Unmanned ship and cable underwater robot cooperative seabed target search path planning method and system based on synthetic aperture sonar

Abstract

The invention discloses a synthetic aperture sonar-based unmanned ship and cable underwater robot cooperative seabed target search path planning method and system, and relates to the technical field of ocean engineering, and the method comprises the steps: constructing a digital twin platform supporting target search; generating a first path in the digital twin platform based on the search task instruction, and controlling the unmanned ship carrying the synthetic aperture sonar to sail along the first path; when the unmanned ship recognizes a suspicious target area in the sailing process, a second path is generated based on the anchoring position of the unmanned ship, and the cable underwater robot is controlled to dive along the second path; when the underwater robot dives to the target depth, the underwater robot with the cable is controlled to carry out approaching fine exploration and confirmation on the suspicious target area; according to a preset task completion standard, whether the seabed target search task is completed is judged, and if the seabed target search task is completed, fusion recording and labeling are carried out on the multi-source data obtained in the search process, so that the efficient, safe and intelligent seabed target search task is realized.

Description

Technical Field

[0001] This invention relates to the field of marine engineering technology, and in particular to a method and system for collaborative seabed target search path planning between unmanned surface vessels and tethered underwater robots based on synthetic aperture sonar. Background Technology

[0002] The ocean covers the vast majority of the Earth's surface and contains abundant resources and important infrastructure. High-precision and high-efficiency exploration of the seabed is of vital importance in fields such as marine resource exploration, marine scientific research, subsea pipeline inspection, and environmental monitoring.

[0003] In recent years, the collaborative exploration model based on "unmanned surface vessels (USVs) as mobile platforms" and "tethered underwater robots (ROVs) as underwater operation units" has become a more cost-effective and operationally flexible solution. However, existing USV-ROV collaborative systems still face multiple technical bottlenecks in complex marine environments: the dynamic management of umbilical cables lacks accurate modeling of the coupling relationship between water flow disturbances, cable deformation, and ROV motion, which can easily lead to excessive towing force, entanglement, or even breakage; underwater communication is susceptible to multipath effects and noise interference, and once interrupted, the ROV loses control, lacking an effective local autonomous emergency mechanism; at the same time, shore-based control systems mostly adopt static task planning, making it difficult to integrate real-time environmental perception and equipment status for online optimization, resulting in rigid exploration paths, weak obstacle avoidance capabilities, and low energy efficiency. Summary of the Invention

[0004] This invention provides a method and system for collaborative seabed target search path planning based on synthetic aperture sonar between unmanned surface vessels and tethered underwater robots, in order to solve the problems of high operating costs, high personnel risks, unstable communication, extensive cable management, rigid path planning and weak response to sudden failures in the complex environment of deep sea exploration systems.

[0005] In a first aspect, embodiments of the present invention provide a method for collaborative seabed target search path planning between an unmanned surface vessel and a tethered underwater robot based on synthetic aperture sonar, comprising: A hierarchical digital twin platform supporting target search is constructed and initialized. The platform includes an L1 prior information layer, an L2 dynamic environment layer, and an L3 real-time detection and collaboration layer. Based on the search mission instructions and L2 dynamic environment layer data, an adaptive path planning strategy is adopted to generate a first path in the digital twin platform, and the unmanned vessel equipped with synthetic aperture sonar is controlled to navigate along the first path and perform real-time imaging. When the unmanned vessel identifies a suspicious target area during navigation, it is controlled to anchor and release a tethered underwater robot. Based on the anchoring position of the unmanned vessel and the suspicious target area, an adaptive cable-tube cooperative control strategy is used to generate a second path, and the tethered underwater robot is controlled to dive along the second path. When the tethered underwater robot dives to the target depth, based on the multi-source sensor data acquired by the tethered underwater robot, the adaptive path planning strategy is used to control the tethered underwater robot to conduct close-range detailed exploration and confirmation of the suspicious target area. Based on the exploration and confirmation results and the preset task completion criteria, it is determined whether the seabed target search task has been completed. If completed, the multi-source data obtained during the search process is fused, recorded, and labeled.

[0006] Optionally, the construction and initialization of a hierarchical digital twin platform supporting target search includes an L1 prior information layer, an L2 dynamic environment layer, and an L3 real-time detection and collaboration layer, comprising: Collect and fuse multibeam topographic data, side-scan sonar historical data and known target information of the target sea area to construct the L1 prior information layer; By acquiring real-time and forecasted marine environmental data, an L2 dynamic environmental layer containing water acoustic characteristic parameters is constructed. Based on the detection model of unmanned ships and their onboard synthetic aperture sonar, and the detection model of tethered underwater robots and their sensors, an L3 real-time detection and collaboration layer is constructed. Based on the L1 prior information layer, L2 dynamic environment layer, and L3 real-time detection and collaboration layer, the hierarchical digital twin platform is constructed, and the platform is initialized, loading the target feature library and search strategy.

[0007] Optionally, after constructing and initializing the hierarchical digital twin platform that supports target search, the method further includes: In the simulation environment of the digital twin platform, training scenarios are defined that include different seabed topography, target types, and environmental disturbances; Based on the imaging principle and target recognition algorithm of synthetic aperture sonar, virtual training and testing were conducted on the sonar image processing model and the target recognition model.

[0008] Optionally, the step of generating a first path in the digital twin platform based on search task instructions and L2 dynamic environment layer data using an adaptive path planning strategy, and controlling the unmanned surface vessel equipped with synthetic aperture sonar to navigate along the first path and perform real-time imaging, includes: Based on the range of the target area, the swath width and resolution parameters of the synthetic aperture sonar in the search task instruction, and combined with the water acoustic data of the L2 dynamic environmental layer, an adaptive path planning strategy is adopted to generate a first path in the digital twin platform. The unmanned vessel is controlled to navigate along the first path, and the synthetic aperture sonar data transmitted back is processed in real time to generate an underwater acoustic image. The acoustic image is analyzed in real time based on a target recognition algorithm, and suspicious target areas are automatically marked.

[0009] Optionally, when the unmanned vessel identifies a suspicious target area during navigation, it controls the unmanned vessel to anchor and release the tethered underwater robot. Based on the anchoring position of the unmanned vessel and the suspicious target area, an adaptive cable-and-tube cooperative control strategy is used to generate a second path, and the tethered underwater robot is controlled to dive along the second path, including: When an unmanned surface vessel (USV) identifies a suspicious target area during navigation, the USV's edge computing nodes combine its own positioning information with the target position in the synthetic aperture sonar image to calculate the coordinates of the suspicious target. Based on the geographic coordinates of the suspected target, the anchoring position of the unmanned vessel is determined; Based on the anchoring position of the unmanned vessel and the geographical coordinates of the suspected target, an adaptive cable-tube cooperative control strategy is adopted to generate a second path; A release command is issued to release the tethered underwater robot, and the tethered underwater robot is controlled to descend along the second path to approach the coordinates of the suspected target.

[0010] Optionally, when the tethered underwater robot dives to the target depth, based on the multi-source sensor data acquired by the tethered underwater robot, the adaptive path planning strategy is used to control the tethered underwater robot to conduct close-range detailed exploration and confirmation of the suspicious target area, including: When the underwater robot descends to the target depth, it acquires multi-source sensor data based on the optical camera, high-resolution profile sonar, and multi-parameter sensors mounted on the tethered underwater robot. Based on the multi-source sensor data, an adaptive path planning strategy is used to generate a third path, and the tethered underwater robot is controlled to conduct close-range, detailed exploration and confirmation of the suspicious target along the third path.

[0011] Optionally, based on the exploration and confirmation results and preset task completion criteria, it is determined whether the seabed target search task is completed. If completed, the multi-source data acquired during the search process is fused, recorded, and labeled, including: Based on the exploration and confirmation results and the preset task completion criteria, it is determined whether the seabed target search task has been completed. If the task is completed, the global scanning sonar images, close-range exploration optical and acoustic data, target coordinates and attribute information are stored in association. Based on the full-process data of this mission, the sonar image processing model, target recognition model, and adaptive path planning strategy in the digital twin platform were optimized and updated.

[0012] Optional, also includes: If the communication between the unmanned vessel and the shore-based control center is interrupted, the edge node of the unmanned vessel will autonomously decide on the subsequent scanning path or the order of exploration of the discovered suspicious targets based on the acquired scanning data, the preset search priority and the digital twin simulation. If communication between the tethered underwater robot and the unmanned vessel is interrupted, the underwater robot will execute a preset emergency exploration action or a safe surfacing procedure based on the last command and the local AI module.

[0013] Secondly, embodiments of the present invention provide a collaborative seabed target search path planning system based on synthetic aperture sonar for unmanned surface vessels and tethered underwater robots. The system is used to execute the collaborative seabed target search path planning method based on synthetic aperture sonar for unmanned surface vessels and tethered underwater robots as described in any one of the embodiments of the present invention, including: The building module is used to build and initialize a hierarchical digital twin platform that supports target search. The platform includes an L1 prior information layer, an L2 dynamic environment layer, and an L3 real-time detection and collaboration layer. The first path generation module is used to generate a first path in the digital twin platform based on the search task instructions and L2 dynamic environment layer data, using an adaptive path planning strategy, and control the unmanned vessel equipped with synthetic aperture sonar to navigate along the first path and perform real-time imaging. The second path generation module is used to control the unmanned vessel to anchor and release the tethered underwater robot when the unmanned vessel identifies a suspicious target area during navigation. Based on the anchoring position of the unmanned vessel and the suspicious target area, an adaptive cable-tube cooperative control strategy is used to generate a second path and control the tethered underwater robot to dive along the second path. The exploration module is used to control the tethered underwater robot to conduct close-range, detailed exploration and confirmation of the suspicious target area based on the multi-source sensor data acquired by the tethered underwater robot and the adaptive path planning strategy when the tethered underwater robot dives to the target depth. The judgment module is used to determine whether the seabed target search task is completed based on the exploration and confirmation results and the preset task completion standards. If completed, the multi-source data obtained during the search process is fused, recorded and labeled.

[0014] Optionally, the building module is specifically used for: Collect and fuse multibeam topographic data, side-scan sonar historical data and known target information of the target sea area to construct the L1 prior information layer; By acquiring real-time and forecasted marine environmental data, an L2 dynamic environmental layer containing water acoustic characteristic parameters is constructed. Based on the detection model of unmanned ships and their onboard synthetic aperture sonar, and the detection model of tethered underwater robots and their sensors, an L3 real-time detection and collaboration layer is constructed. Based on the L1 prior information layer, L2 dynamic environment layer, and L3 real-time detection and collaboration layer, the hierarchical digital twin platform is constructed, and the platform is initialized, loading the target feature library and search strategy.

[0015] In this embodiment, a hierarchical digital twin platform provides a high-fidelity simulation and decision-making basis for the overall task, driving the unmanned vessel to perform large-scale intelligent scanning and real-time target identification along a dynamically optimized first path. After detecting a suspicious target, it seamlessly switches to a second path generated by an adaptive cable-tube collaborative strategy, guiding the tethered underwater robot to dive safely and accurately. Then, based on multi-source perception data, a third path is generated to perform close-range detailed exploration and confirmation. Finally, the task is completed through intelligent judgment and the entire process data is integrated and recorded, thereby significantly improving the efficiency, accuracy, environmental adaptability and system intelligence level of the overall operation.

[0016] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 The flowchart shows a method for collaborative seabed target search path planning based on synthetic aperture sonar between an unmanned surface vessel and a tethered underwater robot, as provided in Embodiment 1 of the present invention. Figure 2 This is a framework diagram of a collaborative seabed target search path planning system based on synthetic aperture sonar for unmanned surface vessels and tethered underwater robots, provided in Embodiment 3 of the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0020] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] Application Overview Existing USV-ROV collaborative systems suffer from three main drawbacks. First, the lack of precise dynamic modeling and collaborative control of the umbilical cable connecting the USV and ROV makes it prone to excessive drag in complex currents, leading to a high risk of cable entanglement or breakage. Second, the system heavily relies on continuous and stable communication; if communication between the USV and shore-based facilities or between the USV and the ROV is interrupted, the ROV becomes uncontrollable, lacking local autonomous emergency response capabilities and resulting in poor operational safety and continuity. Finally, mission planning is largely based on static presets, failing to incorporate real-time marine environmental data (such as sound velocity profiles and currents) and equipment status for dynamic optimization. This leads to inefficient exploration paths, weak obstacle avoidance capabilities, and an inability to learn and improve during missions.

[0022] This invention constructs a layered digital twin platform integrating a priori information layer (L1), a dynamic environment layer (L2), and a real-time detection collaboration layer (L3), providing a high-fidelity simulation, planning, and decision-making environment for the entire search mission. Based on this, an adaptive cable-tube collaborative control strategy is employed to optimize the ROV's diving path, significantly reducing the risk of cable stress. Through a cloud-edge collaborative decision-making mechanism, the USV edge node and the ROV's local AI can autonomously take over the task when communication is interrupted, ensuring operational continuity. Furthermore, virtual training and post-mission data closure are utilized to continuously optimize sonar imaging, target recognition, and path planning models. Ultimately, this solution achieves a transformation from static planning to dynamic optimization, and from reliance on manual intervention to high autonomy, significantly improving the operational efficiency, environmental adaptability, and system safety of seabed target search.

[0023] Example 1: Figure 1 This is a flowchart of a method for collaborative seabed target search path planning using an unmanned surface vessel (USV) and a tethered underwater robot based on synthetic aperture sonar, provided in Embodiment 1 of the present invention. This embodiment is applicable to seabed target search scenarios. The method is executed by a collaborative seabed target search path planning system based on synthetic aperture sonar. Figure 1 As shown, the method includes: S110. Construct a hierarchical digital twin platform that supports target search and initialize it. The platform includes an L1 prior information layer, an L2 dynamic environment layer, and an L3 real-time detection and collaboration layer. S120. Based on the search task instructions and L2 dynamic environment layer data, an adaptive path planning strategy is adopted to generate a first path in the digital twin platform, and the unmanned vessel equipped with synthetic aperture sonar is controlled to navigate along the first path and perform real-time imaging. S130. When the unmanned vessel identifies a suspicious target area during navigation, the unmanned vessel is controlled to anchor and release the tethered underwater robot. Based on the anchoring position of the unmanned vessel and the suspicious target area, an adaptive cable-tube cooperative control strategy is adopted to generate a second path, and the tethered underwater robot is controlled to dive along the second path. S140. When the tethered underwater robot dives to the target depth, based on the multi-source sensor data acquired by the tethered underwater robot, the adaptive path planning strategy is adopted to control the tethered underwater robot to conduct close-range detailed exploration and confirmation of the suspicious target area. S150. Based on the exploration and confirmation results and the preset task completion criteria, determine whether the seabed target search task is completed. If completed, fuse, record and label the multi-source data obtained during the search process.

[0024] In this embodiment, a hierarchical digital twin platform provides a high-fidelity simulation and decision-making basis for the overall task, driving the unmanned vessel to perform large-scale intelligent scanning and real-time target identification along a dynamically optimized first path. After detecting a suspicious target, it seamlessly switches to a second path generated by an adaptive cable-tube collaborative strategy, guiding the tethered underwater robot to dive safely and accurately. Then, based on multi-source perception data, a third path is generated to perform close-range detailed exploration and confirmation. Finally, the task is completed through intelligent judgment and the entire process data is integrated and recorded, thereby significantly improving the efficiency, accuracy, environmental adaptability and system intelligence level of the overall operation.

[0025] Example 2: The technical solution in this example is a further refinement based on the above examples.

[0026] In step S110, the construction and initialization of the hierarchical digital twin platform supporting target search includes: Collect and fuse multibeam topographic data, side-scan sonar historical data and known target information of the target sea area to construct the L1 prior information layer; The target sea area refers to a specific ocean region designated for seabed searching in a target search mission. Multibeam topographic data refers to data acquired using a shipborne multibeam echo sounder, capable of generating high-precision, high-coverage seabed topography (digital elevation model), which is fundamental to understanding seabed topography. Side-scan sonar historical data refers to acoustic image data acquired by side-scan sonar systems in previous missions. It clearly reflects the texture and undulations of the seabed surface, crucial for identifying the outlines and shapes of targets (such as shipwrecks and reefs). Known target information refers to pre-existing information about targets within the target sea area, such as the coordinates of known shipwrecks, pipeline routes, and the locations of obstacles.

[0027] The L1 prior information layer refers to the first layer of the digital twin platform. It integrates and solidifies all relatively static, non-changing prior knowledge of the seabed environment, primarily providing a basic geographic information framework for subsequent mission planning and simulation. Specifically, the L1 static layer is constructed as a digital elevation model (DEM). It can retrieve raw multibeam bathymetry data, raw side-scan sonar image data, and nautical charts or databases labeled with known target locations and attributes from national marine databases, commercial data service providers, and historical mission archives. Data processing is performed on the collected raw data, including: data cleaning to remove outliers and noise; coordinate unification by converting all data to a unified coordinate system (such as WGS-84) and vertical datum; and meshing by using interpolation algorithms to generate a high-resolution, regularly meshed DEM, i.e., a seabed topographic map, from the multibeam discrete point cloud data. The processed side-scan sonar images are then registered and overlaid with the DEM to enrich seabed texture information. Known objects (such as shipwrecks or pipelines) are processed into solid models with 3D shape, location, and attributes (type, height), and then "placed" in the correct position on the DEM. The processed DEM, texture maps, and 3D object models are then imported into a 3D engine (such as Unity, Unreal Engine, or Cesium) to create an interactive and queryable 3D static underwater scene. This scene is the L1 prior information layer, which serves as the "static map" base for the entire digital twin world.

[0028] By acquiring real-time and forecasted marine environmental data, an L2 dynamic environmental layer containing water acoustic characteristic parameters is constructed. Real-time and forecasted marine environmental data can refer to both currently measured (real-time) and future predicted (forecasted) marine parameters, such as three-dimensional current velocity, current direction, temperature, salinity, wind speed, and waves. Water acoustic properties parameters refer to parameters describing the propagation characteristics of sound waves in seawater, such as sound velocity profiles and absorption coefficients. These parameters directly affect the detection performance and imaging quality of sonar equipment.

[0029] The L2 dynamic environment layer is the second layer of the digital twin platform, integrating and simulating dynamically changing marine environmental fields. It provides environmental context for mission simulation and real-time control, and is crucial for evaluating operational windows, optimizing paths, and predicting equipment stresses.

[0030] Specifically, forecast data for a target sea area over a future period (e.g., 72 hours) can be obtained via API from regional ocean forecasting centers (such as ROMS and HYCOM models), including current velocity, direction, temperature, and salinity at different depths. Simultaneously, real-time observation data from ocean buoys and satellites is received for correction. A regional ocean numerical model is deployed in the cloud or on a high-performance computing node, simulating seawater movement based on physical equations. Data assimilation algorithms, such as ensemble Kalman filtering (EnKF), are run. The acquired real-time observation data is "injected" into the ocean model, dynamically correcting the model's initial state and parameters to make the model's forecast output closer to reality. The assimilated model outputs a high spatiotemporal resolution three-dimensional ocean environmental field covering the target area for the future period. Based on the assimilated temperature, salinity, and depth data, empirical formulas (such as the Mackenzie formula) are used to calculate the sound velocity profile of the entire water body, thus obtaining the acoustic characteristics parameters of the water body. The calculated three-dimensional dynamic ocean environment field (current velocity, temperature, sound speed, etc.) is used as a time-varying attribute field and correlated and superimposed with the static topography of layer L1. In the digital twin platform, layer L2 is represented as a dynamic environmental data layer that "flows" on top of the static topography of layer L1 and can be queried and predicted.

[0031] Based on the unmanned vessel and its onboard synthetic aperture sonar detection model, and the tethered underwater robot and its sensor detection model, an L3 real-time detection and collaboration layer is constructed. The unmanned surface vessel (USV) and its onboard synthetic aperture sonar (SAS) detection model refers to a mathematical model that accurately simulates the motion of the USV and the operation of its SAS within a digital twin. This includes the USV's dynamic model and the SAS's acoustic imaging algorithm model, used to predict the USV's navigation status and the SAS's detection effectiveness. The tethered underwater vehicle (ROV) and its sensor detection model refers to a mathematical model that accurately simulates the ROV's motion and the operation of its onboard sensors (such as optical cameras, robotic arms, and profiling sonar) within a digital twin. This includes the ROV's hydrodynamic model, the cable's dynamic model, and the measurement models of each sensor.

[0032] The L3 real-time detection and collaboration layer is the third layer of the digital twin platform, focusing on millisecond-level high-fidelity synchronous simulation and prediction of USV, ROV entities and their interactions. It is the core for realizing real-time monitoring, predictive control, and virtual training.

[0033] Specifically, a six-DOF rigid body dynamics model can be established for the USV, including its mass, inertia, center of buoyancy, center of gravity, and thruster model. A six-DOF hydrodynamic model can be established for the ROV, and a discrete mass model or finite element model can be established for its umbilical cable to simulate the cable's deformation and tension in the water flow. An acoustic imaging signal processing model can be established for the SAS to simulate the generation of sonar images under a given motion trajectory and marine environment. Measurement models can be established for the ROV's optical cameras, sonar, and other sensors to simulate their observation data (such as image noise and field of view). Due to the enormous computational demands of the above high-fidelity models, they can be simplified to meet real-time control requirements. This can be achieved by using mathematical methods such as intrinsic orthogonal decomposition to extract the core dynamic features of the original complex model and generate a computationally faster reduced-order surrogate model. Additionally, a large amount of historical or simulation data can be used to train neural network models to approximate the dynamic response of the original system.

[0034] The simplified real-time simulation models of USVs, ROVs, cables, and sensors are integrated, enabling them to receive real-time sensor data (such as GPS, IMU, depth, and camera footage) from real devices via a data interface to update their simulation state and maintain synchronization with the physical world. This virtual USV / ROV system, capable of millisecond-level simulation and bidirectional synchronization with the physical world, constitutes the L3 real-time detection and collaboration layer.

[0035] Based on the L1 prior information layer, L2 dynamic environment layer, and L3 real-time detection and collaboration layer, the hierarchical digital twin platform is constructed, and the platform is initialized, loading the target feature library and search strategy.

[0036] The target feature library can refer to a database that stores acoustic, optical, and geometric feature data of various targets to be searched (such as shipwrecks of different types, pipe joints, and mineral samples). It is used to train and drive AI target recognition algorithms. The search strategy can refer to a series of rules, algorithms, and decision-making logics formulated to complete a specific search task, such as path planning algorithms, suspicious target identification rules, and the collaborative working mechanism between USVs and ROVs.

[0037] Specifically, this layered digital twin platform adopts a typical cloud-edge architecture. The cloud (shore-based center) is responsible for storing and running massive amounts of L1 and L2 layer data and models, performing large-scale simulations and AI model training. The edge (USV onboard computer) deploys a lightweight L3 real-time layer model, responsible for real-time synchronization, rapid simulation, and local control. Middleware such as ROS and DDS are used to achieve high-speed, reliable data communication between the three layers and between the platform and real USV / ROV equipment.

[0038] The initialization process specifically includes: selecting a target sea area for the user at the shore-based center. The system loads L1 layer scene data for that area from the database, calls the L2 layer model to generate an environmental forecast for the area, and retrieves the corresponding USV / ROV L3 layer simulation model from the model library. These three are packaged into a "task package." The task package is distributed to the edge computing nodes of the USV via the satellite network. The edge nodes start the L3 layer program, decompress the task package, and run a local digital twin sandbox consistent with the cloud environment locally. The USV's sensors begin to work, measuring environmental data (such as flow velocity). The edge nodes run a data assimilation algorithm, dynamically correcting the local environmental model in the L3 layer using the measured data, transitioning it from a "forecast-based" state to a "close to real-world measurement" state, completing the "warm-up." Simultaneously with initialization, the target feature library (for AI recognition) and the search strategy (such as path planning algorithms and collaborative rules) formulated for this task are loaded into the corresponding modules of the platform (such as the AI ​​recognition module and task planning module), enabling the platform to possess the "knowledge" and "methods" to perform specific search tasks.

[0039] In this embodiment, by constructing a hierarchical digital twin platform comprising an L1 prior information layer, an L2 dynamic environment layer, and an L3 real-time detection and collaboration layer, high-fidelity full-scale modeling of static seabed geography, dynamic marine environment, and equipment dynamics is achieved. This provides a unified and accurate virtual environment foundation for mission planning, real-time simulation, and collaborative control, significantly improving the system's environmental perception and predictive decision-making capabilities.

[0040] Optionally, after building and initializing the layered digital twin platform, the following may also be included: In the simulation environment of the digital twin platform, training scenarios are defined that include different seabed topography, target types, and environmental disturbances; The simulation environment of the digital twin platform refers to a fully virtualized marine operating space created based on a pre-constructed hierarchical digital twin platform. It integrates the static terrain of layer L1, the dynamic marine environment of layer L2, and the equipment dynamics model of layer L3, enabling high-fidelity simulation of real-world physical laws and sensor feedback; it serves as a sandbox for "virtual training." Training scenarios refer to a series of virtual task situations specifically designed for training AI models within the simulation environment. Each scenario defines specific initial conditions, environmental parameters, and task objectives, acting as the "test questions" for model learning. Seabed topography refers to the virtual seabed features set in the training scenarios, such as flat seabeds, seamounts, canyons, and slopes, used to train the system's path planning and obstacle avoidance capabilities under different geographical conditions. Target types refer to the categories of various underwater objects to be searched, such as shipwrecks, pipelines, containers, mineral nodules, and specific biological communities. In the training scenarios, the characteristic representations of these targets in acoustic / optical images need to be defined. Environmental disturbances can refer to dynamic interference factors that exist in the real ocean and pose challenges to operations. These mainly include water flow (ocean currents of different speeds, directions, and profiles), visibility (simulating the effect of turbid water on optics), and simulated communication delays or interruptions.

[0041] Specifically, an initialized hierarchical digital twin platform is utilized. Diverse seabed mesh models (such as flat areas, slopes, reef areas, and pipelines / ditches) are selected or programmatically generated from the terrain library of the L1 prior information layer. 3D models and their acoustic / optical properties of various targets are retrieved from the target feature library. The scene is parameterized, and the selected terrain is "arranged" in the simulation engine, potentially with random perturbations to generate subtle variations in terrain height. 3D models of different target types are randomly or systematically "placed" on the scene terrain. Different target attitudes (such as a capsized shipwreck) and partially buried states can be set to increase the difficulty of identification. Environmental disturbances are injected by extracting or randomly generating 3D flow field data of different intensities and directions from the L2 dynamic environment layer model and applying it to the simulation environment to affect the dynamics and cable morphology of USVs / ROVs. Water optical parameters are adjusted to simulate the visual effects of camera images under different water qualities, such as clear and turbid water. Different signal-to-noise ratio speckle noise and multipath echo interference are injected into the generated sonar images.

[0042] By using script programming, the above-mentioned elements such as terrain, target, and disturbance are combined and randomized to automatically generate thousands to tens of thousands of training scenarios, ensuring scenario diversity and preventing model overfitting.

[0043] Based on the imaging principle and target recognition algorithm of synthetic aperture sonar, virtual training and testing were conducted on the sonar image processing model and the target recognition model.

[0044] The imaging principle of synthetic aperture sonar describes how sonar, through the movement of a platform (such as a USV), coherently processes multiple pulse echoes to "synthesize" a larger virtual aperture, ultimately obtaining a seabed image with a resolution far exceeding that of the actual physical aperture. Target recognition algorithms can refer to computer programs or mathematical models used to automatically detect and classify targets of interest in acoustic or optical images, typically deep learning-based neural networks such as convolutional neural networks (CNNs). It is the core object being trained. Sonar image processing models refer to algorithmic modules that perform a series of operations on raw synthetic aperture sonar echo data, including preprocessing, imaging, denoising, and enhancement. In virtual training, this model may be integrated into the simulation pipeline to generate high-quality simulated sonar images, or its parameters may themselves be objects that need optimization. Target recognition models specifically refer to AI models that, after training, can automatically output the target location, category, and confidence level from processed sonar images; this is the core output of virtual training.

[0045] Specifically, within a predefined training scenario, a virtual USV is controlled to navigate along a planned or random path. Based on the imaging principles of synthetic aperture sonar, the simulation engine not only calculates the USV's trajectory but also simulates and calculates the echo signal for each pulse based on acoustic propagation models, seabed material scattering models, and injected noise. These simulated echo signals are then input into a sonar image processing model (or a standard imaging algorithm) to synthesize high-fidelity simulated sonar images. Because the scene is fully controlled, each simulated image carries precise pixel-level annotations (i.e., whether each pixel belongs to a target and, if so, what type).

[0046] Model training can use a massive dataset of generated, labeled simulated sonar images as a training set, inputting it into a target recognition algorithm (such as a deep learning network to be trained). The algorithm continuously compares its predictions with the actual labels (calculating the loss function), backpropagates to adjust the network's internal parameters, and ultimately learns to accurately identify targets from complex sonar images. This process can also be combined with reinforcement learning, allowing the agent (AI model) to learn how to optimize its search strategy through interaction with the simulation environment. Similarly, image processing model optimization can utilize simulated data to optimize parameters in the sonar image processing model (such as filtering parameters and gain compensation parameters) to make the final generated images more conducive to target recognition.

[0047] Model testing will create a completely new set of simulated test scenarios that were never used during training. The trained target recognition model will be run on this test set, with simulated sonar images as input. The system will automatically evaluate the model's performance metrics on the test set, such as target detection accuracy, recall, false alarm rate, and robustness under different terrain and disturbance conditions. Only models that pass the test evaluation will be deployed to real-world USV-ROV collaborative systems.

[0048] In this embodiment, by defining diverse training scenarios in the simulation environment of the digital twin platform, and by conducting virtual training and testing of the image processing and target recognition models based on the principle of synthetic aperture sonar imaging, large-scale, low-cost, and high-efficiency pre-training and verification of the core AI model are achieved. This effectively improves the robustness and recognition accuracy of the model under actual complex sea conditions and reduces the risks of on-site operations.

[0049] In step S120, based on the search mission instructions and L2 dynamic environment layer data, an adaptive path planning strategy is used to generate a first path in the digital twin platform, and the unmanned surface vessel equipped with synthetic aperture sonar is controlled to navigate along the first path and perform real-time imaging, including: Based on the range of the target area, the swath width and resolution parameters of the synthetic aperture sonar in the search task instruction, and combined with the water acoustic data of the L2 dynamic environmental layer, an adaptive path planning strategy is adopted to generate a first path in the digital twin platform. The search mission instruction refers to the command issued by the operator or higher-level system, specifying the basic parameters of the current seabed target search operation. Its core content includes the scope of the target area, usually defined by a set of latitude and longitude coordinates. The first path refers to the optimal global navigation trajectory planned for an unmanned surface vessel (USV) equipped with synthetic aperture sonar (SAR) to perform large-scale initial screening. Its goal is to achieve efficient and comprehensive acoustic coverage of the entire target area. SAR can refer to an advanced underwater acoustic imaging device. By receiving sound echoes from multiple locations during the movement of the USV and performing coherent processing, it can "synthesize" a virtual aperture much larger than the physical size, thereby obtaining a high-resolution seabed image. Its key parameters include swath width (the width of the seabed that can be covered by a single ping) and resolution (the smallest resolvable target size). The L2 dynamic environmental layer's acoustic data refers to the dynamic data describing the acoustic characteristics of seawater in the digital twin platform, primarily sound velocity profiles. The speed of sound varies significantly with depth, temperature, and salinity, affecting the propagation path, focusing, and imaging quality of sound waves. These are critical environmental factors that must be considered when planning sonar detection paths. Adaptive path planning strategies refer to intelligent planning algorithms that dynamically adjust or optimize paths based on real-time or up-to-date environmental information (such as L2 layer data), task constraints, and equipment status, rather than using a fixed route.

[0050] Specifically, the path planning algorithm first calculates the minimum number of parallel survey lines and their approximate orientation required to achieve 100% area coverage based on the target area's range and the swath width of the synthetic aperture sonar. Simultaneously, to ensure imaging quality, adjacent survey lines need to maintain a certain degree of overlap. Furthermore, the algorithm also queries the acoustic data (especially sound velocity profiles) of the L2 dynamic environment layer. Sound waves propagate through non-uniform water bodies, causing their paths to bend. The algorithm uses the sound velocity profile for ray tracing calculations to predict the actual coverage area of ​​the sound waves, thereby fine-tuning the preset survey line spacing to ensure effective detection of all areas on the seabed. If L2 layer data shows the presence of strong internal waves, thermoclines, or other water structures that would severely degrade sonar performance in a certain area, the adaptive strategy will attempt to plan a path around that area or adjust the navigation parameters when passing through it. The "adaptive path planning strategy" will simulate and evaluate multiple candidate paths in a digital twin platform, using the shortest total travel time, lowest total energy consumption, or highest overall coverage quality as objective functions. The final generated "first path" is a global scan route with optimal overall performance under given sonar parameters and dynamic environmental constraints.

[0051] The unmanned vessel is controlled to navigate along the first path, and the synthetic aperture sonar data transmitted back is processed in real time to generate an underwater acoustic image. Among them, seabed acoustic images refer to two-dimensional images generated by processing synthetic aperture sonar data, reflecting the seabed topography and the intensity of object reflection. Similar to optical photographs taken from the air, but depicting the acoustic features of the seabed.

[0052] Specifically, the generated "first path" (a series of waypoints) is sent to the UV's flight controller. This controller integrates data from GPS, inertial navigation unit, compass, and Doppler log to calculate its precise position, velocity, and attitude in real time. It then drives the thrusters through a closed-loop control algorithm, enabling the USV to navigate precisely along the preset path and resist wind and current interference. During navigation, the synthetic aperture sonar continuously emits acoustic pulses and receives echoes. The sonar system must precisely record the arrival time of each echo signal, along with the USV's high-precision position and attitude data at that moment. The transmitted raw acoustic data and synchronized motion data are sent to the edge computing node onboard the USV. The synthetic aperture sonar imaging algorithm running on the node performs core processing: using high-precision USV motion data, it corrects echo phase errors caused by platform turbulence and offset. Furthermore, it aligns and superimposes echo signals received from different locations at the same scattering point on the seabed based on their phase relationships. This process is equivalent to "synthesizing" a huge virtual array, thereby generating high-resolution underwater acoustic images with a resolution far exceeding that of ordinary side-scan sonar.

[0053] The acoustic image is analyzed in real time based on a target recognition algorithm, and suspicious target areas are automatically marked.

[0054] Among them, target recognition algorithms can refer to a set of automated computer programs (usually based on deep learning models, such as convolutional neural networks) used to analyze acoustic images, identify areas in the image that are significantly different from the surrounding seabed background features, and judge them as suspicious target areas.

[0055] Specifically, target recognition algorithms (such as deep learning models), pre-trained on massive amounts of virtual data within the digital twin platform, are deployed on the edge computing nodes of the USV. When a new acoustic image is generated, it is immediately fed into the algorithm for forward propagation inference. The algorithm automatically extracts multi-level features from the image (such as edges, textures, shapes, and intensity distributions) and compares and calculates them with its internally learned "target feature library" to determine whether a pre-defined category of target (such as a shipwreck or container) exists in the image and its location. When the algorithm's confidence level exceeds a set threshold, the region of the suspicious target is automatically marked on the image (usually with a rectangular bounding box). Simultaneously, the system uses the pixel coordinates of the target bounding box in the image, combined with the USV's geographical location at the time the image was generated, the sonar's geometric model, and imaging parameters, to reverse-calculate the real-world geographical coordinates of the suspicious target region. Finally, the bounding box, target category, confidence level, and geographical coordinates are packaged into a structured alert message and updated to the L3 real-time layer of the digital twin platform for operator review or to trigger subsequent ROV approach and reconnaissance procedures.

[0056] In this embodiment, by adaptively generating a first path based on the mission area, sonar parameters, and dynamic acoustic environment data, and controlling the unmanned vessel to navigate along this path, perform real-time imaging, and automatically analyze and label suspicious targets based on AI algorithms, efficient, intelligent, and automated initial screening of a large sea area is achieved. This transforms traditional manual interpretation into online real-time perception and early warning, greatly improving search speed and the efficiency of discovering suspicious targets.

[0057] In step S130, when the unmanned vessel identifies a suspicious target area during navigation, it is controlled to anchor and release a tethered underwater robot. Based on the anchoring position of the unmanned vessel and the suspicious target area, an adaptive cable-and-tube cooperative control strategy is used to generate a second path, and the tethered underwater robot is controlled to descend along the second path. This includes: When an unmanned surface vessel (USV) identifies a suspicious target area during navigation, the USV's edge computing nodes combine its own positioning information with the target position in the synthetic aperture sonar image to calculate the coordinates of the suspicious target. The navigation process refers to the stage where the unmanned surface vessel (USV) is performing a large-scale acoustic scan along a pre-planned "first path" (global scan path). A suspicious target area can refer to a seabed area that the synthetic aperture sonar onboard the USV determines with high confidence, based on real-time imaging and AI analysis, as potentially containing the object to be searched. Suspicious target coordinates refer to the absolute location of the suspicious target in the real world, typically expressed as three-dimensional coordinates using latitude, longitude, and depth (or altitude). The USV edge computing node refers to a high-performance embedded computer installed on the USV. It is responsible for running the L3 real-time layer of the digital twin platform, processing sonar data, executing local AI inference and real-time control algorithms, and serves as the "on-site brain" for collaborative operations. Self-positioning information refers to the high-precision geographical location (latitude and longitude), attitude (roll, pitch, heading), and speed information acquired in real-time by sensors such as GPS and inertial navigation systems. The target position in the synthetic aperture sonar image refers to the row and column coordinates of the "suspicious target" labeled by the AI ​​algorithm in the acoustic image pixel coordinate system.

[0058] Specifically, the unmanned surface vessel's (USV) self-positioning information is provided by an inertial navigation system that integrates GPS, IMU, and DVL data. This means that at the precise moment the current sonar image is generated, the USV's high-precision geographical location and attitude are obtained. The AI ​​algorithm provides the target's pixel coordinates within the sonar image.

[0059] Based on the time from sound wave transmission to reception (inherent in the raw sonar data), the slant range between the target and the sonar transducer is calculated. Combining the sonar's installation geometry, the USV's attitude (roll and pitch), and the target's pixel coordinates (corresponding to a specific beam pointing angle), a spherical or cylindrical geometric model is used to calculate the target's three-dimensional relative position vector with respect to the USV sonar array. This relative position vector is then transformed using coordinate rotation and translation based on the USV's absolute geographical location and heading at that moment, ultimately calculating the target's latitude, longitude, and depth in the geodetic coordinate system—its geographic coordinates.

[0060] Based on the geographic coordinates of the suspected target, the anchoring position of the unmanned vessel is determined; The anchoring position refers to the latitude and longitude of the USV's hull (or a specific reference point) when it finally performs anchoring operations. Furthermore, the anchoring position must ensure that, after anchoring, the ROV beneath the USV is within the effective operating radius of the umbilical cable, enabling it to cover and safely operate over suspicious targets.

[0061] Specifically, a potential anchorage area is delineated on the water surface, centered on the coordinates of the suspected target and with the maximum safe operating length of the ROV umbilical cable as the radius. Within this area, the L1 layer (seabed topography) and L2 layer (current and predicted current and wind fields) of the digital twin platform are consulted to select locations with flat seabeds, suitable seabed sediment for anchoring, and minimal combined wind / current / wave forces as candidate anchoring points. In the L3 real-time layer of the digital twin platform, rapid simulations are performed on the candidate anchoring points to predict the drift range of the USV after anchoring and the stress on the anchor chain. The point where the simulation results show the most stable positioning and the safest mooring force is selected. Considering all these factors, an optimal anchoring location is determined. This location may not be directly above the target, but rather a point to its side or above that provides the best operating window and safety margin.

[0062] Based on the anchoring position of the unmanned vessel and the geographical coordinates of the suspected target, an adaptive cable-tube cooperative control strategy is adopted to generate a second path; The adaptive cable-and-tube cooperative control strategy refers to an advanced control method. It uses a high-fidelity dynamic model integrated into a digital twin to predict the future state of the system comprising the USV, umbilical cable, and ROV in real time, and continuously optimizes control commands (winch deployment / retraction, ROV propulsion) to achieve safe, efficient, and low-stress cooperative motion. Its "adaptive" nature is reflected in its ability to adjust according to the real-time environment (water flow) and mission status. The second path specifically refers to the three-dimensional trajectory of a tethered underwater robot (ROV) as it descends from the USV release point and approaches a suspected target on the seabed. This path is the optimal trajectory dynamically generated by the model-predictive controller.

[0063] Specifically, in the digital twin (L3 real-time layer) of the USV edge node, the dynamic model of the current USV, the dynamic model of the ROV, and the finite element model of the umbilical cable are loaded, and the current anchoring position (as the system starting point) and the coordinates of the suspected target (as the endpoint) are input. The controller acquires or predicts the initial state of the ROV in real time through rolling optimization to generate a path. In each control cycle (e.g., per second), the controller performs a rapid forward simulation within the digital twin. An optimization problem is solved: finding a series of control commands (how the winch should raise and lower the cable, how much thrust the ROV thruster should provide) over a future period starting from the current moment, minimizing a cost function. This cost function simultaneously considers: ensuring the ROV eventually reaches the target coordinates; penalizing excessive cable bending, excessive stress, and proximity to obstacles; reducing total energy consumption; and avoiding drastic changes in control commands. The result of the optimization solution, i.e., the optimal evolution sequence of the system state over a future period, essentially defines a dynamic, smooth, and safe "second path." For example, this cost function can be a scalar function used to mathematically quantify and evaluate the "quality" of all possible system behavior trajectories (control sequences, i.e., real-time second paths) within the prediction time domain in each control cycle. Multiple control objectives, such as ensuring the ROV reaches the target coordinates, penalizing excessive cable bending, excessive stress, approaching obstacles, and reducing total energy consumption, are uniformly mapped to a minimizeable scalar value through weighted summation. Different weight coefficients are then assigned based on the importance of different objectives. The optimization algorithm dynamically generates the optimal "second path" for the current moment by finding the control sequence that minimizes this function value. However, the controller only executes the first step of the sequence. As the ROV begins to descend, real sensor data (water flow, attitude) is fed back, the digital twin is updated accordingly, and the controller re-performs the optimization solution based on the new state in the next cycle, thereby dynamically adjusting and generating subsequent paths. Therefore, the "second path" is generated in real-time and continuously optimized.

[0064] A release command is issued to release the tethered underwater robot, and the tethered underwater robot is controlled to descend along the second path to approach the coordinates of the suspected target.

[0065] The release command can refer to a command issued by the control system that initiates a series of mechanical actions to safely lower the ROV from the USV deck into the water.

[0066] Specifically, the control system issues a command, the A-frame or crane on the USV deck is activated, lifting the ROV and swinging it overboard before slowly lowering it into the water. Once in the water, the ROV thrusters activate, and the winch begins to release the umbilical cable in a controlled manner. The coordinated control commands (i.e., the instantaneous execution scheme of the "second path") generated by the "adaptive cable-and-rope coordinated control strategy" are decomposed into specific winch speed setpoints and thrust commands for each ROV thruster. These commands are sent to the USV's winch control system and the ROV's thruster control system for execution. Simultaneously, sensors on the ROV and cable provide real-time feedback on actual position, velocity, tension, attitude, and other data. The control system compares the actual state with the desired state of the "second path," forming a closed-loop feedback. Based on the feedback and the latest environmental forecasts, the model predictive controller regenerates a new set of optimized commands in the next control cycle, and this cycle repeats, driving the entire "USV-cable-ROV" system to safely and accurately dive towards the target coordinates along the dynamically optimized "second path." The entire process ensured that the ROV could resist drifting in complex water currents, maintain cable safety, and ultimately stabilize above the target.

[0067] In this embodiment, by accurately calculating the target coordinates, optimizing the unmanned vessel anchor positioning, and using an adaptive cable-tube collaborative control strategy to dynamically generate and execute a second path, the ROV is controlled to dive, achieving a safe, stable, and precise approach from the surface platform to the underwater target point. This effectively solves the problems of excessive cable stress, entanglement, and significant impact from water flow disturbances, ensuring the smooth initiation of subsequent detailed exploration.

[0068] In step S140, when the tethered underwater robot descends to the target depth, based on the multi-source sensor data acquired by the tethered underwater robot, the adaptive path planning strategy is used to control the tethered underwater robot to conduct close-range detailed exploration and confirmation of the suspected target area, including: When the underwater robot descends to the target depth, it acquires multi-source sensor data based on the optical camera, high-resolution profile sonar, and multi-parameter sensors mounted on the tethered underwater robot. Target depth refers to the calculated depth of the suspected target, typically near the seabed, and is the depth at which the ROV needs to reach and conduct detailed exploration during its dives. Optical cameras refer to the high-definition cameras mounted on the ROV, used to acquire real-time video streams and still images of the target under illumination conditions. They are core sensors for visual guidance in appearance recognition, condition assessment, and fine-grained operations (such as robotic arm operations). High-resolution profile sonar refers to an acoustic imaging device that provides high-resolution cross-sectional images perpendicular to the seabed. It is crucial for understanding the target's internal structure, buried parts, and interface with the seabed, especially in turbid waters where it can compensate for the limitations of optical equipment. Multi-parameter sensors typically refer to environmental monitoring devices such as CTD sensors (measuring conductivity / salinity, temperature, and depth), used to acquire water environmental parameters at the operational site to aid data interpretation and system modeling. Multi-source sensor data refers to the collection of raw or pre-processed data simultaneously acquired from different types of sensors, including optical, acoustic, and environmental sensors. Its "multi-source" nature is reflected in the diversity of data modalities (images, acoustic spectra, numerical values) and information dimensions (appearance, structure, environment).

[0069] Specifically, once the ROV reaches the target depth and stabilizes near the suspected target, the control system automatically or via operator command activates and optimizes relevant sensors. The optical camera activates a high-powered underwater light, adjusts the focus, aperture, and white balance, and begins capturing high-definition video streams and periodic high-resolution photographs. The high-resolution profile sonar adjusts the sonar frequency, range, and scanning mode, initiating continuous fan-shaped or lateral cross-sectional scanning of the target, generating a sequence of acoustic images reflecting the target's profile characteristics. Multi-parameter sensors continuously record data such as depth, water temperature, and salinity at the current work site.

[0070] All sensor data is embedded with precise timestamps and location tags (linked to the ROV's positioning data). Optical video and images, raw sonar echo or image data, and numerical data such as CTD are uploaded in real time and synchronously to the edge computing node of the USV via fiber optic channels within the umbilical cable. At the USV edge node, this data is preprocessed and packaged into a structured "multi-source sensor data" stream, which is used for local real-time processing and also selectively uploaded to the cloud-based shore-based center via satellite links.

[0071] Based on the multi-source sensor data, an adaptive path planning strategy generates a third path, and the tethered underwater robot is controlled to conduct close-range, detailed exploration and confirmation of the suspicious target along the third path.

[0072] Among them, the adaptive path planning strategy is an intelligent algorithm that can dynamically generate or optimize movement paths based on real-time sensing information. In this step, it plans the optimal observation path in real time based on multi-source data acquired after the ROV approaches. The third path specifically refers to the movement trajectory executed by the ROV when conducting close-range detailed exploration around a suspicious target. The core objective of this path is to guide the ROV to conduct a full-coverage, multi-angle scan and observation of the target from the best perspective, rather than simply approaching a single point. Close-range detailed exploration and confirmation: This refers to the process by which the ROV, at a location very close to the suspicious target (usually within a few meters), uses high-resolution sensors to conduct a detailed examination to ultimately determine its attributes (such as whether it is a shipwreck, pipeline, mineral deposit, etc.) and status.

[0073] Specifically, USV edge nodes or ROV local AI units process acquired multi-source data in real time. For example, computer vision algorithms are used to analyze optical images to identify the target's outline and feature points; sonar images are processed to understand the target's three-dimensional structure and its relative position to the seabed. The processing results, along with the original sensor data, are fused in the L3 real-time layer of the digital twin platform to dynamically update the accurate 3D model of the suspected target and the surrounding environment in the virtual model. The "adaptive path planning strategy" is based on the aforementioned real-time fused perception information and performs rapid simulation and decision-making within the digital twin. Its goal is to generate a path that allows the ROV to optimally complete its exploration mission. Evaluation criteria include: whether it can cover all key parts of the target, such as the bow, midships, and stern of a shipwreck; whether it can acquire identification features from the best angles (such as front, side, and top); whether it maintains a safe distance from the target, seabed, or other obstacles; and whether the total path length is short and the time consumption is minimal.

[0074] The algorithm may dynamically plan various "third path" modes, such as: Circular scanning: controlling the ROV to cruise around the target in a circle or square for all-around observation; Fixed-point detailed investigation: planning a series of stopping points, allowing the ROV to hover at each point for multi-sensor concentrated detection; Contour tracking: guiding the ROV to fly along the edge of the target or a specific structure (such as a pipe). The planned "third path" (a series of waypoints or continuous trajectories) is translated into specific ROV thruster control commands. The control system drives the ROV to move precisely along this path. During the movement, a cloud-edge collaboration mechanism plays a role: the edge side handles real-time obstacle avoidance and stable tracking, while the cloud side may perform in-depth analysis of complex situations and provide strategy optimization suggestions. Throughout the execution of the "third path," multi-source sensor data is continuously acquired and analyzed. Once conclusive identification evidence is obtained (such as a clear image of the shipwreck's name or confirmation of the pipe interface type), or a pre-set comprehensive scan is completed, the system can determine that the exploration is complete and perform final confirmation of the target (including target type, precise coordinates, status, etc.), updating the results to the digital twin platform and mission log. If the target does not match, it is marked as "excluded."

[0075] In this embodiment, by acquiring multi-source sensor data after the ROV approaches, and generating a third path in real time based on this data using an adaptive path planning strategy, the ROV is controlled to conduct close-range, detailed exploration and confirmation of the target. This achieves close-range, multi-angle, and adaptive all-round observation of suspicious targets, ensuring the final accuracy and reliability of target identification, and enabling the acquisition of the target's detailed features and status information.

[0076] In step S150, based on the exploration and confirmation results and the preset task completion criteria, it is determined whether the seabed target search task is completed. If completed, the multi-source data acquired during the search process is fused, recorded, and labeled, including: Based on the exploration and confirmation results and the preset task completion criteria, it is determined whether the seabed target search task has been completed. The exploration and confirmation results refer to the final conclusions reached after the tethered underwater vehicle (ROV) performs the "close-up detailed exploration and confirmation" step. For example, the target may be "confirmed" to be a certain type of shipwreck, with precise coordinates and image evidence; or the target may be "excluded" and confirmed to be a natural reef. Pre-defined mission completion criteria refer to a set of quantitative or qualitative indicators pre-defined before the mission begins to determine whether the mission can be completed. These typically include: target achievement criteria: such as a specific target being found and confirmed; area coverage criteria: such as 100% acoustic scan coverage of a designated area; and system resource / environmental limitations: such as reaching the maximum operating time, energy depletion, or sea conditions deteriorating to a safety threshold. Multi-source data refers to all types of data generated throughout the mission process. This mainly includes global scanning sonar images from the "first path" phase, optical (video / image) and acoustic (profile sonar) data from the close-up exploration in the "third path" phase, as well as target coordinates and attribute information, equipment status logs, and environmental data throughout the mission. Integration and annotation refers to the process of integrating, associating, and adding metadata to the aforementioned multi-source data, making it a data package with a unified index that is easy to retrieve and analyze.

[0077] Specifically, the system aggregates "exploration and confirmation results," such as "Target A has been confirmed as a shipwreck, coordinates AAA," and real-time monitoring of "task status," such as current area coverage, remaining battery power, and operation time. This information is then compared item by item with the "preset task completion criteria": Criterion 1: Target Confirmation: If the task is to search for a specific target, and the "exploration and confirmation results" show that the target has been found and confirmed, then the "target achieved" condition is met. Criterion 2: Area Coverage: If the task is to conduct a general survey of a certain area, the system calculates the proportion of the scanned area to the total area of ​​the target area. When the proportion reaches a preset value (e.g., 100%) and there are no major omissions, the "area coverage" condition is met. Criterion 3: Resource / Environmental Constraints: The system continuously monitors battery power and time. If the remaining battery power is below the safe return threshold, or the operation time reaches the preset upper limit, or the L2 dynamic layer forecast indicates that future sea conditions will exceed the safe operation range, then the "forced termination" condition is triggered, and the task must be judged as completed (possibly unplanned completion).

[0078] When any preset completion criterion is met, the USV edge node or shore-based center will generate a "mission completion recommendation" along with the judgment criteria. This recommendation will be presented to the operator at the shore-based control center for final review and confirmation. The operator can view the mission summary dashboard in the digital twin platform, verify key data, and make a final decision to "confirm completion" or "continue supplementary exploration".

[0079] If the task is completed, the global scanning sonar images, close-range exploration optical and acoustic data, target coordinates and attribute information are stored in association. Global scanning sonar images refer to wide-area seabed acoustic images covering the entire target area generated by synthetic aperture sonar onboard an unmanned surface vessel (ROV) during the "first path" scan. Close-range exploration optical and acoustic data specifically refer to close-range, high-definition video, photographs, and acoustic profile images acquired by the ROV during the "third path" phase using its optical cameras and high-resolution profiling sonar. Target coordinates and attribute information refer to the precise three-dimensional geographic coordinates of confirmed targets, along with descriptive information such as their type, size, and status. Association-based storage refers to a data management method that links data from different sources and of different types through shared key indexes (such as mission ID, timestamp, and geographic coordinates) and stores them in a logically unified database, ensuring that the inherent connections between data are preserved.

[0080] Specifically, after the mission is completed, data scattered across USV edge nodes, ROV local caches, and the cloud are aggregated to the data management platform at the shore-based center. The data is cleaned, deredundanted, and unified in time reference and coordinate system. A unique mission ID is created for this mission. All data entities are associated through one or more of the following key indexes: Mission ID: All data belongs to this mission. Timestamp: A time stamp accurate to milliseconds, which can associate data from different sensors at the same time, such as aligning optical images with acoustic images and ROV pose. Geographic coordinates: Each pixel in the sonar image, the target in the optical image, and the ROV's trajectory are all associated with a unified geographic reference system through coordinate transformation. The target's coordinates and attribute information serve as the core index for all exploration data of that target. The processed data is stored in a spatiotemporal database or object storage system, including: global scan sonar images, stored as a geographic raster data layer, along with imaging parameters and coverage metadata. Close-range exploration yields optical and acoustic data, including videos, images, and sonar cross-sectional data, which are stored as files. However, their metadata (capture time, location, sensor parameters, and associated target ID) is extracted into a relational database for efficient retrieval. Target coordinates and attribute information are stored as structured records in the database. Each record contains the target ID, type, confidence level, precise coordinates, and links to all related evidence files (images, videos).

[0081] Through this "associative storage," when a review is needed, all related wide-area sonar images, close-up optical photographs, acoustic profiles, and the trajectory and attitude of the ROV during observation can be retrieved instantly using a single target ID, forming a complete and traceable "data story chain."

[0082] Based on the full-process data of this mission, the sonar image processing model, target recognition model, and adaptive path planning strategy in the digital twin platform were optimized and updated.

[0083] The "full-process data" refers to all data records generated during this mission, from initialization, planning, searching, exploration to completion, including sensor data, control commands, decision logs, and intermediate model results. The "sonar image processing model" refers to the algorithm module (potentially containing adjustable parameters) used to process raw sonar echo data and generate high-quality acoustic images. The "target recognition model" refers to the artificial intelligence model (such as a convolutional neural network) used to automatically detect and classify targets from acoustic or optical images. The "adaptive path planning strategy" refers to the intelligent planning algorithm and its parameter set used during the generation of the "first, second, and third paths" to dynamically adjust according to the environment and mission status.

[0084] Specifically, the "full-process data" of this task is extracted from the "associative storage" database as a high-quality training and validation dataset. In particular, the target area data labeled with results (confirmation or exclusion) is extremely valuable.

[0085] The optimization of the sonar image processing model involves using the raw sonar echo data collected in this mission and the known high-quality imaging results (images verified by humans) to adjust the parameters in the imaging algorithm (such as filter coefficients and motion compensation parameters), or using data-driven methods to train a neural network model that can better suppress the noise specific to the sea area in this mission, so as to achieve higher imaging quality in similar environments in the future.

[0086] The target recognition model optimization involves adding precisely labeled acoustic / optical images (positive samples: confirmed targets; negative samples: excluded false targets or difficult samples) acquired in this task to the model's training set. These AI models are then retrained or fine-tuned using powerful computing capabilities in the cloud. This allows the model to learn new target features and false alarm patterns encountered in this task, resulting in more accurate identification and a lower false alarm rate in future tasks. The "reinforcement learning" mentioned in the document can also utilize the operation sequence and results of this task as experience replay to optimize decision-making strategies.

[0087] The optimization of the adaptive path planning strategy specifically involves analyzing the actual execution performance data of the "first, second, and third paths" in this task. For example, comparing the deviation between the planned and actual paths optimizes the USV's path tracking controller parameters. Analyzing the cable stress data during the "second path" dive under specific water flow conditions optimizes the cost function weights or model predictive controller parameters in the "adaptive cable-tube cooperative control strategy." Finally, successful observed path patterns from the "third path" are summarized and abstracted into rules to enrich the expert knowledge base of the "adaptive path planning strategy."

[0088] The optimized new model and strategy parameters, after testing and verification, will be packaged into a new software version and deployed to the digital twin platform and USV / ROV edge computing nodes before the next task. In this way, each time the system executes a real task, it becomes a learning opportunity, thereby achieving continuous iterative optimization that makes it smarter with use.

[0089] In this embodiment, the task completion is judged based on the exploration results and preset standards, and multi-source data such as global scanning and close exploration are stored in a correlated manner. Finally, the core model of the platform is optimized and updated based on the full-process data, realizing closed-loop management of tasks, in-depth mining of data value, and continuous self-evolution of system capabilities. This forms a virtuous cycle of "execution-learning-optimization" and continuously improves the overall operation performance.

[0090] Optionally, the seabed exploration mission further includes: If the communication between the unmanned vessel and the shore-based control center is interrupted, the edge node of the unmanned vessel will autonomously decide on the subsequent scanning path or the order of exploration of the discovered suspicious targets based on the acquired scanning data, the preset search priority and the digital twin simulation. The communication interruption between the unmanned surface vessel (USV) and the shore-based control center refers to a failure of the main communication link (usually satellite communication) between the USV and the remote command center (cloud), resulting in the inability to issue commands or transmit data. The shore-based control center, located on land or on the mothership, is the command hub responsible for overall monitoring, mission planning, advanced decision-making, and data analysis. The unmanned surface vessel edge node refers to the high-performance embedded computer installed on the USV. It is the carrier of "edge computing," carrying the L3 real-time layer of the digital twin platform, the local AI model, and autonomous control logic, and is the highest-level decision-making unit on-site after the interruption. Acquired scan data refers to the synthetic aperture sonar image data and its corresponding area coverage information that the USV had already collected and processed before the communication interruption. Preset search priorities refer to the rules predefined during the mission planning phase to guide the order of mission execution. For example, "prioritize 100% coverage of area A" or "once a type B target is detected, immediately pause scanning and prioritize exploration." Digital twin simulation refers to the L3 real-time layer digital twin model running on the USV edge node, synchronized with the physical world. After communication is interrupted, it can operate independently of real data input, performing future task deductions and simulations based on its current state in a virtual environment. Autonomous decision-making refers to the intelligent behavior of edge nodes, relying on built-in algorithms and rule bases, to independently plan subsequent actions when cloud commands are unavailable. Subsequent scanning path refers to the autonomously planned navigation route for continuing the search of the remaining area before the USV has completed a global scan of the target area during communication interruption. The order of exploration for identified suspicious targets refers to the order in which edge nodes autonomously decide to approach and explore multiple suspicious targets when they have been identified but not yet confirmed by the ROV.

[0091] Specifically, the USV edge nodes continuously monitor the "heartbeat" signal with the shore-based system. Once a communication interruption is confirmed, the "autonomous operation mode" is immediately activated. The edge nodes retrieve all acquired scan data and clearly mark the scanned and unscanned areas in the digital twin (L3 layer). At the same time, all discovered but unconfirmed suspicious targets, along with their locations and preliminary attributes, are listed.

[0092] Priority-based decision tree judgment: Edge nodes query preset search priority rules. Typical decision logic includes: Rule A (Coverage Priority): If the core task is to conduct a general survey of a certain area, priority drives the system to continue scanning. In the digital twin, the edge node simulates and generates one or more "subsequent scanning paths" with the goal of "most efficiently completing the coverage of the remaining area," and selects the optimal one to execute. Rule B (Target Confirmation Priority): If the core task is to find a specific target, and high-confidence suspicious points have been found, priority drives the system to explore them first. The edge node sorts all found targets (e.g., by confidence level, distance), simulates the estimated time and success probability of exploring each target with an ROV in the digital twin, and thus decides an optimal "exploration order." After making an initial decision (whether to continue scanning or explore target X), the edge node performs a rapid simulation in the digital twin to predict the USV's navigation status, energy consumption, and estimated task completion time under this decision. Only when the simulation results meet the requirements of safety and mission feasibility will the decision be finally executed and transformed into actual control commands for the USV (and subsequently the ROV).

[0093] If communication between the tethered underwater robot and the unmanned vessel is interrupted, the underwater robot will execute a preset emergency exploration action or a safe surfacing procedure based on the last command and the local AI module.

[0094] The communication interruption between the tethered underwater robot (ROV) and the unmanned surface vessel (USV) refers to a failure in the data link established via the umbilical cable, preventing the ROV from receiving real-time control commands from the USV and from uploading sensor data. The last command refers to the last valid control command packet received and successfully executed by the ROV from the USV just moments before the communication interruption. The local AI module refers to the embedded AI computing unit deployed on the ROV itself. It has certain target recognition, scene understanding, and simple decision-making capabilities and can work independently when disconnected from the USV. Pre-programmed emergency exploration actions refer to a set of safe and conservative operational instructions pre-programmed and stored locally on the ROV. For example, "If interrupted during detailed exploration, continue to complete one orbital observation and then hover." The safe ascent procedure refers to a set of standardized operational sequences that ensure the ROV can be safely recovered in an emergency, typically including: stopping operations, retracting the robotic arm, surfacing vertically or along a pre-programmed safe path below the USV, and sending an acoustic response signal.

[0095] Specifically, the ROV continuously monitors its communication link with the USV. Upon interruption, it immediately switches to "local autonomous mode." The ROV's controller parses the last instruction received before the communication interruption. For example, the last instruction might be "performing orbital observation of target A." Simultaneously, the local AI module quickly analyzes current sensor data (depth, attitude, camera footage) to understand its current mission stage, such as being in the middle of orbital observation. If the last instruction is a clear operational command, and the local environment assessment indicates safety, the local AI module will take over control and attempt to complete the remaining parts of the command. For example, if the interruption occurred while orbiting a target, the AI ​​will attempt to autonomously complete the remaining orbital path based on local visual or sonar data, acquiring as much data as possible. All data will be cached locally. The safe ascent procedure will be prioritized under any of the following conditions: The last instruction is a safety-related command, such as "surface" or "recover." The local AI assesses environmental hazards: strong water currents are detected, the ROV's attitude is abnormal, or visibility is rapidly decreasing. Pre-set actions are completed or fail: emergency reconnaissance actions are completed, or obstacles that cannot be resolved locally are encountered during execution. Safety timer reached: The interruption time exceeds the preset safety waiting time, such as 5 minutes. Safe ascent procedure executed: The ROV immediately stops all exploratory actions, and the controller initiates the standardized recovery procedure: adjusting its attitude to positive buoyancy, activating the vertical thrusters or ascending at a small angle, while simultaneously attempting to send its own position signal to the USV via the underwater transponder. It will continuously attempt to reconnect with the USV until successfully recovered.

[0096] In this embodiment, two emergency mechanisms together constitute the system's "neural redundancy" in the unreliable deep-sea environment. The autonomous decision-making of the USV edge nodes ensures mission continuity, allowing the mission to continue intelligently even if communication with the rear is lost. The ROV's local emergency procedures ensure equipment survivability, ensuring that the safety of personnel and equipment is the highest priority under any circumstances.

[0097] Example 3: Figure 2 This is a framework diagram of a collaborative seabed target search path planning system based on synthetic aperture sonar for an unmanned surface vessel and a tethered underwater robot, provided in Embodiment 3 of the present invention. Figure 2 As shown, the system includes: The construction module 210 is used to build and initialize a hierarchical digital twin platform that supports target search. The platform includes an L1 prior information layer, an L2 dynamic environment layer, and an L3 real-time detection and collaboration layer. The first path generation module 220 is used to generate a first path in the digital twin platform based on the search task instructions and L2 dynamic environment layer data, using an adaptive path planning strategy, and control the unmanned vessel equipped with synthetic aperture sonar to navigate along the first path and perform real-time imaging. The second path generation module 230 is used to control the unmanned vessel to anchor and release the tethered underwater robot when the unmanned vessel identifies a suspicious target area during navigation. Based on the anchoring position of the unmanned vessel and the suspicious target area, an adaptive cable-tube cooperative control strategy is adopted to generate a second path and control the tethered underwater robot to dive along the second path. The exploration module 240 is used to control the tethered underwater robot to conduct close-range detailed exploration and confirmation of the suspicious target area based on the multi-source sensor data acquired by the tethered underwater robot and the adaptive path planning strategy when the tethered underwater robot dives to the target depth. The judgment module 250 is used to determine whether the seabed target search task is completed based on the exploration and confirmation results and the preset task completion standard. If completed, the multi-source data obtained during the search process is fused, recorded and labeled.

[0098] Optionally, the building module 210 is specifically used for: Collect and fuse multibeam topographic data, side-scan sonar historical data and known target information of the target sea area to construct the L1 prior information layer; By acquiring real-time and forecasted marine environmental data, an L2 dynamic environmental layer containing water acoustic characteristic parameters is constructed. Based on the detection model of unmanned ships and their onboard synthetic aperture sonar, and the detection model of tethered underwater robots and their sensors, an L3 real-time detection and collaboration layer is constructed. Based on the L1 prior information layer, L2 dynamic environment layer, and L3 real-time detection and collaboration layer, the hierarchical digital twin platform is constructed, and the platform is initialized, loading the target feature library and search strategy.

[0099] Optionally, the building module 210 is specifically used for: In the simulation environment of the digital twin platform, training scenarios are defined that include different seabed topography, target types, and environmental disturbances; Based on the imaging principle and target recognition algorithm of synthetic aperture sonar, virtual training and testing were conducted on the sonar image processing model and the target recognition model.

[0100] Optionally, the first path generation module 220 is specifically used for: Based on the range of the target area, the swath width and resolution parameters of the synthetic aperture sonar in the search task instruction, and combined with the water acoustic data of the L2 dynamic environmental layer, an adaptive path planning strategy is adopted to generate a first path in the digital twin platform. The unmanned vessel is controlled to navigate along the first path, and the synthetic aperture sonar data transmitted back is processed in real time to generate an underwater acoustic image. The acoustic image is analyzed in real time based on a target recognition algorithm, and suspicious target areas are automatically marked.

[0101] Optionally, the second path generation module 230 is specifically used for: When an unmanned surface vessel (USV) identifies a suspicious target area during navigation, the USV's edge computing nodes combine its own positioning information with the target position in the synthetic aperture sonar image to calculate the coordinates of the suspicious target. Based on the geographic coordinates of the suspected target, the anchoring position of the unmanned vessel is determined; Based on the anchoring position of the unmanned vessel and the geographical coordinates of the suspected target, an adaptive cable-tube cooperative control strategy is adopted to generate a second path; A release command is issued to release the tethered underwater robot, and the tethered underwater robot is controlled to descend along the second path to approach the coordinates of the suspected target.

[0102] Optional, the exploration module 240 is specifically used for: When the underwater robot descends to the target depth, it acquires multi-source sensor data based on the optical camera, high-resolution profile sonar, and multi-parameter sensors mounted on the tethered underwater robot. Based on the multi-source sensor data, an adaptive path planning strategy generates a third path, and the tethered underwater robot is controlled to conduct close-range, detailed exploration and confirmation of the suspicious target along the third path.

[0103] Optional, the judgment module 250 is specifically used for: Based on the exploration and confirmation results and the preset task completion criteria, it is determined whether the seabed target search task has been completed. If the task is completed, the global scanning sonar images, close-range exploration optical and acoustic data, target coordinates and attribute information are stored in association. Based on the full-process data of this mission, the sonar image processing model, target recognition model, and adaptive path planning strategy in the digital twin platform were optimized and updated.

[0104] Optionally, it also includes: an interrupt module, specifically used for: If the communication between the unmanned vessel and the shore-based control center is interrupted, the edge node of the unmanned vessel will autonomously decide on the subsequent scanning path or the order of exploration of the discovered suspicious targets based on the acquired scanning data, the preset search priority and the digital twin simulation. If communication between the tethered underwater robot and the unmanned vessel is interrupted, the underwater robot will execute a preset emergency exploration action or a safe surfacing procedure based on the last command and the local AI module.

[0105] The synthetic aperture sonar-based unmanned surface vessel and tethered underwater robot collaborative seabed target search path planning system provided in the embodiments of the present invention can execute the synthetic aperture sonar-based unmanned surface vessel and tethered underwater robot collaborative seabed target search path planning method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0106] Example 4: In some embodiments, a method for collaborative seabed target search path planning between an unmanned surface vessel (USV) and a tethered underwater robot based on synthetic aperture sonar can be implemented as a computer program tangibly contained in a computer-readable storage medium. The computer program for implementing the method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0107] In the context of this invention, a computer-readable storage medium stores computer instructions that, when executed by a processor, implement the method provided by this invention for collaborative seabed target search path planning between an unmanned surface vessel and a tethered underwater robot based on synthetic aperture sonar. The computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. The computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, the computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0108] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0109] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for collaborative seabed target search path planning between an unmanned surface vessel and a tethered underwater robot based on synthetic aperture sonar, characterized in that, include: A hierarchical digital twin platform supporting target search is constructed and initialized. The platform includes an L1 prior information layer, an L2 dynamic environment layer, and an L3 real-time detection and collaboration layer. Based on the search mission instructions and L2 dynamic environment layer data, an adaptive path planning strategy is adopted to generate a first path in the digital twin platform, and the unmanned vessel equipped with synthetic aperture sonar is controlled to navigate along the first path and perform real-time imaging. When the unmanned vessel identifies a suspicious target area during navigation, it is controlled to anchor and release a tethered underwater robot. Based on the anchoring position of the unmanned vessel and the suspicious target area, an adaptive cable-tube cooperative control strategy is used to generate a second path, and the tethered underwater robot is controlled to dive along the second path. When the tethered underwater robot dives to the target depth, based on the multi-source sensor data acquired by the tethered underwater robot, the adaptive path planning strategy is used to control the tethered underwater robot to conduct close-range detailed exploration and confirmation of the suspicious target area. Based on the exploration and confirmation results and the preset task completion criteria, it is determined whether the seabed target search task has been completed. If completed, the multi-source data obtained during the search process is fused, recorded, and labeled.

2. The method according to claim 1, characterized in that, The construction and initialization of a hierarchical digital twin platform supporting target search are described. The platform includes an L1 prior information layer, an L2 dynamic environment layer, and an L3 real-time detection and collaboration layer, comprising: Collect and fuse multibeam topographic data, side-scan sonar historical data and known target information of the target sea area to construct the L1 prior information layer; By acquiring real-time and forecasted marine environmental data, an L2 dynamic environmental layer containing water acoustic characteristic parameters is constructed. Based on the detection model of unmanned ships and their onboard synthetic aperture sonar, and the detection model of tethered underwater robots and their sensors, an L3 real-time detection and collaboration layer is constructed. Based on the L1 prior information layer, L2 dynamic environment layer, and L3 real-time detection and collaboration layer, the hierarchical digital twin platform is constructed, and the platform is initialized, loading the target feature library and search strategy.

3. The method according to claim 1, characterized in that, After constructing and initializing the hierarchical digital twin platform that supports target search, the following steps are also included: In the simulation environment of the digital twin platform, training scenarios are defined that include different seabed topography, target types, and environmental disturbances; Based on the imaging principle and target recognition algorithm of synthetic aperture sonar, virtual training and testing were conducted on the sonar image processing model and the target recognition model.

4. The method according to claim 1, characterized in that, The process involves generating a first path in a digital twin platform based on search mission instructions and L2 dynamic environment layer data, using an adaptive path planning strategy. This path is then used to control an unmanned surface vessel equipped with synthetic aperture sonar to navigate along the first path and perform real-time imaging. The process includes: Based on the range of the target area, the swath width and resolution parameters of the synthetic aperture sonar in the search task instruction, and combined with the water acoustic data of the L2 dynamic environmental layer, an adaptive path planning strategy is adopted to generate a first path in the digital twin platform. The unmanned vessel is controlled to navigate along the first path, and the synthetic aperture sonar data transmitted back is processed in real time to generate an underwater acoustic image. The acoustic image is analyzed in real time based on a target recognition algorithm, and suspicious target areas are automatically marked.

5. The method according to claim 1, characterized in that, When the unmanned surface vessel (USV) identifies a suspicious target area during navigation, it is controlled to anchor and release a tethered underwater robot. Based on the USV's anchoring position and the suspicious target area, an adaptive cable-and-tube cooperative control strategy is used to generate a second path, and the tethered underwater robot is controlled to descend along the second path, including: When an unmanned surface vessel (USV) identifies a suspicious target area during navigation, the USV's edge computing nodes combine its own positioning information with the target position in the synthetic aperture sonar image to calculate the coordinates of the suspicious target. Based on the geographic coordinates of the suspected target, the anchoring position of the unmanned vessel is determined; Based on the anchoring position of the unmanned vessel and the geographical coordinates of the suspected target, an adaptive cable-tube cooperative control strategy is adopted to generate a second path; A release command is issued to release the tethered underwater robot, and the tethered underwater robot is controlled to descend along the second path to approach the coordinates of the suspected target.

6. The method according to claim 1, characterized in that, When the tethered underwater robot descends to the target depth, based on the multi-source sensor data acquired by the tethered underwater robot, the adaptive path planning strategy is used to control the tethered underwater robot to conduct close-range, detailed exploration and confirmation of the suspicious target area, including: When the underwater robot descends to the target depth, it acquires multi-source sensor data based on the optical camera, high-resolution profile sonar, and multi-parameter sensors mounted on the tethered underwater robot. Based on the multi-source sensor data, an adaptive path planning strategy is used to generate a third path, and the tethered underwater robot is controlled to conduct close-range, detailed exploration and confirmation of the suspicious target along the third path.

7. The method according to claim 1, characterized in that, Based on the exploration and confirmation results and the preset task completion criteria, it is determined whether the seabed target search task is completed. If completed, the multi-source data acquired during the search process is fused, recorded, and labeled, including: Based on the exploration and confirmation results and the preset task completion criteria, it is determined whether the seabed target search task has been completed. If the task is completed, the global scanning sonar images, close-range exploration optical and acoustic data, target coordinates and attribute information are stored in association. Based on the full-process data of this mission, the sonar image processing model, target recognition model, and adaptive path planning strategy in the digital twin platform were optimized and updated.

8. The method according to claim 1, characterized in that, Also includes: If the communication between the unmanned vessel and the shore-based control center is interrupted, the edge node of the unmanned vessel will autonomously decide on the subsequent scanning path or the order of exploration of the discovered suspicious targets based on the acquired scanning data, the preset search priority and the digital twin simulation. If communication between the tethered underwater robot and the unmanned vessel is interrupted, the underwater robot will execute a preset emergency exploration action or a safe surfacing procedure based on the last command and the local AI module.

9. A collaborative seabed target search path planning system based on synthetic aperture sonar for unmanned surface vessels and tethered underwater robots, characterized in that, The system is used to execute the collaborative seabed target search path planning method based on synthetic aperture sonar for unmanned surface vessels and tethered underwater robots as described in any one of claims 1-8, including: The building module is used to build and initialize a hierarchical digital twin platform that supports target search. The platform includes an L1 prior information layer, an L2 dynamic environment layer, and an L3 real-time detection and collaboration layer. The first path generation module is used to generate a first path in the digital twin platform based on the search task instructions and L2 dynamic environment layer data, using an adaptive path planning strategy, and control the unmanned vessel equipped with synthetic aperture sonar to navigate along the first path and perform real-time imaging. The second path generation module is used to control the unmanned vessel to anchor and release the tethered underwater robot when the unmanned vessel identifies a suspicious target area during navigation. Based on the anchoring position of the unmanned vessel and the suspicious target area, an adaptive cable-tube cooperative control strategy is used to generate a second path and control the tethered underwater robot to dive along the second path. The exploration module is used to control the tethered underwater robot to conduct close-range, detailed exploration and confirmation of the suspicious target area based on the multi-source sensor data acquired by the tethered underwater robot and the adaptive path planning strategy when the tethered underwater robot dives to the target depth. The judgment module is used to determine whether the seabed target search task is completed based on the exploration and confirmation results and the preset task completion criteria. If completed, the multi-source data obtained during the search process is fused, recorded and labeled.

10. The system according to claim 9, characterized in that, The building module is specifically used for: Collect and fuse multibeam topographic data, side-scan sonar historical data and known target information of the target sea area to construct the L1 prior information layer; By acquiring real-time and forecasted marine environmental data, an L2 dynamic environmental layer containing water acoustic characteristic parameters is constructed. Based on the detection model of unmanned ships and their onboard synthetic aperture sonar, and the detection model of tethered underwater robots and their sensors, an L3 real-time detection and collaboration layer is constructed. Based on the L1 prior information layer, L2 dynamic environment layer, and L3 real-time detection and collaboration layer, the hierarchical digital twin platform is constructed, and the platform is initialized, loading the target feature library and search strategy.

Images