples of known types, covering typical reef areas and man-made structures like docks, which may be obtained through field surveys or high-resolution image interpretation. Machine learning algorithms such as Support Vector Machines (SVM), Random Forests, or neural networks are used for training. During training, cross-validation is employed to optimize model parameters, such as the kernel function and penalty coefficient for SVMs, and the number and depth of trees in random forests, aiming to improve the overall accuracy of the classifier to a high level, for example, above 85%. The trained classifier then predicts the type of each pixel in the entire land-water segmentation map, outputting preliminary land cover type labels.

[0051] For uncertain areas or misclassified pixels in the prediction results, spatial context analysis is used for further discrimination. Based on the principle of spatial continuity of land cover distribution, this method analyzes the type distribution characteristics within the surrounding neighborhood (e.g., 8-neighborhood or larger windows) of the target pixel, correcting prediction results that are isolated or conflict with surrounding types. The final result is an accurate land-water boundary map with land cover type labels, clearly distinguishing reefs from artificial structures. This method, combining machine learning and spatial analysis, significantly improves the automation level and interpretation accuracy of land cover classification, reduces the intensity of human intervention, and is suitable for detailed mapping of large-scale coastal areas.

[0052] Methods for correcting land cover types in land-water separation maps based on differentiation models include: The predicted land cover types output by the classifier are post-processed, and morphological filtering is used to eliminate isolated misclassified pixels. A spatial distribution rule library for land cover types is established, including the natural distribution characteristics of reef areas and the regular geometric shape features of artificial structures such as wharves. Areas that violate the spatial distribution rules are manually and interactively corrected, and the original multispectral remote sensing images and synthetic aperture radar images are compared and verified during the correction process. The corrected land cover type data is fused with the tidal-corrected land-water boundary map to output an accurate land-water boundary map with complete topological information. Each boundary line segment in the accurate land-water boundary map is labeled with a land cover type attribute. At the same time, a land cover classification accuracy report is generated, recording the classification accuracy indicators and correction statistics for various land cover types.

[0053] In the post-processing and fusion output stage of the classification results, a series of rule bases and human interaction measures are used to further improve the topological quality and practical value of the land cover classification map. This process first performs post-processing on the land cover type prediction results output by the machine learning classifier, using morphological filtering methods, such as using 3×3 or 5×5 structuring elements for opening or closing operations, to eliminate isolated misclassified pixels and small noise patches in the image, making homogeneous land cover regions more continuous and complete.

[0054] Subsequently, a spatial distribution rule base is established based on prior knowledge of land cover types to identify and correct areas that violate general distribution patterns. The rule base includes rules such as the irregular natural contours and clustered distribution characteristics of reef areas, while artificial structures at wharves often exhibit regular geometric shapes (e.g., straight lines, right angles), specific orientations, and spatial relationships with port functions. The system can automatically detect potential anomalies and prompt for interactive manual correction. Operators can access original multispectral remote sensing images and SAR images for comparison and verification, making judgments based on spectral, textural, and spatial features, and editing incorrect classifications to ensure the reliability of the correction results.

[0055] The corrected land cover type data is then fused with the tidal-corrected land-water separation map to generate an accurate land-water boundary map containing complete topological information and land cover type attribute labels. Each boundary segment in the map not only has geometric location information but also land cover type attributes. A land cover classification accuracy report is also output, summarizing classification accuracy indicators for various land cover types (such as producer accuracy and user accuracy) and correction statistics. This post-processing and fusion mechanism greatly ensures the data quality and direct usability of the final results, enabling it to better support subsequent 3D terrain modeling and spatial analysis applications.

[0056] In another technical solution, step S4 specifically includes: Gaussian filtering was applied to the accurate land-water boundary map, and convolution was performed using a Gaussian kernel with a standard deviation of 0.5. The image gradient magnitude and direction of the accurate land-water boundary map were calculated, and convolution was performed in the horizontal and vertical directions using a 3×3 Sobel operator. The gradient magnitude was refined using non-maximum suppression, retaining local maxima. Dual threshold detection and edge connection were used, with the high threshold set to 0.7 times the maximum gradient magnitude and the low threshold set to 0.4 times the high threshold. Broken edge segments were connected using an edge tracking algorithm. The extracted edges were refined to sub-pixel level using B-spline curve fitting, achieving an edge localization accuracy of 0.1 pixels. The refined edges were precisely registered with the digital surface model data, and the nearest neighbor method was used to assign a corresponding elevation value to each edge point. The elevation assignment results were quality controlled, and outlier elevation values ​​exceeding the normal range were removed.

[0057] This process extracts sub-pixel-level shoreline contours from a precise land-water boundary map and assigns elevation values ​​to the contour points, achieving high-precision positioning of the shoreline boundary and accurate correlation with elevation information. Gaussian filtering preprocessing is applied to the precise land-water boundary map using a Gaussian kernel function with a small standard deviation, such as 0.5, for convolution operations to smooth image noise while preserving important edge information. Subsequently, the gradient magnitude and direction of the image are calculated, typically using the Sobel operator for convolution in both the horizontal and vertical directions. The operator size is commonly 3×3, thus obtaining the gradient intensity and direction at each pixel. This gradient information reveals the severity and trend of grayscale changes in the image and is the fundamental basis for edge detection.

[0058] After obtaining gradient information, non-maximum suppression (NMS) is used to refine the gradient magnitude. By comparing the neighboring values ​​of each pixel in its gradient direction, only local maxima are retained, resulting in a finer and more accurate edge contour. Next, a dual-threshold detection method is used for edge connection. A high threshold and a low threshold are set; for example, the high threshold can be 0.7 times the maximum gradient magnitude, and the low threshold can be 0.4 times the high threshold. Points with gradient magnitudes higher than the high threshold are identified as strong edge points, while points between the high and low thresholds are identified as weak edge points. An edge tracking algorithm connects the weak edge points to the strong edge points, forming complete and continuous edge segments. To further improve edge localization accuracy, sub-pixel-level refinement algorithms such as B-spline curve fitting are used to optimize the extracted pixel-level edges, achieving an actual edge localization accuracy within 0.1 pixels, significantly improving the geometric accuracy of the shoreline contour.

[0059] Finally, the refined sub-pixel-level shoreline contour is precisely registered with the digital surface model (DSM) generated from the LiDAR point cloud to ensure complete spatial correspondence. Using interpolation methods such as the nearest neighbor method, each point on the contour is assigned an elevation value corresponding to its location in the DSM, thus achieving the fusion of shoreline geometry and elevation attributes. Strict quality control is applied to the elevation assignment results, such as setting a reasonable elevation range and removing elevation anomalies that significantly deviate from the normal range. The final output is wharf shoreline contour data with high-precision elevation attributes, providing reliable input for the construction of a 3D terrain model.

[0060] In another technical solution, after assigning elevation values ​​to the shoreline, elevation data optimization is performed, including the following steps: A height anomaly detection model was established to identify potential height anomalies by analyzing the height change rate and spatial distribution characteristics of adjacent points. A moving window filtering method was used to smooth the height data, with a window size of 5×5 pixels, and a median filtering algorithm was used to eliminate noise in the height data. A topological consistency check was performed on the height data of the wharf shoreline contour to verify whether the height changes of adjacent points met the requirements of terrain continuity. Special processing rules were established for the unique vertical shoreline structure of the wharf area, and a linear interpolation method was used to supplement missing height data caused by occlusion. Finally, complete wharf shoreline contour data with height attributes was generated, and a height data quality report was output, recording the number of height anomalies and their processing status.

[0061] The shoreline contour data with assigned elevation values ​​is optimized and quality controlled to address noise, anomalies, and inconsistencies, ensuring the accuracy and reliability of the final terrain model. An elevation anomaly detection model is established, identifying potential elevation anomalies by analyzing the elevation change rate of adjacent points, local statistical characteristics (such as mean and standard deviation), and spatial distribution patterns (such as isolated points and clustering). For example, points with sudden elevation changes in flat areas or spatially isolated points on continuous shorelines may be marked as anomalies requiring further processing.

[0062] To address detected noise and anomalies, a moving window filtering method is used to smooth the elevation data. A commonly used window size is 5×5 pixels, and the median filtering algorithm is selected because it effectively removes salt-and-pepper noise while preserving edge features well. After filtering, a topological consistency check is performed on the elevation data of the wharf shoreline outline to ensure that the elevation changes of adjacent points conform to the natural continuity of the terrain. For example, the slope of the elevation changes is checked to ensure that it is within a reasonable physical range, avoiding steep cliff-like changes unless there is indeed a vertical shoreline structure in the area.

[0063] For special structures such as vertical embankments and breakwaters commonly found in wharf areas, their prominent geometric features and susceptibility to data acquisition occlusion necessitate the establishment of special processing rules. Linear interpolation or geometric model-based interpolation methods are employed to reasonably interpolate areas with missing elevation data due to occlusion, restoring complete elevation information. Finally, optimized and complete wharf shoreline contour data with elevation attributes is generated, along with a detailed elevation data quality report. This report records the number of outliers found, the processing methods used, and the post-processing data quality assessment indicators, providing data reliability assurance for subsequent 3D modeling.

[0064] Through the aforementioned systematic optimization process, the quality and consistency of shoreline elevation data can be significantly improved, effectively overcoming the limitations of a single data source and laying a solid data foundation for generating high-precision, high-reliability 3D terrain models of wharf shorelines. Compared with traditional methods, this optimization strategy is more targeted and adaptable, and can better handle elevation data problems in complex environments.

[0065] It should be noted that although the steps are described in a specific order above, this does not mean that they must be performed in that order. In fact, some of these steps can be executed concurrently, or even in a different order, as long as the required functionality is achieved. The number of devices and processing scale described herein are for simplification of the invention; applications, modifications, and variations of this invention will be readily apparent to those skilled in the art.

[0066] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A method for reconstructing wharf shoreline topography based on remote sensing data, characterized in that, Includes the following steps: S1: Acquire multi-temporal multispectral remote sensing images, synthetic aperture radar images, lidar point cloud data, and corresponding tidal water level data of the target wharf area. S2: Perform atmospheric and geometric corrections on multispectral remote sensing images, calculate the improved normalized water index, and generate an initial land-water separation map. S3: Based on tidal water level data, the initial land-water separation map is tidal corrected, and the VV polarization and VH polarization features of synthetic aperture radar imagery are used to identify land features in the corrected land-water separation map. By calculating the difference in backscattering intensity and texture features between the synthetic aperture radar imagery in the reef area and the artificial structure area of ​​the wharf, a distinction model is established. Based on the distinction model, the land feature type is corrected in the land-water separation map to generate an accurate land-water boundary map. S4: Process lidar point cloud data to generate a digital surface model. Use the Canny edge detection operator to extract the wharf shoreline contour with sub-pixel accuracy from the accurate land-water boundary map, and combine it with the digital surface model to assign elevation values ​​to the shoreline. S5: Spatially overlay the wharf shoreline outline with elevation values ​​onto multispectral remote sensing images, and generate a three-dimensional terrain model of the wharf shoreline using a triangular mesh construction algorithm.

2. The method for reconstructing wharf shoreline topography based on remote sensing data according to claim 1, characterized in that, Step S1 specifically includes the following steps: The system acquires Sentinel-2 multispectral imagery and Sentinel-1 synthetic aperture radar imagery of the target wharf area using a satellite remote sensing platform. The multispectral remote sensing images are acquired over at least one tidal cycle, including image data from both high and low tide phases. Lidar point cloud data of the target wharf area is acquired using an airborne lidar measurement system, with a point cloud density of at least 16 points per square meter. Tidal level data is obtained from tidal observation records from marine monitoring stations near the target wharf area, including instantaneous tide height values ​​precisely corresponding to the acquisition time of each multispectral remote sensing and synthetic aperture radar image. The acquisition time of the lidar point cloud data is matched with the transit time of the multispectral remote sensing images to calculate the time difference. When the time difference exceeds a set threshold, the lidar point cloud data is corrected for tidal difference. All remote sensing data are uniformly converted to the same coordinate system, and the spatial resolution is resampled to a consistent scale. The data is also cropped according to the boundary of the target wharf area to form a standardized input dataset.

3. The method for reconstructing wharf shoreline topography based on remote sensing data according to claim 1, characterized in that, Step S5 specifically includes the following steps: A topological check was performed on the wharf shoreline contour data with elevation values ​​to examine the continuity and closure of the contour lines. A constrained Delaunay triangulation algorithm was used to construct a triangular mesh, constrained by the wharf shoreline contour. During the triangular mesh generation process, the maximum area of ​​the triangles was limited to 50 square meters, and the minimum angle was limited to 25 degrees. True-color textures were generated based on the red, green, and blue bands of multispectral remote sensing imagery, and the textures were mapped onto the surface of the triangular mesh through image registration. For buildings and artificial structures in the wharf area, a feature-preserving mesh optimization algorithm was used to adaptively subdivide and smooth the triangular mesh. A multi-level detail model was established, and triangular mesh models of different resolutions were generated through a mesh simplification algorithm. A 3D terrain model including complete geometric and texture information was output, along with a model quality report recording the number of meshes, texture resolution, and model accuracy.

4. The method for reconstructing wharf shoreline topography based on remote sensing data according to claim 1, characterized in that, Step S2 specifically includes the following steps: Atmospheric correction was performed on the multispectral remote sensing image using dark target subtraction. Geometric fine correction was then performed on the multispectral remote sensing image based on ground control points until the positioning error was less than one pixel. Based on the corrected multispectral remote sensing image, the improved Normalized Difference Water Index (MNDWI) was calculated using the green band and shortwave infrared band. The calculation formula is MNDWI=(Green-SWIR) / (Green+SWIR), where Green is the reflectance of the green band and SWIR is the reflectance of the shortwave infrared band. An adaptive thresholding method was used to segment the distribution image of the improved MNDWI. The segmentation threshold was determined by analyzing the histogram features of the improved MNDWI. Morphological processing was performed on the segmented binary image. Opening operations were used to eliminate small-area noise, and closing operations were used to fill holes to generate an initial land-water segmentation map. The threshold parameters and morphological processing parameters used in the segmentation process were recorded.

5. The method for reconstructing wharf shoreline topography based on remote sensing data according to claim 4, characterized in that, After generating the initial land-water separation map, quality verification and optimization are performed, including the following steps: The initial land-water segmentation map was validated using synthetic aperture radar (SAR) imagery acquired concurrently. By comparing the consistency between the improved normalized water index segmentation results and the backscattering characteristics of the SAR imagery, potentially misclassified areas were identified. For the unique breakwater structures and breakwater shadow areas of the wharf area, a misclassified area detection model based on multi-feature fusion was established. This model comprehensively utilizes the improved normalized water index value, texture features, and spatial context information. The detected misclassified areas were manually corrected, and the corrected land-water segmentation map was used as input for subsequent tidal correction. Simultaneously, a quality evaluation report was established, recording the number, area ratio, and spatial distribution characteristics of misclassified areas.

6. The method for reconstructing wharf shoreline topography based on remote sensing data according to claim 1, characterized in that, Step S3 specifically includes: A tide level-shoreline positional relationship model was established based on tidal level data. The instantaneous tide level influence value of each pixel in the tide level-shoreline positional relationship model was calculated using linear interpolation. Each pixel in the initial land-water segmentation map was reclassified according to its corresponding elevation value and instantaneous tide level value to generate a tidal-corrected land-water segmentation map. The synthetic aperture radar image was preprocessed, including radiometric calibration, noise filtering, and terrain correction, and the VV polarization and VH polarization backscattering coefficient features were extracted. The texture feature parameters of the synthetic aperture radar image were calculated, including contrast, correlation, and entropy features based on the gray-level co-occurrence matrix. A backscattering intensity difference model between the reef area and the wharf artificial structure area was established. By analyzing the statistical difference in the ratio of VV polarization and VH polarization backscattering coefficients between the reef area and the wharf artificial structure area, the distinction threshold range was determined.

7. The method for reconstructing wharf shoreline topography based on remote sensing data according to claim 6, characterized in that, The method for establishing the differentiation model is as follows: A ground feature classifier was constructed using a machine learning classification algorithm. The input features of the ground feature classifier included VV polarization backscattering coefficient, VH polarization backscattering coefficient, VV / VH polarization ratio, and texture feature parameters. Training samples of known types were collected, including typical reef area samples and dock artificial structure area samples, to train the classifier. During training, cross-validation was used to optimize the classifier parameters, raising the classification accuracy to over 85%. The trained classifier was then used to predict land cover types for each pixel in the entire land-water segmentation map. For uncertain areas in the land cover type prediction results, spatial context analysis was used to further distinguish them by combining the distribution characteristics of land cover types around the pixel. Finally, an accurate land-water boundary map with land cover type labels was generated, which distinguishes between reef areas and artificial structure areas such as docks.

8. The method for reconstructing wharf shoreline topography based on remote sensing data according to claim 7, characterized in that, Methods for correcting land cover types in land-water separation maps based on differentiation models include: The predicted land cover types output by the classifier are post-processed, and morphological filtering is used to eliminate isolated misclassified pixels. A spatial distribution rule library for land cover types is established, including the natural distribution characteristics of reef areas and the regular geometric shape features of artificial structures such as wharves. Areas that violate the spatial distribution rules are manually and interactively corrected, and the original multispectral remote sensing images and synthetic aperture radar images are compared and verified during the correction process. The corrected land cover type data is fused with the tidal-corrected land-water boundary map to output an accurate land-water boundary map with complete topological information. Each boundary line segment in the accurate land-water boundary map is labeled with a land cover type attribute. At the same time, a land cover classification accuracy report is generated, recording the classification accuracy indicators and correction statistics for various land cover types.

9. The method for reconstructing wharf shoreline topography based on remote sensing data according to claim 1, characterized in that, Step S4 specifically includes: Gaussian filtering was applied to the accurate land-water boundary map, and convolution was performed using a Gaussian kernel with a standard deviation of 0.

5. The image gradient magnitude and direction of the accurate land-water boundary map were calculated, and convolution was performed in the horizontal and vertical directions using a 3×3 Sobel operator. The gradient magnitude was refined using non-maximum suppression, retaining local maxima. Dual threshold detection and edge connection were used, with the high threshold set to 0.7 times the maximum gradient magnitude and the low threshold set to 0.4 times the high threshold. Broken edge segments were connected using an edge tracking algorithm. The extracted edges were refined to sub-pixel level using B-spline curve fitting, achieving an edge localization accuracy of 0.1 pixels. The refined edges were precisely registered with the digital surface model data, and the nearest neighbor method was used to assign a corresponding elevation value to each edge point. The elevation assignment results were quality controlled, and outlier elevation values ​​exceeding the normal range were removed.

10. The method for reconstructing wharf shoreline topography based on remote sensing data according to claim 1, characterized in that, After assigning elevation values ​​to the shoreline, the elevation data is optimized, including the following steps: A height anomaly detection model was established to identify potential height anomalies by analyzing the height change rate and spatial distribution characteristics of adjacent points. A moving window filtering method was used to smooth the height data, with a window size of 5×5 pixels, and a median filtering algorithm was used to eliminate noise in the height data. A topological consistency check was performed on the height data of the wharf shoreline contour to verify whether the height changes of adjacent points met the requirements of terrain continuity. Special processing rules were established for the unique vertical shoreline structure of the wharf area, and a linear interpolation method was used to supplement missing height data caused by occlusion. Finally, complete wharf shoreline contour data with height attributes was generated, and a height data quality report was output, recording the number of height anomalies and their processing status.

Images