Underwater topographic survey method and system based on unmanned ship data acquisition

Abstract

The invention discloses an underwater topographic survey method and system based on unmanned ship data acquisition, and relates to the technical field of underwater topographic survey, and the method comprises the steps: generating a dynamic environment map based on acquired multi-modal environment data; dynamically sensing the underwater environment by using a deep Q learning algorithm based on the dynamic environment map, and predicting a potential topographic change probability; judging whether to update the dynamic environment map or not according to the predicted potential terrain change probability, and planning a navigation path by utilizing the updated dynamic environment map based on an RRT algorithm; collecting topographic data based on the navigation path, generating a three-dimensional topographic model, and detecting a shadow region in the three-dimensional topographic model by using U-Net; according to the detected shadow area, generating a complementary collection point through geometric analysis, adjusting a navigation path according to the complementary collection point, performing complementary collection on the shadow area, and obtaining complementary collection data; and fusing the complementary collection data with the three-dimensional terrain model to generate a terrain survey report. According to the invention, through dynamic environment perception, path planning, shadow area detection and supplementary mining, the efficiency and precision of underwater topographic survey are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of underwater terrain surveying, and in particular to an underwater terrain surveying method and system based on unmanned vessel data acquisition. Background Art

[0002] With the growing demands for underwater resource development, environmental protection, and national defense, underwater terrain survey technology has developed rapidly. Traditional underwater terrain surveys typically rely on manned vessels or fixed-site sonar systems, but suffer from low efficiency, limited coverage, and high costs. Unmanned surface vehicles (USVs), as a new type of intelligent survey platform, offer advantages such as high automation, flexible operation, and the ability to operate in dangerous or complex waters. They are gradually becoming an important tool for underwater terrain surveys. Combining the development of multimodal sensors (such as multi-beam sonar and side-scan sonar) with artificial intelligence algorithms, USVs have significant potential to improve the efficiency and accuracy of underwater terrain surveys. However, existing technologies still have shortcomings in dynamic path planning and data acquisition integrity, and urgently need to be optimized.

[0003] The main problems with existing technologies are: First, the underwater environment is complex and ever-changing. Existing path planning algorithms rely heavily on static environmental data, making it difficult to respond promptly to obstacles or changes in water flow, resulting in reduced survey efficiency and even mission failure. Second, in terrain modeling, uncovered shadow areas often appear due to terrain obstruction or sensor blind spots, affecting model integrity. Although some technologies attempt to address this issue by improving sensor resolution or increasing acquisition density, they lack effective methods that combine real-time path optimization with supplementary acquisition mechanisms, making it difficult to achieve a balanced approach for efficiency and accuracy. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an underwater terrain survey method based on unmanned vessel data collection to solve the problems of insufficient adaptability of dynamic path planning and missing terrain modeling data.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides an underwater terrain survey method based on unmanned vessel data collection, which includes generating a dynamic environment map based on the collected multimodal environment data; based on the dynamic environment map, dynamically sensing the underwater environment using a deep Q learning algorithm to predict the probability of potential terrain changes; judging whether to update the dynamic environment map according to the predicted probability of potential terrain changes, and planning a navigation path using the updated dynamic environment map based on the RRT algorithm; collecting terrain data based on the navigation path and generating a three-dimensional terrain model, and detecting shadow areas in the three-dimensional terrain model using U-Net; generating supplementary sampling points through geometric analysis based on the detected shadow areas, adjusting the navigation path according to the supplementary sampling points, supplementary sampling of the shadow areas, and obtaining supplementary sampling data; fusing the supplementary sampling data with the three-dimensional terrain model to generate a terrain survey report.

[0007] As a preferred solution of the underwater terrain survey method based on unmanned vessel data collection of the present invention, wherein: the dynamic environment map is generated based on the collected multimodal environment data, and the specific steps are as follows: Collect multimodal environmental data in real time through sensors; multimodal environmental data includes underwater terrain, obstacle distribution, and water flow speed; Bayesian fusion is used to fuse multimodal environmental data, and combined with GPS to spatially locate the multimodal environmental data, and the navigation trajectory of the unmanned vessel is generated based on the spatial positioning; Based on the navigation trajectory of the unmanned vessel and multimodal environmental data, a dynamic environmental map is generated through Delaunay.

[0008] As a preferred solution of the underwater terrain survey method based on unmanned vessel data collection described in the present invention, wherein: based on the dynamic environment map, the deep Q learning algorithm is used to dynamically perceive the underwater environment and predict the probability of potential terrain changes. The specific steps are as follows: Through feature extraction and discretization methods, the data features in the dynamic environment map are converted into the state space of the deep Q learning algorithm, and the data features of the local area of ​​the current underwater environment are used as the state in the state space; Based on the operation of the unmanned ship in the current environment as an action, define the action space; The comprehensive reward value is calculated based on the distance between the unmanned ship and the obstacle, the coverage ratio of the target area and the navigation cost. ; Based on comprehensive reward value , combining the divergence of the water velocity field, terrain depth and obstacle density of the underwater environment to predict the probability of potential terrain changes .

[0009] As a preferred solution of the underwater terrain survey method based on unmanned vessel data collection according to the present invention, the following steps are used to determine whether to update the dynamic environment map based on the predicted probability of potential terrain changes, and to plan the navigation path using the updated dynamic environment map based on the RRT algorithm: Based on the frequency and magnitude of historical terrain changes, a threshold terrain change risk threshold C1 is defined; when When ≥C1, it is considered that the potential change probability of the current regional terrain is high, and the dynamic environment map is updated; when When <C1, it is considered that the potential change probability of the current regional terrain is low and there is no need to update the dynamic environment map; Initialize the starting node of the navigation path and target node Coordinates Initialize the RRT tree ; Set rule constraints and generate sampling points based on the rule constraints in the updated dynamic environment map , and based on the candidate nodes in the current path , find the sampling point The nearest node ; Connect the distance sampling points in the current RRT tree T The nearest node and sampling points , and through iterative expansion, generate To the target node navigation path.

[0010] As a preferred solution of the underwater terrain survey method based on unmanned vessel data collection according to the present invention, the terrain data is collected based on the navigation path and a three-dimensional terrain model is generated, and the shadow area in the three-dimensional terrain model is detected using U-Net. The specific steps are as follows: The unmanned ship visits the path nodes in sequence along the planned navigation path, and collects the initial underwater terrain data through the sensor equipment at each path node; the path nodes include the starting node of the unmanned ship on the navigation path , distance sampling point The nearest node and the target node ; Use Delaunay triangulation method to convert the initial terrain data into triangular mesh data, and then import the triangular mesh data into OpenGL to generate a three-dimensional terrain model; The 3D terrain model is projected into a 2D depth map, which is then fed into a U-Net network. The encoder in the U-Net network is used to extract terrain features, and the decoder in the U-Net network is used to decode the terrain features. A binary shadow mask with the same size as the 2D depth map is generated, and shadow areas in the 3D terrain model are identified and marked based on the binary shadow mask.

[0011] As a preferred solution of the underwater terrain survey method based on unmanned vessel data collection according to the present invention, wherein: according to the detected shadow area, supplementary sampling points are generated through geometric analysis, the navigation path is adjusted according to the supplementary sampling points, and supplementary sampling is performed on the shadow area to obtain supplementary sampling data. The specific steps are as follows: Using geometric analysis methods, the boundary contour of the shadow area is extracted to form a boundary point set B; Discretize the shadow area and generate a regular grid covering the shadow area; Based on the boundary point set B, the center coordinates of each regular grid are used as the supplementary mining point to generate the supplementary mining point set V; Calculate the distance between each supplementary mining point in the supplementary mining point set V and the current position of the unmanned vessel, and insert the nearest supplementary mining points into the current navigation path in sequence, and adjust the current navigation path based on the rule constraints; The shadow area is supplemented with samples according to the adjusted navigation path to obtain supplementary data.

[0012] As a preferred solution of the underwater terrain survey method based on unmanned vessel data acquisition according to the present invention, wherein: the supplementary data is integrated with the three-dimensional terrain model to generate a terrain survey report, the specific steps are as follows: Fuse the supplementary mining point data into the initial terrain data of the 3D terrain model to generate complete gridded terrain data; Determine whether the grid cell corresponding to the supplementary mining point is empty; If the grid cell corresponding to the supplementary mining point is empty, the supplementary mining data is directly filled in; the filled supplementary mining point data is integrated into the initial terrain data of the 3D terrain model to generate complete gridded terrain data; If the grid cell corresponding to the supplementary mining point is not empty, the supplementary mining point data is directly fused into the initial terrain data of the three-dimensional terrain model to generate complete gridded terrain data; The three-dimensional terrain model is updated based on the gridded terrain data, and a terrain survey report is generated using the updated three-dimensional terrain model.

[0013] In a second aspect, the present invention provides an underwater terrain survey system based on unmanned vessel data collection, comprising an environment map generation module, a terrain change prediction module, a path planning module, a shadow area detection module, a shadow area supplementary sampling module and a report generation module; the environment map generation module is used to generate a dynamic environment map based on the collected multimodal environment data; the terrain change prediction module is used to dynamically perceive the underwater environment based on the dynamic environment map using a deep Q learning algorithm and predict potential terrain changes; the path planning module is used to determine whether to update the dynamic environment map according to the predicted potential terrain changes, and to plan a navigation path using the updated dynamic environment map based on the RRT algorithm; the shadow area detection module is used to collect terrain data based on the navigation path and generate a three-dimensional terrain model, and to detect shadow areas in the three-dimensional terrain model using U-Net; the shadow area supplementary sampling module is used to generate supplementary sampling points through geometric analysis based on the detected shadow areas, adjust the navigation path according to the supplementary sampling points, supplementary sample the shadow areas, and obtain supplementary sampling data; the report generation module is used to fuse the supplementary sampling data with the three-dimensional terrain model to generate a terrain survey report.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the underwater terrain survey method based on unmanned vessel data acquisition as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the underwater terrain survey method based on unmanned vessel data acquisition as described in the first aspect of the present invention is implemented.

[0016] The present invention improves the efficiency and accuracy of underwater terrain surveys through dynamic environmental perception and path planning, as well as detection and supplementary sampling of shadowed areas. A deep Q-learning algorithm is used to achieve dynamic environmental perception and path optimization, ensuring efficient obstacle avoidance and coverage of the target area by the unmanned vessel. Uncovered areas are detected using a U-Net, combined with geometric analysis to generate supplementary sampling points. The path is then adjusted for precise supplementary sampling, addressing model blind spots. Ultimately, a complete, high-precision three-dimensional terrain model is generated, overcoming the shortcomings of existing technologies in path planning and modeling integrity. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a flow chart of the underwater terrain survey method based on unmanned vessel data collection in Example 1.

[0019] Figure 2 Schematic diagram of the underwater terrain survey system based on unmanned vessel data acquisition in Example 1. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" 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 one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0023] Example 1, with reference to Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides an underwater terrain survey method based on unmanned vessel data collection, comprising the following steps: S1: Generate a dynamic environment map based on the collected multimodal environment data, specifically including: S1.1: Collect multimodal environmental data in real time through sensors; multimodal environmental data includes underwater terrain, obstacle distribution, and water flow velocity.

[0024] It should be noted that the above data provides basic support for generating dynamic environmental maps, planning safe navigation routes, and accurately surveying underwater terrain.

[0025] S1.2: Use Bayesian fusion to fuse multimodal environmental data, and combine it with GPS to spatially locate the multimodal environmental data, and generate the navigation trajectory of the unmanned ship based on the spatial positioning.

[0026] Furthermore, after preprocessing the multimodal environmental data collected by each sensor, extracting target features and calibrating data noise, the present invention uses the Bayesian formula to fuse this data. By setting prior probabilities (such as obstacle location estimates based on historical data) and likelihood functions (such as the probability distribution of sonar echo intensity and water flow disturbances), the confidence level of each modal data is quantified, and the posterior probability is calculated to achieve data fusion, thereby improving the accuracy and consistency of environmental data. Subsequently, combined with the precise latitude and longitude information provided by GPS, the fused multimodal data is mapped to a unified geographic coordinate system to ensure that the unmanned vessel can accurately locate itself and the surrounding environmental features, such as the exact location of underwater terrain and obstacles. Based on this high-precision spatial positioning data, the system generates an optimized and safe navigation trajectory, guiding the unmanned vessel to avoid potential dangers and efficiently complete tasks, such as bypassing newly confirmed obstacles or adjusting the path to adapt to changes in water flow.

[0027] It should be noted that when performing Bayesian fusion of multimodal environmental data, the data collected by each sensor is first preprocessed, including denoising, feature extraction, and time synchronization, to ensure data quality and consistency. Next, by setting prior probabilities (preliminary estimates of environmental states based on existing knowledge or historical data) and likelihood functions (describing the relationship between new observations and different environmental hypotheses), the sonar-detected obstacle locations are combined with information such as water flow disturbances. These probabilities are updated using the Bayesian formula to calculate posterior probabilities, thereby quantifying the confidence level of each modal data and achieving effective data fusion. For example, when determining the precise location of underwater obstacles, the present invention not only considers the potential obstacle location indicated by sonar echo intensity but also incorporates changes in the water velocity field to infer water flow disturbances that may affect the accuracy of sonar readings. Ultimately, Bayesian inference is used to produce a more reliable obstacle location result. This process significantly improves the accuracy and reliability of environmental perception, providing a solid foundation for subsequent path planning and obstacle avoidance.

[0028] S1.3: Generate a dynamic environment map using Delaunay based on the UAV’s navigation trajectory and multimodal environment data.

[0029] Furthermore, based on the unmanned vessel's real-time trajectory and multimodal environmental data, the present invention dynamically generates and continuously updates an environmental map using the Delaunay triangulation method, accurately reflecting features such as underwater terrain, obstacle distribution, and water velocity. As the unmanned vessel follows a planned or adaptively adjusted path, it continuously collects discrete point data using sensors (such as sonar, visual cameras, and water velocity meters). This data includes information such as terrain depth, obstacle boundaries, and water velocity. This multimodal data is combined with the unmanned vessel's current position (provided by GPS) to ensure that each measured point is accurately mapped to a geographic coordinate system. The Delaunay triangulation algorithm is then applied to transform these discrete points into an optimal, non-overlapping triangular mesh, constructing a continuous three-dimensional environmental model. As the unmanned vessel continues along its trajectory, new multimodal data is continuously collected and integrated into the existing map, enabling the environmental map to be updated in real time, dynamically reflecting changes in the current underwater environment. This method of combining the unmanned ship's navigation trajectory and multimodal environmental data not only improves the accuracy and consistency of environmental perception, but also provides the unmanned ship with reliable obstacle avoidance and path optimization support, ensuring that it can perform tasks safely and efficiently in complex environments.

[0030] S2: Based on the dynamic environment map, the deep Q-learning algorithm is used to dynamically perceive the underwater environment and predict the probability of potential terrain changes. Specifically, it includes: S2.1: Through feature extraction and discretization, the data features in the dynamic environment map are converted into the state space of the deep Q-learning algorithm, with the data features of the current local area of ​​the underwater environment as the states in the state space. Data features include terrain depth, obstacle location, water flow intensity, etc.

[0031] S2.2: Take the operation of the unmanned ship in the current environment as an action and define the action space.

[0032] Furthermore, the actions in the action space of the unmanned ship can include a series of operations related to navigation and mission execution, such as moving forward, backward, turning left, turning right, staying, and changing the navigation speed. These actions are refined according to the physical characteristics of the unmanned ship (such as turning radius, maximum speed, acceleration, etc.) and environmental constraints (such as water flow direction, obstacle location and terrain characteristics). For example, in a complex underwater environment, the action space can also be expanded to specific operations such as "bypassing obstacles" or "adjusting depth" to adapt to mission requirements. In addition, each action corresponds to different navigation results (such as path length, energy consumption, survey coverage, etc.) S2.3: Calculate the comprehensive reward value based on the distance between the unmanned vessel and the obstacle, the target area coverage ratio, and the navigation cost. The specific expression is:

[0033] in, Indicates that the unmanned ship is in state Next action The distance between the time and the obstacle, Indicates that the unmanned ship is in state Next action The coverage ratio of the target area surveyed at that time, Indicates that the unmanned ship is in state Next action The cost of sailing, It represents the comprehensive reward value when the unmanned ship performs action a in state s.

[0034] It should be explained that is to calculate the shortest distance between the unmanned ship and the obstacle (calculated by Euclidean distance), It is calculated by the grid coverage of the current path. It is calculated based on path length and energy consumption.

[0035] The present invention uses sensors to detect the distance between the unmanned boat and the obstacle in real time, and uses the Euclidean distance to calculate the closest distance. To evaluate obstacle avoidance safety.

[0036] The present invention calculates the number of grid cells covered by the current path of the unmanned ship and the total number of grid cells in the target area according to the ratio of the number of grid cells covered by the current path of the unmanned ship to the total number of grid cells in the target area. Next action Target area coverage ratio .

[0037] The present invention performs actions according to the unmanned ship The path length, speed and energy consumption model are used to calculate the unmanned ship in the state The navigation cost required to complete the action .

[0038] S2.4: Based on comprehensive reward value , combined with the divergence of the water velocity field, terrain depth, and obstacle density of the underwater environment, the probability of potential terrain change is predicted, and the expression is:

[0039] in, Indicates the coordinate position and time The probability of potential terrain changes under the forecast, Indicates the water flow at the coordinate position and time The degree of divergence under Indicates coordinate position and time The depth of the terrain, is the attenuation coefficient of terrain depth on water disturbance, Indicates the coordinate position and time The obstacle density under The weight coefficient that represents the impact of obstacles on terrain changes, Indicates coordinate position and time water flow rate.

[0040] It should be noted that the water flow is at the coordinate position and time The degree of divergence under , which is calculated by the divergence of the water velocity field, and the expression is:

[0041] in, The water velocity field is The direction component, The water velocity field is The direction component, is the partial derivative operator.

[0042] S3: Determine whether to update the dynamic environment map based on the predicted probability of potential terrain changes, and plan the navigation path using the updated dynamic environment map based on the RRT algorithm. Specifically, it includes: S3.1: Define a threshold terrain change risk threshold, C1, based on the historical frequency and magnitude of terrain changes.

[0043] S3.2: When When ≥C1, it is considered that the potential change probability of the terrain in the current area is high, and the dynamic environment map is updated.

[0044] S3.3: When When <C1, it is considered that the potential change probability of the terrain in the current area is low and there is no need to update the dynamic environment map.

[0045] It should be noted that the process of updating the dynamic environment map is to collect the terrain depth, water flow speed and obstacle data in the area in real time, and merge or replace the new data with the original map. When the risk threshold C1 is exceeded, the terrain depth is re-collected through sonar (for example, from 10 meters to 9.5 meters), the water flow sensor updates the water flow velocity (for example, from 1.2 meters / second to 1.5 meters / second), and new obstacles are detected (such as the position (x+10, y−5)). These data are replaced or weightedly fused into the corresponding grid to generate an updated dynamic environment map to reflect the latest underwater environment characteristics.

[0046] S3.4: Initialize the starting node of the navigation path and target node 's coordinates.

[0047] S3.5: Initialize RRT tree .

[0048] S3.6: Set rule constraints and generate sampling points based on the rule constraints in the updated dynamic environment map , and based on the current RRT tree Candidate nodes in , find the sampling point The nearest node , the expression is:

[0049] in, is the distance sampling point in the current RRT tree T The nearest node, is a candidate node in the tree T.

[0050] It should be noted that the specific constraints set by the rules are: Environmental information such as terrain depth, water velocity, and obstacle distribution is integrated to provide constraints for the unmanned vessel's navigation path planning. For example, if the map shows that the water velocity in a certain area exceeds 2 meters per second (strong current zone), this area is designated as "impassable"; if the terrain depth in an area is less than 3 meters (shallow water), it is marked as a "low priority" area; and areas with dense obstacles are designated as "obstacle avoidance zones," where the unmanned vessel's path must maintain a safe distance of at least 5 meters. These constraints will limit the generation of sampling points and guide the unmanned vessel to choose safe and low-cost paths for navigation and surveying.

[0051] S3.7: Connect the sampling points in the current RRT tree T The nearest node and sampling points , and through iterative expansion, generate To the target node navigation path.

[0052] Specifically: Calculate the sampling points according to the distance formula (such as Euclidean distance) The distance to all nodes in the tree, select the node with the smallest distance As the starting point of the connection. Then, from Towards Generate an incremental path (expanded at a fixed step size) in the direction, and check whether the path meets the rule constraints in the dynamic environment map (such as obstacle avoidance area, water flow influence, etc.). If the path is valid, add the newly expanded node to the tree T and repeat the above process, gradually expanding the scope of the tree until the path connects the starting node With the target node , and finally generate the optimal navigation path that meets all constraints.

[0053] S4: Collect terrain data based on the navigation path and generate a 3D terrain model. Use U-Net to detect shadow areas in the 3D terrain model. Specifically, it includes: S4.1: The unmanned vessel visits the path nodes in sequence along the planned navigation path, and at each path node, collects the initial underwater terrain data through the sensor equipment; the path node includes the starting node of the unmanned vessel on the navigation path , distance sampling point The nearest node and the target node .

[0054] S4.2: Use the Delaunay triangulation method to convert the initial terrain data into triangular mesh data, and import the triangular mesh data into OpenGL to generate a three-dimensional terrain model.

[0055] S4.3: Project the 3D terrain model into a 2D depth map and input the 2D depth map into a U-Net network. Use the encoder in the U-Net network to extract terrain features and use the decoder in the U-Net network to decode the terrain features. Generate a binary shadow mask with the same size as the 2D depth map. Identify and mark shadow areas in the 3D terrain model based on the binary shadow mask.

[0056] Specifically, the 3D terrain model is converted into a 2D depth map through orthographic or perspective projection. The resulting 2D image retains the terrain's depth information (such as undulations and elevation changes) and is passed as input to the U-Net network. The U-Net encoder extracts terrain features from the terrain map, such as the boundaries and shape characteristics of shadow areas, through multiple layers of convolution and pooling operations. Subsequently, the U-Net decoder gradually restores the image resolution through deconvolution and upsampling operations, and generates a binary shadow mask that matches the size of the 2D depth map. Based on the generated binary shadow mask, the identified shadow areas are remapped back to the 3D terrain model, accurately marking the shadow areas in the 3D terrain model for subsequent data re-collection or terrain analysis, where "1" indicates a shadow area and "0" indicates a non-shadow area.

[0057] S5: Generate supplementary sampling points based on the detected shadow area through geometric analysis, adjust the navigation path according to the supplementary sampling points, supplementary sample the shadow area, and obtain supplementary sampling data, which specifically includes: S5.1: Use geometric analysis methods to extract the boundary contours of the shadow area and form a boundary point set B.

[0058] For example, through morphological operations (such as edge detection or convex hull algorithm), the boundaries of the shadow area are extracted to ensure that the supplementary sampling points cover the edges of the shadow area.

[0059] S5.2: Based on the boundary point set B, the center coordinates of each regular grid are used as supplementary mining points to generate a supplementary mining point set V. The regular grid is obtained by discretizing the shadow area.

[0060] S5.3: Calculate the distance between each supplementary mining point in the supplementary mining point set V and the current position of the unmanned vessel, and insert the nearest supplementary mining points into the current navigation path in sequence, and adjust the current navigation path based on the rule constraints.

[0061] Furthermore, based on the set of supplementary sampling points V, the distance between each supplementary sampling point and the UAV's current position is first calculated (e.g., using the Euclidean distance formula), and the supplementary sampling points are sorted from closest to farthest. The closest supplementary sampling points are then inserted into the current navigation path, ensuring that the UAV prioritizes visiting closer supplementary sampling points, thereby optimizing navigation efficiency. When inserting supplementary sampling points, the path is adjusted to avoid high-risk areas, taking into account the constraints in the dynamic environment map (e.g., obstacle avoidance zones, shallow water areas, or strong currents).

[0062] For example, if the UAV is currently located at (x0, y0), and the set of supplementary collection points is {(x1, y1), (x2, y2), (x3, y3)}, and the distances are calculated and sorted as (x2, y2) < (x1, y1) < (x3, y3), the UAV will first navigate to (x2, y2). If the path from (x0, y0) to (x2, y2) passes through an obstacle area, a path planning algorithm (such as A* or RRT) will adjust the navigation route based on the dynamic environment map to bypass the obstacle and reach the supplementary collection point, ensuring safety and navigation efficiency.

[0063] S5.4: Conduct additional sampling of the shadowed area according to the adjusted navigation path to obtain additional sampling data.

[0064] It should be noted that supplementary sampling of shadowed areas based on the adjusted navigation path can effectively cover shadowed areas that were originally missed due to complex terrain or sensor blind spots. By optimizing the path, flight time and energy consumption are reduced while ensuring the integrity and accuracy of the supplementary data. This not only significantly improves the integrity and resolution of the 3D terrain model, but also provides more reliable data support for subsequent terrain analysis, environmental assessment, and navigation decision-making, avoiding potential risks or misjudgments caused by data gaps.

[0065] S6: Fusion of supplementary mining data with the 3D terrain model to generate a terrain survey report, including: S6.1: Fuse the supplementary mining point data into the initial terrain data of the three-dimensional terrain model to generate complete gridded terrain data.

[0066] S6.2: Determine whether the grid cell corresponding to the supplementary mining point is empty.

[0067] S6.2.1: If the grid cell corresponding to the supplementary mining point is empty, directly fill it with the supplementary mining data; merge the filled supplementary mining point data into the initial terrain data of the three-dimensional terrain model to generate complete gridded terrain data.

[0068] For example, when the unmanned vessel collects new depth data in a shadow area, if there is no data in the grid cell at that location, the newly collected data is directly filled into the empty cell.

[0069] S6.2.2: If the grid cell corresponding to the supplementary mining point is not empty, directly merge the supplementary mining point data into the initial terrain data of the three-dimensional terrain model to generate complete gridded terrain data.

[0070] For example, when old data already exist at the supplementary sampling point, the old and new data are averaged or fused according to a certain algorithm to update the data of the grid unit to ensure the accuracy of the terrain model.

[0071] S6.2.3: Update the three-dimensional terrain model based on the gridded terrain data and generate a terrain survey report using the updated three-dimensional terrain model.

[0072] Furthermore, after all supplemental data has been properly integrated into the 3D terrain model, complex algorithms are executed to recalculate and optimize terrain features, ensuring that each grid cell accurately reflects the latest survey results and performing detailed analysis of slope, depth changes, and potential geological structures. Based on the updated, high-precision 3D terrain model, a detailed underwater terrain survey report is automatically generated. This report not only summarizes the terrain changes, specifically noting the locations of new highlands, depressions, and obstacles that may affect navigation safety, but also provides comparative analysis with previous terrain data to help users understand terrain change trends and their potential impacts. It also includes a series of intuitive charts and images to assist professionals in in-depth research and decision-making.

[0073] This embodiment also provides an underwater terrain survey system based on unmanned vessel data collection, including: an environment map generation module, a terrain change prediction module, a path planning module, a shadow area detection module, a shadow area supplementary sampling module, and a report generation module; the environment map generation module is used to generate a dynamic environment map based on the collected multimodal environment data; the terrain change prediction module is used to dynamically perceive the underwater environment based on the dynamic environment map and predict potential terrain changes using a deep Q learning algorithm; the path planning module is used to determine whether to update the dynamic environment map based on the predicted potential terrain changes, and plan a navigation path using the updated dynamic environment map based on the RRT algorithm; the shadow area detection module is used to collect terrain data based on the navigation path and generate a three-dimensional terrain model, and use U-Net to detect shadow areas in the three-dimensional terrain model; the shadow area supplementary sampling module is used to generate supplementary sampling points through geometric analysis based on the detected shadow areas, adjust the navigation path according to the supplementary sampling points, supplementary sample the shadowed areas, and obtain supplementary sampling data; the report generation module is used to fuse the supplementary sampling data with the three-dimensional terrain model to generate a terrain survey report.

[0074] This embodiment also provides a computer device, which is suitable for the underwater terrain survey method based on unmanned ship data collection, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the underwater terrain survey method based on unmanned ship data collection proposed in the above embodiment.

[0075] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0076] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the underwater terrain survey method based on unmanned vessel data acquisition as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0077] In summary, this invention improves the efficiency and accuracy of underwater terrain surveys through dynamic environmental perception and path planning, as well as shadow area detection and supplementary sampling. It utilizes a deep Q-learning algorithm for dynamic environmental perception and path optimization, ensuring the unmanned vessel effectively avoids obstacles and covers the target area. Uncovered areas are detected using U-Net, combined with geometric analysis to generate supplementary sampling points. The path is adjusted for precise supplementary sampling, addressing model blind spots. Ultimately, a complete, high-precision three-dimensional terrain model is generated, overcoming the shortcomings of existing technologies in path planning and modeling integrity.

[0078] Example 2, referring to Table 1, is the second embodiment of the present invention. In order to further verify the technical solution of the present invention, experimental simulation data of an underwater terrain survey method based on unmanned vessel data collection is provided.

[0079] To demonstrate the advantages of underwater topographic survey methods based on data collection using unmanned vessels, an underwater topographic survey experiment was designed. The experimental area was selected within a 500m x 500m coastal waters with a water depth ranging from 3.2m to 14.8m. The area contained natural obstacles (such as reefs and shoals) and complex current disturbances. The unmanned vessel, equipped with multimodal sensors (multibeam sonar, current meter, and GPS), collected underwater environmental data. Using dynamic perception algorithms and path planning methods, the vessel completed topographic surveys and supplemented acquisitions in shadowed areas.

[0080] First, the unmanned vessel uses multimodal sensors (such as multi-beam sonar, current meters, and GPS) in the test area to collect real-time data on underwater terrain depth, water flow velocity, and obstacle distribution. The multimodal data is processed using a Bayesian fusion algorithm and combined with Delaunay triangulation to generate a preliminary environmental map. The environmental map is dynamically updated according to the unmanned vessel's navigation trajectory, providing a basis for subsequent terrain change prediction and path planning.

[0081] Secondly, based on the dynamic environment map, the deep Q learning algorithm is used to perceive the characteristic changes of the underwater environment and predict the probability of potential terrain changes. The system can also mark potential high-risk areas in real time. The RRT algorithm is combined with other algorithms to optimize the navigation path, ensuring that the unmanned vessel avoids high-risk areas while covering the target survey area as much as possible, thereby improving the safety and efficiency of path planning.

[0082] Next, the unmanned vessel visits each node along the planned route, collecting initial terrain data and generating a preliminary 3D terrain model through Delauna triangulation. This 3D terrain model is then projected into a 2D depth map, which is then fed into a U-Net network to detect shadowed areas. Geometric analysis generates supplementary sampling points, and the route is adjusted based on the dynamic environment map to supplementary sampled areas and update missing data.

[0083] Finally, the supplementary data was fused with the initial terrain data to generate complete gridded terrain data, and the 3D terrain model was updated. Key data indicators of the pre- and post-sampling models, including terrain depth accuracy, shadow area ratio, path coverage, navigation path length, and energy consumption, were analyzed. Ultimately, the post-sampling terrain model showed significant improvements in coverage, accuracy, and efficiency, validating the superiority of the proposed method.

[0084] In this experiment, existing technologies employed traditional underwater terrain survey methods, primarily using multi-beam sonar aboard unmanned vessels to collect underwater terrain data, sequentially covering target areas along a fixed navigation path. The resulting 3D terrain model relied solely on pre-set data collection paths and failed to dynamically perceive environmental changes. This led to data gaps and reduced accuracy in complex terrain and high-velocity areas (such as shallows and areas with dense obstacles). Furthermore, existing technologies lacked effective mechanisms for detecting and supplementing shadow areas, resulting in a high proportion of shadowed areas. This limited the integrity and accuracy of the 3D terrain model, making it difficult to meet the requirements for accurate surveying.

[0085] The details are shown in Table 1 below: Table 1 Comparison of underwater topographic survey test data Parameter name Existing technology The present invention Improvement rate (%) unit Terrain depth accuracy (average error) 3.87 1.23 68.2 rice Shadow area ratio 17.46 2.85 83.7 % Path coverage 74.31 96.78 30.2 % Length of flight path 5274.56 4928.34 -6.6 rice Total energy consumption 1215.43 1158.67 -4.7 kilowatt-hour Through the analysis of the data in the above table, it can be clearly seen that the present invention has significant advantages over the existing technology in terms of accuracy, coverage and efficiency of underwater terrain survey. The present invention uses innovative technologies such as dynamic environment perception, path optimization and shadow area supplementary sampling to achieve high-precision and high-coverage surveys of complex terrains, while reducing navigation costs and energy consumption. For example, in terms of terrain depth accuracy, the average error of the existing technology is 3.87 meters, while the present invention significantly reduces the error to 1.23 meters through precise shadow area supplementary sampling and dynamic path adjustment, and the accuracy is improved by 68.2%. The proportion of shadow areas has also been greatly reduced from 17.46% of the existing technology to 2.85%, and the coverage rate has been increased by 83.7%, indicating that the present invention can effectively identify and supplement blind spots that are difficult to cover with traditional methods. This is due to the precise shadow detection of the U-Net network and the supplementary sampling point design generated by geometric analysis.

[0086] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An underwater terrain survey method based on unmanned ship data collection, characterized in that: include, Generate dynamic environment maps based on collected multimodal environment data; Based on the dynamic environment map, the deep Q-learning algorithm is used to dynamically perceive the underwater environment and predict the probability of potential terrain changes; Determine whether to update the dynamic environment map based on the predicted probability of potential terrain changes, and plan the navigation path using the updated dynamic environment map based on the RRT algorithm; Collect terrain data based on the navigation path and generate a 3D terrain model, and use U-Net to detect shadow areas in the 3D terrain model; According to the detected shadow area, the supplementary mining points are generated through geometric analysis, and the navigation path is adjusted according to the supplementary mining points to supplement the shadow area and obtain supplementary mining data; The supplementary mining data is integrated with the three-dimensional terrain model to generate a terrain survey report.

2. The underwater topographic survey method based on unmanned ship data collection according to claim 1, characterized in that: The specific steps of generating a dynamic environment map based on the collected multimodal environment data are as follows: Collect multimodal environmental data in real time through sensors; multimodal environmental data includes underwater terrain, obstacle distribution and water flow speed; The multimodal environmental data are fused by using Bayesian fusion, and the multimodal environmental data are spatially positioned in combination with GPS, and the navigation trajectory of the unmanned ship is generated according to the spatial positioning; Based on the navigation trajectory of the unmanned ship and multimodal environmental data, a dynamic environmental map is generated through Delaunay.

3. The underwater topographic survey method based on unmanned ship data collection according to claim 2, characterized in that: Based on the dynamic environment map, the deep Q learning algorithm is used to dynamically perceive the underwater environment and predict the probability of potential terrain changes. The specific steps are as follows: Through feature extraction and discretization methods, the data features in the dynamic environment map are converted into the state space of the deep Q learning algorithm, and the data features of the local area of ​​the current underwater environment are used as the state in the state space; Take the operation of the unmanned ship in the current environment as an action and define the action space; The comprehensive reward value is calculated based on the distance between the unmanned ship and the obstacle, the coverage ratio of the target area and the navigation cost. ; Based on comprehensive reward value , combining the divergence of the water velocity field in the underwater environment, the depth of the terrain, and the density of obstacles to predict the probability of potential terrain changes .

4. The underwater topographic survey method based on unmanned ship data collection according to claim 3, characterized in that: The specific steps of determining whether to update the dynamic environment map according to the predicted probability of potential terrain changes, and planning the navigation path using the updated dynamic environment map based on the RRT algorithm are as follows: Based on the frequency and magnitude of historical terrain changes, the terrain change risk threshold C1 is defined; when When ≥C1, it is considered that the potential change probability of the terrain in the current area is high, and the dynamic environment map is updated; when When <C1, it is considered that the potential change probability of the current regional terrain is low, and there is no need to update the dynamic environment map; Initialize the starting node of the navigation path and the target node Coordinates Initialize the RRT tree ; Set rule constraints and generate sampling points based on the rule constraints in the updated dynamic environment map , and based on the candidate nodes in the current path , find the sampling point The closest node ; Connect the distance sampling points in the current RRT tree T The closest node and sampling points , and through iterative expansion, generate To the target node navigation path.

5. The underwater topographic survey method based on unmanned ship data collection according to claim 4, characterized in that: The steps of collecting terrain data based on the navigation path and generating a three-dimensional terrain model and detecting shadow areas in the three-dimensional terrain model using U-Net are as follows: The unmanned ship visits the path nodes in sequence along the planned navigation path, and collects the initial underwater terrain data through the sensor equipment at each path node; the path nodes include the starting node of the unmanned ship on the navigation path , distance sampling point The closest node and the target node ; Use Delaunay triangulation method to convert initial terrain data into triangular mesh data, and import the triangular mesh data into OpenGL to generate a three-dimensional terrain model; The 3D terrain model is projected into a 2D depth map, and the 2D depth map is input into the U-Net network. The terrain features are extracted by the encoder in the U-Net network, and the terrain features are decoded by the decoder in the U-Net network to generate a binary shadow mask with the same size as the 2D depth map. The shadow area in the 3D terrain model is identified and marked based on the binary shadow mask.

6. The underwater topographic survey method based on unmanned ship data collection according to claim 5, characterized in that: The method generates supplementary mining points through geometric analysis based on the detected shadow area, adjusts the navigation path according to the supplementary mining points, supplements the mining of the shadow area, and obtains supplementary mining data. The specific steps are as follows: Using geometric analysis methods, the boundary contour of the shadow area is extracted to form a boundary point set B; Discretize the shadow area and generate a regular grid covering the shadow area; Based on the boundary point set B, the center coordinates of each regular grid are used as supplementary mining points to generate a supplementary mining point set V; Calculate the distance between each supplementary mining point in the supplementary mining point set V and the current position of the unmanned ship, and insert the nearest supplementary mining points into the current navigation path in sequence, and adjust the current navigation path based on the rule constraints; The shadow area is supplemented with samples according to the adjusted navigation path to obtain supplementary sampling data.

7. The underwater topographic survey method based on unmanned ship data collection according to claim 6, characterized in that: The above-mentioned steps of fusing the supplementary mining data with the three-dimensional terrain model to generate a terrain survey report are as follows: The supplementary mining point data is integrated into the initial terrain data of the 3D terrain model to generate complete gridded terrain data; Determine whether the grid cell corresponding to the supplementary mining point is empty; If the grid cell corresponding to the supplementary mining point is empty, the supplementary mining data is directly filled in; the filled supplementary mining point data is merged into the initial terrain data of the three-dimensional terrain model to generate complete gridded terrain data; If the grid cell corresponding to the supplementary mining point is not empty, the supplementary mining point data is directly fused into the initial terrain data of the three-dimensional terrain model to generate complete gridded terrain data; The three-dimensional terrain model is updated based on the gridded terrain data, and a terrain survey report is generated using the updated three-dimensional terrain model.

8. An underwater terrain survey system based on unmanned ship data collection, based on the underwater terrain survey method based on unmanned ship data collection according to any one of claims 1 to 7, characterized in that: It includes an environment map generation module, a terrain change prediction module, a path planning module, a shadow area detection module, a shadow area supplementary sampling module and a report generation module; Environmental map generation module: used to generate dynamic environmental maps based on the collected multimodal environmental data; Terrain change prediction module: used to dynamically perceive the underwater environment based on the dynamic environment map and predict potential terrain changes using the deep Q learning algorithm; Path planning module: used to determine whether to update the dynamic environment map based on the predicted potential terrain changes, and plan the navigation path using the updated dynamic environment map based on the RRT algorithm; Shadow area detection module: used to collect terrain data based on the navigation path and generate a 3D terrain model, and use U-Net to detect shadow areas in the 3D terrain model; Shadow area supplementary mining module: It is used to generate supplementary mining points through geometric analysis according to the detected shadow area, adjust the navigation path according to the supplementary mining points, supplementary mine the shadow area, and obtain supplementary mining data; Report generation module: used to integrate the supplementary mining data with the three-dimensional terrain model to generate a terrain survey report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the underwater terrain survey method based on unmanned ship data collection according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the underwater terrain survey method based on unmanned ship data collection according to any one of claims 1 to 7 are implemented.

Images