Underwater terrain intelligent modeling method and device based on river terrain key structure

Abstract

The invention belongs to the technical field of underwater terrain modeling, and particularly relates to an underwater terrain intelligent modeling method and device based on a river terrain key structure. According to the method, the river center line recognition model is constructed through transfer learning, automatic recognition of the complex river form is achieved, the problem that the center line of a bifurcated and curved river reach is difficult to extract is effectively solved, a section measurement scheme is designed based on the length and curvature of the river channel, key coverage of sparse water depth measurement points on the river channel form is achieved, and the measurement cost is reduced. Meanwhile, the precision of underwater terrain modeling is guaranteed, section sideline points, thalweg points and slope change points are automatically recognized through a convolutional neural network model, the manual marking workload is reduced, the section feature point extraction efficiency and accuracy are improved, and the generated riverbed height meets measurement data and also meets the requirement of the riverbed height through a center line curvature constraint interpolation method and a curve smooth interpolation method. The river form continuity is kept, and sharp turning or abnormal areas are avoided.

Description

Technical Field

[0001] This invention belongs to the field of underwater terrain modeling technology, specifically relating to an intelligent underwater terrain modeling method and device based on key structures of river terrain. Background Technology

[0002] With the continuous development of water conservancy projects, river ecological protection, and underwater topography research, higher demands are being placed on the accurate modeling of river underwater topography. Traditional underwater topography modeling methods mainly rely on manual measurement or sparse depth measurement data for interpolation. These methods suffer from problems such as sparse measurement points, low modeling accuracy, and complex and inefficient modeling processes, making it difficult to meet the needs of refined modeling of complex river channels. Furthermore, existing methods often fail to fully utilize information from the river centerline and key river cross-sections, resulting in significant errors in the modeling results at river bends or slope change areas, failing to effectively reflect the true topographic features of the river. In recent years, deep learning technology has made significant progress in image recognition, object detection, and spatial feature extraction, providing new ideas for river underwater topography modeling. It offers feasibility for constructing high-precision, continuous, and highly automated underwater topography models.

[0003] However, existing technologies still have shortcomings. They lack systematic modeling methods for the river centerline and key cross-sectional structures, making it difficult to guarantee the accuracy and continuity of modeling complex river areas. The sparsity and uneven distribution of water depth measurement points affect the accuracy of cross-section generation and riverbed feature extraction. Furthermore, they lack an overall intelligent modeling process that combines deep learning, curvature constraint interpolation, and smooth interpolation, making it impossible to achieve efficient and automated underwater terrain generation. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent underwater topography modeling method and device based on key river topographic structures, which can improve modeling accuracy and automation level, and meet the needs of water conservancy projects, river ecological protection and scientific research analysis.

[0005] The specific technical solution adopted by this invention is as follows: An intelligent underwater topography modeling method based on key river topographic structures includes: Obtain sample river surface data and construct a river centerline sample dataset. Use the river centerline sample dataset to perform transfer learning on the DeepCenterline model to obtain a river centerline recognition model. Extract the river centerline based on the river centerline recognition model. Acquire river channel status data, including river channel length and centerline curvature, and combine the river centerline to generate a cross-sectional water depth measurement scheme. Based on the cross-sectional water depth measurement scheme, obtain multiple sparse water depth measurement points to form a water depth measurement point set. The set of water depth measurement points is clustered using a density clustering algorithm to obtain a set of river channel cross-sections; A river cross-section feature dataset was constructed. Based on the river cross-section dataset, a ResNet-50 convolutional neural network model was trained using transfer learning. The trained convolutional neural network model was then used to automatically identify the cross-sectional feature points of each river cross-section. The cross-sectional feature points include river edge points, thallopath points, and slope change points. Based on the cross-sectional feature points and centerline curvature of each river section, centerline curvature constraint interpolation is performed to generate an interpolation point set at a preset distance along the river direction. Multiple interpolation points in the interpolation point set are connected to form a slope change line, a thalweg line, and a side line, respectively. Based on the interpolation point set, a curve smoothing interpolation method is used, and a river underwater terrain model is generated through preset constraints.

[0006] In a preferred embodiment, sample river surface data is acquired, and a river centerline sample dataset is constructed. The DeepCenterline model is then transferred to this dataset to obtain a river centerline identification model. Based on this model, the river centerline is extracted, including: Acquire sample river surface data and preprocess the sample river surface data, including brightness equalization, noise reduction and edge enhancement; Based on the preprocessed sample river surface data, the river centerline annotation information is extracted through annotation. A sample dataset of river centerlines is constructed based on sample river water surface data and corresponding river centerline annotation information; Based on the river centerline sample dataset, the DeepCenterline model is transferred to the river. By adjusting the parameters of the feature extraction layer and the centerline prediction layer, the DeepCenterline model is adapted to the features of the river water surface image, thus obtaining the river centerline recognition model. Acquire the water surface data of the target river section and automatically extract the river centerline based on the river centerline identification model.

[0007] In a preferred embodiment, river channel state data is acquired, including the river channel length and centerline curvature. Combined with the river centerline, a cross-sectional water depth measurement scheme is generated. Based on this scheme, multiple sparse water depth measurement points are obtained, forming a water depth measurement point set, including: Acquire river channel status data, which includes the river channel length along the river centerline and the corresponding centerline curvature. Based on the river centerline and river channel status data, the river segment is divided into morphological zones to obtain the river segment morphology set, which includes the straight river segment morphology set and the meandering river segment morphology set. Based on the river morphology, corresponding cross-sectional water depth measurement intervals are set for different river morphologies, and cross-sectional water depth measurement schemes are generated. Based on the cross-sectional water depth measurement scheme, multiple sparse water depth measurement points are obtained to form a water depth measurement point set.

[0008] In a preferred embodiment, corresponding cross-sectional water depth measurement intervals are set according to different river segment morphologies within the river segment morphology set, and a cross-sectional water depth measurement scheme is generated, including: Obtain the river segment sinusoidal coefficient and river segment length corresponding to different river segment morphologies in the river segment morphology set; Obtain a measurement table, which includes multiple river segment tortuosity coefficient intervals in the longitudinal index and multiple river segment length intervals in the transverse index, as well as the cross-sectional water depth measurement intervals corresponding to each river segment tortuosity coefficient interval and river segment length interval. Based on the river bend coefficient and river length corresponding to each river segment morphology, the corresponding longitudinal index river bend coefficient interval and transverse index river length interval are matched, and the corresponding cross-sectional water depth measurement interval is obtained from the measurement table. By summing up the water depth measurement intervals of multiple cross-sections, a cross-section water depth measurement scheme is obtained.

[0009] In a preferred embodiment, the set of water depth measurement points is clustered using a density clustering algorithm to obtain a set of river channel cross-sections, including: The neighborhood radius parameter and the minimum number of water depth measurement points within a cluster are set based on the density clustering algorithm. Based on the neighborhood radius parameter and the minimum number of water depth measurement points within the cluster parameter, density clustering is performed on the water depth measurement point set to obtain multiple measurement point clusters; The water depth measurement points corresponding to each measurement point cluster are fitted according to the preset spatial relationship to generate the corresponding river cross section, and then summarized to obtain the river cross section set.

[0010] In a preferred embodiment, a river cross-section feature dataset is constructed. A ResNet-50 convolutional neural network model is trained using transfer learning based on this dataset. The trained convolutional neural network model is then used to automatically identify the cross-sectional feature points of each river cross-section. These feature points include river boundary points, thalveolar points, and slope change points, including: Each river cross section is marked in the data set to obtain cross section marking information, which includes the river marking boundary points, marking thalassium points, and marking slope change points; The set of river cross sections and the corresponding cross section annotation information are combined to form a river cross section feature dataset; Based on the river cross-section feature dataset, a pre-trained ResNet-50 convolutional neural network model is trained by transfer learning so that the convolutional neural network model can learn the spatial distribution and morphological features of different feature points in the river cross-section. The trained convolutional neural network model is applied to the target river cross-section data to automatically identify the cross-sectional feature points of multiple river cross-sections, obtaining the cross-sectional feature points of each river cross-section, including river edge points, thallopath points, and slope change points.

[0011] In a preferred embodiment, based on the cross-sectional feature points and centerline curvature of each river channel section, centerline curvature-constrained interpolation is performed to generate an interpolation point set at a preset distance along the river channel direction. Multiple interpolation points within this set are connected to form a slope change line, a thalweg line, and a boundary line, including: Each cross-sectional feature point is projected onto the centerline of the river channel to obtain the centerline projection position parameters corresponding to the cross-sectional feature points. Curvature constraint interpolation is performed on the cross-sectional feature points based on the centerline curvature, generating multiple interpolation point sequences that satisfy the continuity of the centerline curvature between adjacent cross-sectional feature points; The interpolation point sequence obtained by curvature constraint interpolation is resampled according to a preset spacing along the river channel to form a set of interpolation points uniformly distributed along the river channel. Connect the interpolation points in the interpolation point set according to the feature point category to obtain the slope change line, thalweg line, and edge line.

[0012] In a preferred embodiment, a river underwater topography model is generated based on an interpolation point set, using a curve smoothing interpolation method and pre-defined constraints, including: Grid nodes are constructed based on the interpolation point set. Grid nodes corresponding to cross-sectional feature points are set as known nodes, and grid nodes not occupied by cross-sectional feature points are set as nodes to be interpolated. Obtain preset constraints, including water depth constraints, river channel boundary constraints, and smoothness constraints, and achieve curve smooth interpolation by minimizing the objective function to obtain the interpolation function value; Based on the interpolation function value, the riverbed height of the node to be interpolated is determined, and a complete underwater topographic surface of the river is generated; The underwater terrain surface of the river is smoothed to obtain an underwater terrain model.

[0013] The present invention also provides an underwater terrain intelligent modeling device based on key river topographic structures, used in the above-mentioned underwater terrain intelligent modeling method based on key river topographic structures, comprising: The centerline module is used to acquire sample river surface data and construct a river centerline sample dataset. The DeepCenterline model is then transferred and learned using the river centerline sample dataset to obtain a river centerline recognition model. The river centerline is then extracted based on the river centerline recognition model. The water depth measurement module is used to acquire river channel state data, which includes the river channel length and centerline curvature. Combined with the river centerline, a cross-sectional water depth measurement scheme is generated. Based on the cross-sectional water depth measurement scheme, multiple sparse water depth measurement points are acquired to form a water depth measurement point set. The river cross-section module is used to cluster the set of water depth measurement points according to the density clustering algorithm to obtain the set of river cross-sections; The cross-section feature module is used to construct a river cross-section feature dataset. Based on the river cross-section dataset, the ResNet-50 convolutional neural network model is trained by transfer learning, and the trained convolutional neural network model is used to automatically identify the cross-section feature points of each river cross-section. The cross-section feature points include river edge points, thallopath points, and slope change points. The interpolation point module is used to perform centerline curvature constraint interpolation based on the cross-sectional feature points and centerline curvature of each river section, and generate an interpolation point set at a preset distance along the river direction. Multiple interpolation points in the interpolation point set are connected to form slope change line, thalweg line and edge line respectively. The model generation module uses a curve smoothing interpolation method based on the interpolation point set and generates an underwater terrain model of the river through preset constraints.

[0014] And, an underwater terrain intelligent modeling terminal based on key river topographic structures, comprising: One or more processors; A storage device on which one or more programs are stored; When one or more programs are executed by one or more processors, the one or more processors implement an intelligent underwater terrain modeling method based on key structures of river topography.

[0015] The technical effects achieved by this invention are as follows: This invention constructs a river centerline recognition model through transfer learning, enabling automatic identification of complex river morphologies and effectively solving the problem of difficulty in extracting centerlines in bifurcated and meandering river sections. Based on river length and curvature, a cross-section measurement scheme is designed to achieve key coverage of river morphology by sparse depth measurement points, reducing measurement costs while ensuring the accuracy of underwater topographic modeling. A convolutional neural network model automatically identifies cross-section edge points, thalassium points, and slope change points, reducing manual annotation workload and improving the efficiency and accuracy of cross-section feature point extraction. Through centerline curvature constraint interpolation and curve smoothing interpolation, the generated riverbed height conforms to the measurement data while maintaining the continuity of the river morphology, avoiding sharp turns or abnormal areas. Combining multi-source data with deep learning forms an automated workflow, significantly shortening the modeling cycle while ensuring the accuracy of key topographic structures, providing a reliable data foundation for water conservancy management, waterway maintenance, and river simulation. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method provided by the present invention; Figure 2 This is a system module diagram provided by the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0019] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in a preferred embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0020] Furthermore, the present invention will be described in detail with reference to the schematic diagrams. When describing the embodiments of the present invention in detail, the schematic diagrams are merely examples for ease of explanation and should not limit the scope of protection of the present invention.

[0021] Please see the appendix Figure 1 As shown, an intelligent underwater terrain modeling method based on key river topographic structures is provided, including: S1. Obtain sample river surface data and construct a river centerline sample dataset. Use the river centerline sample dataset to perform transfer learning on the DeepCenterline model to obtain a river centerline recognition model. Extract the river centerline based on the river centerline recognition model. S2. Obtain river channel status data, including river channel length and centerline curvature. Combine the river centerline to generate a cross-sectional water depth measurement scheme. Based on the cross-sectional water depth measurement scheme, obtain multiple sparse water depth measurement points to form a water depth measurement point set. S3. Cluster the water depth measurement point set according to the density clustering algorithm to obtain the river channel cross-section set; S4. Construct a river cross-section feature dataset, train the ResNet-50 convolutional neural network model using transfer learning based on the river cross-section dataset, and use the trained convolutional neural network model to automatically identify the cross-section feature points of each river cross-section. The cross-section feature points include river edge points, thallopath points, and slope change points. S5. Based on the cross-sectional feature points and centerline curvature of each river section, perform centerline curvature constraint interpolation to generate an interpolation point set at a preset distance along the river direction. Multiple interpolation points in the interpolation point set are connected to form a slope change line, a thalweg line, and a side line, respectively. S6. Based on the interpolation point set, a smooth curve interpolation method is used, and a river underwater terrain model is generated through preset constraints.

[0022] As described in steps S1 to S6 above, by acquiring sample river surface data from the national land survey and constructing a river centerline sample dataset, the DeepCenterline model is transferred to obtain a river centerline recognition model. The river centerline is automatically extracted from the river surface image of the target river segment, achieving automatic recognition of complex river centerlines. Based on river channel state data, including channel length and centerline curvature, the river segment is morphologically divided into straight and curved sections. A cross-sectional depth measurement scheme is generated using the river centerline to obtain sparse depth measurement points, forming a depth measurement point set. This ensures coverage of depth measurement points at key locations of river channel morphological changes. A density clustering algorithm is then used for depth measurement. The measurement points are clustered into multiple river cross-sections, resulting in a river cross-section set. This set can automatically adapt to sparse or irregular distribution of water depth points. Based on this set, a river cross-section feature dataset is constructed. A ResNet-50 convolutional neural network (CNN) model is trained using transfer learning to identify cross-section feature points, including river edge points, thalveolar points, and slope change points. The trained model automatically identifies target river cross-section data, achieving rapid and accurate extraction of cross-section feature points. Based on the cross-section feature points of each river cross-section and the curvature of the river centerline, a curvature-constrained interpolation point set is generated between cross-sections, forming a uniformly distributed sequence of interpolation points along the river direction. These points are then analyzed according to their features. Point type connections yield slope lines, thalweg lines, and boundary lines, ensuring that the interpolation points spatially follow the river channel curvature. A grid of nodes is constructed on the interpolation point set, with nodes corresponding to cross-sectional feature points designated as known nodes and the remaining nodes designated as nodes to be interpolated. Smooth curve interpolation is achieved by minimizing an objective function that includes smoothness constraints, boundary line constraints, and water depth constraints, yielding the riverbed height and generating a complete underwater topographic model. This model is then smoothed to ensure continuous and natural topography. A river centerline recognition model is constructed through transfer learning, enabling automatic identification of complex river morphologies and effectively solving the problem of extracting centerlines from bifurcated and meandering river sections. A cross-sectional measurement scheme is designed based on river channel length and curvature to achieve sparse water depth measurement. The measurement points provide crucial coverage of river channel morphology, reducing measurement costs while ensuring the accuracy of underwater topographic modeling. A convolutional neural network model automatically identifies cross-section edge points, thallopath points, and slope change points, reducing manual annotation workload and improving the efficiency and accuracy of cross-section feature point extraction. Through centerline curvature constraint interpolation and curve smoothing interpolation, the generated riverbed height conforms to the measurement data while maintaining the continuity of the river channel morphology, avoiding sharp turns or abnormal areas. Combining multi-source data with deep learning creates an automated workflow, significantly shortening the modeling cycle while ensuring the accuracy of key topographic structures (such as thallopaths, slope changes, and edge points), providing a reliable data foundation for water conservancy management, waterway maintenance, and river channel simulation.

[0023] In a preferred implementation, sample river surface data is acquired, and a river centerline sample dataset is constructed. The DeepCenterline model is then transferred to this dataset to obtain a river centerline recognition model. The river centerline is extracted based on this model, including: S101. Obtain sample river surface data and preprocess the sample river surface data, including brightness equalization, noise reduction and edge enhancement. S102. Based on the preprocessed sample river surface data, extract the river centerline annotation information through annotation. S103. Construct a river centerline sample dataset based on the sample river water surface data and the corresponding river centerline annotation information; S104. Based on the river centerline sample dataset, transfer learning is performed on the DeepCenterline model. By adjusting the parameters of the feature extraction layer and the centerline prediction layer, the DeepCenterline model is adapted to the features of the river water surface image, thus obtaining the river centerline recognition model. S105. Obtain the water surface data of the target river section and automatically extract the river centerline based on the river centerline identification model.

[0024] As described in steps S101 to S105 above, river surface image data provided by national census or remote sensing is acquired, and preprocessing operations such as brightness equalization, noise removal, and edge enhancement are performed on the image data to improve the recognizability of river texture features. Based on the preprocessed river surface data, river centerline annotation information is extracted through manual annotation and semi-automatic annotation methods. The preprocessed river surface images are paired with the corresponding centerline annotation information to construct a river centerline sample dataset. This dataset contains rich river channel morphology samples, covering the curvature and width variations of different river sections. The constructed dataset is used to perform transfer learning on the DeepCenterline model. By adjusting the parameters of the feature extraction layer and the centerline prediction layer, the model can be adapted to river surface conditions. Image features are used to form a river centerline recognition model (Rever-DeepCenterline). Transfer learning utilizes the pre-trained knowledge of the existing model to accelerate the training process and improve the model's adaptability to complex river morphology. The trained river centerline recognition model is applied to the river surface image of the river section to be tested to achieve automatic extraction of the river centerline. It can obtain the continuous centerline of the river channel, including branches and bends. The river centerline recognition model constructed through transfer learning can adapt to the image features of different river sections and achieve accurate extraction of complex river centerlines (including bends and branches). The transfer learning method enables the model to maintain high recognition ability under different river sections and different image conditions, and can handle river sections with large variations in channel width and complex textures.

[0025] In a preferred embodiment, river channel state data is acquired, including the river channel length and centerline curvature. Combined with the river centerline, a cross-sectional water depth measurement scheme is generated. Based on this scheme, multiple sparse water depth measurement points are obtained, forming a water depth measurement point set, including: S201. Obtain river channel status data, which includes the river channel length along the river centerline and the centerline curvature at the corresponding location. S202. Based on the river centerline and river channel status data, the river segment is divided into morphological zones to obtain the river segment morphology set, which includes the straight river segment morphology set and the meandering river segment morphology set. S203. Based on the river section morphology, set corresponding cross-sectional water depth measurement intervals for different river section morphologies and generate cross-sectional water depth measurement schemes. S204. Based on the cross-sectional water depth measurement scheme, obtain multiple sparse water depth measurement points to form a water depth measurement point set.

[0026] As described in steps S201 to S204 above, the river channel length and corresponding centerline curvature data along the river centerline direction are obtained. River channel length reflects the spatial extension of the river, while centerline curvature reflects the degree of river channel curvature and morphological complexity. Based on the river centerline and channel condition data, the river segment is divided into morphological zones: straight sections and meandering sections. Straight sections show slow channel changes, while meandering sections show significant changes. For different river segment morphologies, corresponding cross-sectional depth measurement intervals are set. For example, straight sections can have larger intervals (e.g., 0.5–1 km), while meandering sections add key cross-sectional measurement points to capture channel changes. A complete cross-sectional depth measurement scheme is generated by combining the river segment morphology set, achieving optimized layout of measurement points. Following the cross-sectional depth measurement scheme, the... Multiple sparse depth measurement points are acquired using unmanned surface vessel single-beam echo sounders or other echo sounding equipment, forming a depth measurement point set. The collected data can cover key locations in the river section. Based on the cross-section planning of the river channel state and centerline information, the measurement points can be rationally deployed, avoiding blind measurements and significantly improving the efficiency of depth measurement. River section morphology zoning ensures that depth information in curved and key river sections is fully collected, guaranteeing that underwater topographic data covers the main features of the river. By generating cross-section depth measurement schemes, unmanned surface vessels or automated echo sounding equipment can be guided to perform data collection tasks, reducing human intervention and improving the intelligence and repeatability of the measurement process. The zonal measurement strategy can adapt to straight, curved, and complex river sections, ensuring the measurement accuracy and data representativeness of different river sections and enhancing the generalization ability of the model.

[0027] In a preferred embodiment, corresponding cross-sectional water depth measurement intervals are set according to different river segment morphologies, and a cross-sectional water depth measurement scheme is generated, including: S2031. Obtain the river segment curvature coefficient and river segment length corresponding to different river segment morphologies in the river segment morphology set; S2032. Obtain a measurement table, wherein the measurement table includes multiple river segment curvature coefficient intervals in the longitudinal index and multiple river segment length intervals in the transverse index, as well as the cross-sectional water depth measurement intervals corresponding to each river segment curvature coefficient interval and river segment length interval. S2033. Based on the river bend coefficient and river length corresponding to each river segment morphology, match the corresponding longitudinal index river bend coefficient interval and the transverse index river length interval, and obtain the corresponding cross-sectional water depth measurement interval from the measurement table. S2034. Summarize the water depth measurement intervals of multiple cross-sections to obtain the cross-section water depth measurement scheme.

[0028] As described in steps S2031 to S2034 above, for each river segment morphology, its river segment sinuosity coefficient and river segment length are obtained. The river segment sinuosity coefficient reflects the degree of sinuosity of the river segment, and the length reflects the spatial range of the river segment. A preset measurement table is obtained. This table is generated based on the river segment sinuosity, length, and modeling accuracy requirements through specifications, historical experience, or simulation analysis. It records multiple river segment sinuosity coefficient intervals with a vertical index and multiple river segment length intervals with a horizontal index. Each sinuosity coefficient interval and length interval corresponds to a cross-sectional water depth measurement interval. The measurement table associates the geometric characteristics of the river segment with the measurement interval to achieve quantitative management. For each river segment, a matching is performed based on its sinuosity coefficient and length. The corresponding tortuosity coefficient and length intervals in the measurement table are used to obtain the recommended cross-sectional water depth measurement intervals, ensuring that the measurement intervals are scientific and reasonable and can reflect changes in river morphology. The cross-sectional water depth measurement intervals obtained from each river segment are summarized to form a complete cross-sectional water depth measurement scheme, which is used to guide subsequent water depth measurement data collection. The cross-sectional intervals are determined by combining the river segment tortuosity coefficient and length with the measurement table, so that the intervals are larger in straight segments and smaller in curved segments, ensuring that water depth changes in key river segments are fully captured. The measurement density is increased in curved segments and the intervals are moderate in straight segments to ensure that key underwater topographic features (such as thalweg, slope changes, and edge lines) are accurately reflected, thereby improving the accuracy of subsequent modeling.

[0029] In a preferred embodiment, the set of water depth measurement points is clustered according to a density clustering algorithm to obtain a set of river channel cross-sections, including: S301. Set the neighborhood radius parameter and the minimum number of water depth measurement points within the cluster based on the density clustering algorithm; S302. Based on the neighborhood radius parameter and the minimum number of water depth measurement points within the cluster parameter, density clustering is performed on the water depth measurement point set to obtain multiple measurement point clusters. S303. Fit the water depth measurement points corresponding to each measurement point cluster according to the preset spatial relationship to generate the corresponding river cross section, and summarize them to obtain the river cross section set.

[0030] As described in steps S301 to S303 above, necessary parameters are set for the density clustering algorithm (DBSCAN). The neighborhood radius is used to define the spatial proximity of the measurement points. The preset number of water depth measurement points within a cluster (i.e., the minimum number of water depth measurement points) is used to determine the minimum number of points required to form an effective cluster. Based on the set neighborhood radius and minimum number of points, density clustering is performed on the water depth measurement point set. The algorithm identifies high-density areas and groups high-density adjacent measurement points into a cluster, while classifying discrete points as noise, resulting in multiple measurement point clusters. Each cluster represents a potential cross-sectional area in the river channel. The water depth measurement points within each measurement point cluster are fitted according to a preset spatial relationship to generate the corresponding river channel cross-section. For example, the points within the cluster can be sorted and a curve fitted according to the centerline direction. Finally, all cross-sections are summarized to obtain a complete set of river channel cross-sections. The density clustering algorithm can identify high-density areas as cross-sections, while treating isolated points as noise, avoiding interference from outliers in cross-section generation and improving data reliability.

[0031] In a preferred embodiment, a river cross-section feature dataset is constructed. A ResNet-50 convolutional neural network model is trained using transfer learning based on this dataset. The trained convolutional neural network model is then used to automatically identify the cross-sectional feature points of each river cross-section. These feature points include river boundary points, thalveolar points, and slope change points, including: S401. Mark each river section in the river section collection to obtain the section marking information, which includes the river marking edge point, the marking thalweg point and the marking slope change point; S402. Combine the river cross-section set with the corresponding cross-section annotation information to form a river cross-section feature dataset; S403. Based on the river cross-section feature dataset, transfer learning training is performed on the pre-trained ResNet-50 convolutional neural network model so that the convolutional neural network model can learn the spatial distribution and morphological features of different feature points in the river cross-section. S404. Apply the trained convolutional neural network model to the target river cross-section data to automatically identify the cross-sectional feature points of multiple river cross-sections, and obtain the cross-sectional feature points of each river cross-section, including river edge points, thalassium points, and slope change points.

[0032] As described in steps S401 to S404 above, each cross-section in the river cross-section set is manually or semi-automatically labeled to obtain cross-section labeling information, including river boundary points, thallopath points, and slope change points. The river cross-section set is then combined with the corresponding cross-section labeling information to construct a river cross-section feature dataset. This dataset contains a pair of inputs (cross-section images or depth data) and outputs (labeled feature point locations), serving as the basis for training a convolutional neural network (CNN) model. Based on a pre-trained ResNet-50 CNN model, the feature extraction layer and output layer are adapted and trained through transfer learning. The model is able to identify the spatial distribution and morphological features of different feature points within a river cross-section. Through training, the convolutional neural network model learns the local and global structural features of the cross-sectional curve, thus possessing the ability to automatically identify cross-sectional feature points. The trained convolutional neural network is applied to the target river cross-sectional data, automatically outputting the coordinates of feature points for each cross-section, including river edge points, thallopath points, and slope change points, achieving automated extraction of cross-sectional feature points. Utilizing the powerful feature extraction capabilities of the convolutional neural network model, it can effectively identify edge points, thallopath points, and slope change points in complex river sections, overcoming the subjectivity and errors of manual identification.

[0033] In a preferred embodiment, based on the cross-sectional feature points and centerline curvature of each river channel section, centerline curvature-constrained interpolation is performed to generate an interpolation point set at a preset distance along the river channel direction. Multiple interpolation points within the interpolation point set are connected to form a slope change line, a thalweg line, and a boundary line, including: S501. Project each cross-sectional feature point to the centerline of the river channel to obtain the centerline projection position parameters corresponding to the cross-sectional feature points. S502. Perform curvature constraint interpolation on the cross-sectional feature points based on the centerline curvature, and generate multiple interpolation point sequences that satisfy the continuity of the centerline curvature between adjacent cross-sectional feature points. S503. Resample the interpolation point sequence obtained by curvature constraint interpolation according to the preset spacing along the river channel to form a set of interpolation points uniformly distributed along the river channel. S504. Connect the interpolation points in the interpolation point set according to the feature point category to obtain the slope change line, thalweg line and edge line.

[0034] As described in steps S501 to S504 above, each cross-sectional feature point (including slope change points, thallophon points, and edge points) is projected onto the river centerline. The corresponding projection position parameters of each feature point on the centerline are obtained to establish the spatial relationship between the cross-sectional feature points and the centerline. Using the curvature information of the centerline, interpolation is performed on adjacent cross-sectional feature points along the river direction to generate an interpolation point sequence that satisfies the continuity of the centerline curvature. This ensures that the interpolation points maintain consistency with the river centerline in terms of lateral distance and river segment morphology, guaranteeing that the generated slope change lines, thallophon lines, and edge lines naturally conform to the river channel morphology. The curvature-constrained interpolation results in… The interpolation point sequence is resampled at a preset interval (e.g., every 10 meters) to form a set of interpolation points evenly distributed along the river channel. The interpolation point set is then connected according to the feature point category to form river slope change lines, thalweg lines, and boundary lines, which respectively reflect the changes in riverbed slope, the deepest point of the riverbed, and the river boundary line. Curvature constraint interpolation ensures that the lines generated by the feature points match the shape of the river centerline, so that the curvature and width changes of the river channel are naturally reflected in the feature lines. By resampling along the river channel to generate uniform interpolation points, the data discontinuity caused by uneven spacing between feature points between cross sections is reduced, improving the continuity and accuracy of underwater terrain modeling.

[0035] In a preferred embodiment, a river underwater topography model is generated based on an interpolation point set, using a curve smoothing interpolation method and pre-defined constraints, including: S601. Construct grid nodes based on the interpolation point set, set the grid nodes corresponding to the cross-section feature points as known nodes, and set the grid nodes not occupied by the cross-section feature points as nodes to be interpolated. S602. Obtain preset constraints, including water depth constraints, river channel boundary constraints, and smoothness constraints. Then, achieve curve smooth interpolation by minimizing the objective function to obtain the interpolation function value. S603. Based on the interpolation function value, determine the riverbed height of the node to be interpolated and generate a complete underwater topographic surface of the river. S604. Smooth the underwater terrain surface of the river to obtain an underwater terrain model.

[0036] As described in steps S601 to S604 above, regular or irregular grid nodes are constructed based on the interpolation point set along the river channel direction and cross-sectional feature points. Nodes containing cross-sectional feature points are marked as known nodes, and the remaining nodes are set as nodes to be interpolated. The constraints required for underwater terrain modeling are obtained, including water depth constraints (ensuring the water depth at known cross-sectional feature points remains true), river channel boundary constraints (ensuring the shape of the river channel boundary matches the actual boundary), and smoothness constraints (controlling the smoothness of riverbed surface changes to avoid overly sharp or unnatural undulations). Under these constraints, curve smoothing interpolation is performed by minimizing the objective function to obtain the interpolation function values ​​for the nodes to be interpolated, ensuring that the riverbed elevation distribution conforms to the known... The measured values ​​are kept naturally smooth. Based on the interpolation function value obtained by curve smoothing interpolation, the riverbed height is determined for the nodes to be interpolated, thus forming a continuous and complete underwater river topography surface. The generated underwater river topography surface is post-processed and smoothed to further eliminate local discontinuities or noise, improve the smoothness and visualization effect of the topography surface, and generate an underwater topography model. By constraining known cross-sectional feature points and interpolation points, the height of the generated riverbed surface is ensured to be accurate, reflecting the actual riverbed variation characteristics. The curve smoothing interpolation method combined with smoothness constraints generates a naturally smooth riverbed surface, avoiding discontinuities or abrupt changes in discrete points. It can uniformly process curved, bifurcated or widened river sections to generate an underwater topography model that fits the actual river channel morphology.

[0037] Please see the appendix Figure 2 As shown, the present invention also provides an underwater terrain intelligent modeling device based on key river topographic structures, used in the above-mentioned underwater terrain intelligent modeling method based on key river topographic structures, comprising: The centerline module is used to acquire sample river surface data and construct a river centerline sample dataset. The DeepCenterline model is then transferred and learned using the river centerline sample dataset to obtain a river centerline recognition model. The river centerline is then extracted based on the river centerline recognition model. The water depth measurement module is used to acquire river channel state data, which includes the river channel length and centerline curvature. Combined with the river centerline, a cross-sectional water depth measurement scheme is generated. Based on the cross-sectional water depth measurement scheme, multiple sparse water depth measurement points are acquired to form a water depth measurement point set. The river cross-section module is used to cluster the set of water depth measurement points according to the density clustering algorithm to obtain the set of river cross-sections; The cross-section feature module is used to construct a river cross-section feature dataset. Based on the river cross-section dataset, the ResNet-50 convolutional neural network model is trained by transfer learning, and the trained convolutional neural network model is used to automatically identify the cross-section feature points of each river cross-section. The cross-section feature points include river edge points, thallopath points, and slope change points. The interpolation point module is used to perform centerline curvature constraint interpolation based on the cross-sectional feature points and centerline curvature of each river section, and generate an interpolation point set at a preset distance along the river direction. Multiple interpolation points in the interpolation point set are connected to form slope change line, thalweg line and edge line respectively. The model generation module uses a curve smoothing interpolation method based on the interpolation point set and generates an underwater terrain model of the river through preset constraints.

[0038] The above describes the following modules: The centerline module acquires sample river surface data and constructs a river centerline dataset. It then performs transfer learning on the DeepCenterline model to obtain a river centerline recognition model (Rever-DeepCenterline). This model automatically extracts the river centerline of the target river segment. The depth measurement module acquires river channel state data, including the channel length along the centerline direction and the corresponding centerline curvature. Based on the river segment morphology and channel state, it generates a cross-sectional depth measurement scheme, acquiring sparse depth measurement points to form a depth measurement point set. The river cross-section module clusters the depth measurement point set using a density-based clustering algorithm (DBSCAN) to generate a river cross-section set. Each measurement point cluster is fitted into a cross-section according to a preset spatial relationship, realizing the transformation of sparse measurement data into cross-sectional structures. The cross-section feature module constructs a river cross-section feature dataset and performs transfer learning on a pre-trained ResNet-50 convolutional neural network model based on this dataset to automatically identify the cross-sectional feature points of each river cross-section. The module includes river boundary points, thalweg points, and slope change points. The interpolation point module performs centerline curvature constraint interpolation based on cross-sectional feature points and centerline curvature, generating a set of interpolation points at preset distances (e.g., every 10 meters) along the river channel. These interpolation points are connected according to feature point categories to form slope change lines, thalweg lines, and boundary lines, achieving continuous representation of key riverbed structures. The model generation module constructs a grid of nodes based on the interpolation point set, setting cross-sectional feature points as known nodes and the remaining nodes as nodes to be interpolated, obtaining water depth, channel boundary, and smoothness constraints. By minimizing the objective function to achieve smooth curve interpolation, the riverbed height of the nodes to be interpolated is obtained, generating a complete and continuous underwater topographic surface of the river. After smoothing, the final underwater topographic model is obtained. By combining the river centerline, cross-sectional feature points and water depth measurement data, the height of the generated riverbed model is ensured to be realistic and accurate. The curve smooth interpolation method, combined with smoothness constraints, realizes a natural and continuous riverbed surface, avoiding irregularities or abrupt changes caused by discrete points. It can handle curved, widened or bifurcated river sections and generate an underwater model that fits the actual river topography.

[0039] And, an underwater terrain intelligent modeling terminal based on key river topographic structures, comprising: One or more processors; A storage device on which one or more programs are stored; When one or more programs are executed by one or more processors, the one or more processors implement an intelligent underwater terrain modeling method based on key structures of river topography.

[0040] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A method for intelligent underwater topography modeling based on key structures of river topography, characterized in that, include: Obtain sample river surface data and construct a river centerline sample dataset. Use the river centerline sample dataset to perform transfer learning on the DeepCenterline model to obtain a river centerline recognition model. Extract the river centerline based on the river centerline recognition model. Acquire river channel status data, including river channel length and centerline curvature, and combine the river centerline to generate a cross-sectional water depth measurement scheme. Based on the cross-sectional water depth measurement scheme, obtain multiple sparse water depth measurement points to form a water depth measurement point set. The set of water depth measurement points is clustered using a density clustering algorithm to obtain a set of river channel cross-sections; A river cross-section feature dataset was constructed. Based on the river cross-section dataset, a ResNet-50 convolutional neural network model was trained using transfer learning. The trained convolutional neural network model was then used to automatically identify the cross-sectional feature points of each river cross-section. The cross-sectional feature points include river edge points, thallopath points, and slope change points. Based on the cross-sectional feature points and centerline curvature of each river section, centerline curvature constraint interpolation is performed to generate an interpolation point set at a preset distance along the river direction. Multiple interpolation points in the interpolation point set are connected to form a slope change line, a thalweg line, and a side line, respectively. Based on the interpolation point set, a curve smoothing interpolation method is used, and a river underwater terrain model is generated through preset constraints.

2. The underwater terrain intelligent modeling method based on key river topographic structures according to claim 1, characterized in that, Obtain sample river surface data and construct a river centerline sample dataset. Use this dataset to perform transfer learning on the DeepCenterline model to obtain a river centerline recognition model. Extract the river centerline based on this model, including: Acquire sample river surface data and preprocess the sample river surface data, including brightness equalization, noise reduction and edge enhancement; Based on the preprocessed sample river surface data, the river centerline annotation information is extracted through annotation. A sample dataset of river centerlines is constructed based on sample river water surface data and corresponding river centerline annotation information; Based on the river centerline sample dataset, the DeepCenterline model is transferred to the river. By adjusting the parameters of the feature extraction layer and the centerline prediction layer, the DeepCenterline model is adapted to the features of the river water surface image, thus obtaining the river centerline recognition model. Acquire the water surface data of the target river section and automatically extract the river centerline based on the river centerline identification model.

3. The underwater terrain intelligent modeling method based on key river topographic structures according to claim 1, characterized in that, Acquire river channel state data, including channel length and centerline curvature. Combine this data with the river centerline to generate a cross-sectional water depth measurement scheme. Based on this scheme, obtain multiple sparse water depth measurement points to form a water depth measurement point set, including: Acquire river channel status data, which includes the river channel length along the river centerline and the corresponding centerline curvature. Based on the river centerline and river channel status data, the river segment is divided into morphological zones to obtain the river segment morphology set, which includes the straight river segment morphology set and the meandering river segment morphology set. Based on the river morphology, corresponding cross-sectional water depth measurement intervals are set for different river morphologies, and cross-sectional water depth measurement schemes are generated. Based on the cross-sectional water depth measurement scheme, multiple sparse water depth measurement points are obtained to form a water depth measurement point set.

4. The underwater terrain intelligent modeling method based on key river topographic structures according to claim 3, characterized in that, Based on the river segment morphology, corresponding cross-sectional water depth measurement intervals are set for different river segment morphologies, and a cross-sectional water depth measurement scheme is generated, including: Obtain the river segment sinusoidal coefficient and river segment length corresponding to different river segment morphologies in the river segment morphology set; Obtain a measurement table, which includes multiple river segment tortuosity coefficient intervals in the longitudinal index and multiple river segment length intervals in the transverse index, as well as the cross-sectional water depth measurement intervals corresponding to each river segment tortuosity coefficient interval and river segment length interval. Based on the river bend coefficient and river length corresponding to each river segment morphology, the corresponding longitudinal index river bend coefficient interval and transverse index river length interval are matched, and the corresponding cross-sectional water depth measurement interval is obtained from the measurement table. By summing up the water depth measurement intervals of multiple cross-sections, a cross-section water depth measurement scheme is obtained.

5. The underwater terrain intelligent modeling method based on key river topographic structures according to claim 1, characterized in that, The set of water depth measurement points is clustered using a density clustering algorithm to obtain a set of river channel cross-sections, including: The neighborhood radius parameter and the minimum number of water depth measurement points within a cluster are set based on the density clustering algorithm. Based on the neighborhood radius parameter and the minimum number of water depth measurement points within the cluster parameter, density clustering is performed on the water depth measurement point set to obtain multiple measurement point clusters; The water depth measurement points corresponding to each measurement point cluster are fitted according to the preset spatial relationship to generate the corresponding river cross section, and then summarized to obtain the river cross section set.

6. The underwater terrain intelligent modeling method based on key river topographic structures according to claim 1, characterized in that, A river cross-section feature dataset was constructed. Based on this dataset, a ResNet-50 convolutional neural network model was trained using transfer learning. The trained model was then used to automatically identify the cross-sectional feature points for each river cross-section. These feature points include river boundary points, thallopath points, and slope change points, including: Each river cross section is marked in the data set to obtain cross section marking information, which includes the river marking boundary points, marking thalassium points, and marking slope change points; The set of river cross sections and the corresponding cross section annotation information are combined to form a river cross section feature dataset; Based on the river cross-section feature dataset, a pre-trained ResNet-50 convolutional neural network model is trained by transfer learning so that the convolutional neural network model can learn the spatial distribution and morphological features of different feature points in the river cross-section. The trained convolutional neural network model is applied to the target river cross-section data to automatically identify the cross-sectional feature points of multiple river cross-sections, obtaining the cross-sectional feature points of each river cross-section, including river edge points, thallopath points, and slope change points.

7. The underwater terrain intelligent modeling method based on key river topographic structures according to claim 1, characterized in that, Based on the cross-sectional feature points and centerline curvature of each river channel section, centerline curvature-constrained interpolation is performed to generate an interpolation point set at a preset distance along the river channel direction. Multiple interpolation points within this set are connected to form slope change lines, thalweg lines, and edge lines, including: Each cross-sectional feature point is projected onto the centerline of the river channel to obtain the centerline projection position parameters corresponding to the cross-sectional feature points. Curvature constraint interpolation is performed on the cross-sectional feature points based on the centerline curvature, generating multiple interpolation point sequences that satisfy the continuity of the centerline curvature between adjacent cross-sectional feature points; The interpolation point sequence obtained by curvature constraint interpolation is resampled according to a preset spacing along the river channel to form a set of interpolation points uniformly distributed along the river channel. Connect the interpolation points in the interpolation point set according to the feature point category to obtain the slope change line, thalweg line, and edge line.

8. The underwater terrain intelligent modeling method based on key river topographic structures according to claim 1, characterized in that, Based on the interpolation point set, a curve smoothing interpolation method is used, and a river underwater terrain model is generated through preset constraints, including: Grid nodes are constructed based on the interpolation point set. Grid nodes corresponding to cross-sectional feature points are set as known nodes, and grid nodes not occupied by cross-sectional feature points are set as nodes to be interpolated. Obtain preset constraints, including water depth constraints, river channel boundary constraints, and smoothness constraints, and achieve curve smooth interpolation by minimizing the objective function to obtain the interpolation function value; Based on the interpolation function value, the riverbed height of the node to be interpolated is determined, and a complete underwater topographic surface of the river is generated; The underwater terrain surface of the river is smoothed to obtain an underwater terrain model.

9. An intelligent underwater terrain modeling device based on key river topographic structures, applied to the intelligent underwater terrain modeling method based on key river topographic structures as described in any one of claims 1 to 8, characterized in that, include: The centerline module is used to acquire sample river surface data and construct a river centerline sample dataset. The DeepCenterline model is then transferred and learned using the river centerline sample dataset to obtain a river centerline recognition model. The river centerline is then extracted based on the river centerline recognition model. The water depth measurement module is used to acquire river channel state data, which includes the river channel length and centerline curvature. Combined with the river centerline, a cross-sectional water depth measurement scheme is generated. Based on the cross-sectional water depth measurement scheme, multiple sparse water depth measurement points are acquired to form a water depth measurement point set. The river cross-section module is used to cluster the set of water depth measurement points according to the density clustering algorithm to obtain the set of river cross-sections; The cross-section feature module is used to construct a river cross-section feature dataset. Based on the river cross-section dataset, the ResNet-50 convolutional neural network model is trained by transfer learning, and the trained convolutional neural network model is used to automatically identify the cross-section feature points of each river cross-section. The cross-section feature points include river edge points, thallopath points, and slope change points. The interpolation point module is used to perform centerline curvature constraint interpolation based on the cross-sectional feature points and centerline curvature of each river section, and generate an interpolation point set at a preset distance along the river direction. Multiple interpolation points in the interpolation point set are connected to form slope change line, thalweg line and edge line respectively. The model generation module uses a curve smoothing interpolation method based on the interpolation point set and generates an underwater terrain model of the river through preset constraints.

10. An intelligent underwater terrain modeling terminal based on key river topographic structures, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When one or more programs are executed by one or more processors, the one or more processors implement the underwater terrain intelligent modeling method based on key river terrain structures as described in any one of claims 1 to 8.

Images