An air-sea joint near-shore sea floating garbage low-disturbance removal control method and system

Abstract

The present application relates to a kind of air-sea linkage's near-shore sea garbage low-disturbance removal control method and system, belong to marine garbage processing field.The present application identifies sea garbage according to sea video, and obtains the geodetic geographic coordinates of sea garbage;According to the geodetic geographic coordinates of sea garbage, garbage cleaning route is planned for water surface garbage cleaner;The cleaning effect evaluation is carried out to the cleaning operation of water surface garbage cleaner, and water surface garbage cleaner is dynamically adjusted according to the evaluation result.The present application can be carried out wide area sea garbage sensing and water surface platform scheduling by unmanned aerial vehicle, realize the efficient search and salvage route accurate planning of sea garbage under air-sea linkage;Meanwhile, through systematic acousto-optic control and non-destructive flexible recovery structure design, eliminate the interference and damage caused to marine organisms and their habitat, realize the large-scale efficient removal and environment-friendly recovery of near-shore sea garbage.Provide technical and equipment support for the ecological management of high-value and ecologically sensitive sea areas.

Description

Technical Field

[0001] This invention relates to the field of marine debris treatment technology, and in particular to a method and system for low-disturbance removal and control of nearshore marine debris that combines air and sea operations. Background Technology

[0002] The continuous accumulation of marine debris in nearshore waters has caused multiple harms to the marine ecological environment and socio-economic conditions. Marine debris easily leads to accidental ingestion and mechanical entanglement by seabirds and sea turtles, damaging their health. Microplastics formed from the degradation of marine debris can accumulate organic pollutants and heavy metals in the water, and through the bioaccumulation and amplification effect of food webs, pose a threat to human food safety and public health. At the same time, high densities of nearshore marine debris also significantly impact the economic benefits of coastal tourism and aquaculture. Therefore, there is an urgent need to establish efficient and precise early interception and removal programs for nearshore marine debris.

[0003] There are three main types of existing nearshore marine debris cleanup solutions: manned vessel mechanical salvage, static physical barrier interception, and unmanned platform intelligent recycling. Manned vessel salvage is the most widely used, but it suffers from drawbacks such as high cost, low efficiency, and high omission rate. Static physical barrier interception can easily block the migration channels of marine life, even causing accidental capture of organisms, and poses safety hazards to ship navigation. In recent years, with the rapid breakthroughs in artificial intelligence technologies such as visual perception and deep learning, intelligent waste recycling solutions based on unmanned platforms have gradually matured. They overcome the drawbacks of high cost and high randomness of manual salvage, as well as the poor ecological compatibility of passive interception, and have become the optimal development direction for efficient and precise management of nearshore marine debris.

[0004] While current unmanned platform intelligent recycling solutions have initially achieved flexible and autonomous cleanup of marine debris, their global information acquisition capabilities are severely insufficient due to the integration of debris identification and route planning modules into the platform itself and the extremely low observation viewpoint. This results in a small effective search range and a wide blind spot. Furthermore, existing equipment generally neglects the control of sound, light, and electrical pollution in its design. Its propulsion noise, active sonar pulses (depth sounding / obstacle avoidance), high-frequency sensor signals, and lighting can easily disturb surrounding marine life, and propellers and water pumps can easily cause mechanical damage to marine organisms. Therefore, these solutions cannot be applied to ecologically sensitive areas such as marine breeding grounds, habitats of rare and endangered species, coral reefs, seagrass beds, and marine nature reserves.

[0005] Therefore, this invention proposes an air-sea integrated low-disturbance removal system and control method for nearshore marine debris in high-value and ecologically sensitive sea areas, overcoming the shortcomings of existing unmanned platforms such as limited field of vision and poor ecological compatibility, and realizing wide-area accurate search and environmentally friendly recycling of nearshore marine debris. Summary of the Invention

[0006] This invention overcomes the shortcomings of the prior art and provides a method and system for low-disturbance removal and control of nearshore marine debris that combines air and sea operations.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The first aspect of this invention provides a method for low-disturbance removal and control of nearshore marine debris through air-sea coordination, comprising the following steps: By deploying drone airports in a pre-defined area and utilizing the built-in weather stations and marine hydrological interfaces of the drone airports, wind vector and tidal dynamics data of the operational sea area can be obtained in real time. Flight path planning for the UAV is performed based on wind vector and tidal dynamics data of the aforementioned sea area. The system continuously collects ocean surface videos using drones traveling along the flight path, identifies marine debris based on the videos, and obtains the geographic coordinates of the marine debris. Based on the geographic coordinates of the marine debris, a debris-collecting route is planned for the surface debris sweeper; The cleaning effect of the surface garbage sweeper was evaluated, and the sweeper was dynamically adjusted based on the evaluation results. Operation data and logs were collected to build a database of the spatiotemporal evolution of marine debris, predict the trend of garbage accumulation, and optimize the garbage identification model.

[0008] Furthermore, in the air-sea integrated method for the low-disturbance removal and control of nearshore marine debris, unmanned aerial vehicle (UAV) airports are deployed in a pre-defined area, and the wind vector and tidal dynamics data of the operational sea area are obtained in real time using the built-in meteorological stations and marine hydrological interfaces of the UAV airports. Specifically: The drone airport is equipped with a miniature weather station and interfaces for acquiring tidal and hydrological data, and carries a multi-rotor drone with a built-in high-precision real-time dynamic positioning module, laser rangefinder and flight control unit; The multi-rotor drone is equipped with an edge AI computing chip, DDR5 memory and SSD storage, and uses a private network radio and 5G dual communication module to send target coordinates and cleaning routes to the water surface garbage cleaner in real time, and to receive the cleaning status feedback from the cleaner. The drone airport can be deployed in three ways: shipborne, shore-based, or offshore. It can use the built-in meteorological station and marine hydrological interface to obtain real-time wind vector and tidal dynamics data of the operating sea area.

[0009] Furthermore, in the air-sea integrated method for the low-disturbance removal and control of nearshore marine debris, the flight path of the UAV is planned based on the wind vector and tidal dynamics data of the operating area, specifically as follows: Based on the wind vector of the operating sea area, orthogonal decomposition is performed along the direction perpendicular to the coastline to obtain the onshore wind component, and it is determined whether the onshore wind component reaches the preset threshold for easy displacement of marine debris. When the onshore wind component reaches the preset threshold for easy displacement of marine debris, it is confirmed that the current wind force has the characteristics of transporting debris to the shore, and tidal phase analysis is performed at the same time. When it is in the middle of the high tide to the high tide and low tide period, it is confirmed that the hydrological conditions meet the characteristics of nearshore retention of marine debris. The surface wind current velocity induced by the onshore wind component is vector-superimposed with the current tidal current velocity to obtain the comprehensive drift velocity vector of marine debris. When the direction of the overall drift velocity vector of the marine debris points to the preset nearshore salvage area, and the estimated remaining time for the marine debris to reach the salvage area is less than or equal to the preset time threshold, it is determined that the current time is for marine debris search and cleanup, and the drone is used to plan a reciprocating bow-shaped drone flight path.

[0010] Furthermore, in the air-sea integrated method for the low-disturbance removal and control of nearshore marine debris, a drone traveling along its flight path continuously collects sea surface video, identifies marine debris based on the sea surface video, and obtains the geographic coordinates of the marine debris. Specifically: Acquire sea surface videos and construct a sea surface video dataset. Build an initial deep learning network and embed a bidirectional routing attention module in the backbone of the deep learning network.

[0011] Images from the sea surface video dataset are input into the backbone network for feature extraction; a directed graph is constructed based on the extracted features, and the directed graph is input into the bidirectional routing attention module; the bidirectional routing attention module dynamically selects the key-value pairs most relevant to the marine debris target, thereby extracting and generating the deep semantic features of the target.

[0012] An adaptive upsampling module is constructed in the neck network of the deep learning network to replace traditional deconvolution or interpolation upsampling. During feature pyramid fusion in the neck network, the adaptive upsampling module can aggregate contextual information within a larger receptive field and reconstruct kernels based on content prediction, generating richer features.

[0013] The deep semantic features output by the bidirectional routing attention module are concatenated with the shallow texture features output by the adaptive upsampling module at multiple levels. A scale-balanced attention module is introduced into the fusion network to dynamically calculate and allocate fusion weights for feature maps at different levels, outputting high-quality multi-scale fusion features. Based on the multi-scale fusion features and the corresponding real labels, the deep learning network is iteratively trained to obtain a marine debris recognition model.

[0014] Furthermore, in the air-sea integrated nearshore marine debris removal and control method, marine debris in sea surface videos is identified based on the aforementioned debris identification model, and its location information in the image is marked. Based on the location information and combined with UAV pose parameters and ranging data, the geodetic coordinates of the marine debris are calculated. Specifically, this includes: The video is inspected frame by frame using a garbage identification model. When marine debris is identified and the identification confidence level is greater than a preset threshold, the coordinates of the geometric center point of the debris patch in the image pixel coordinate system are extracted. Simultaneously acquire high-precision GNSS position information, real-time attitude data, and vertical height relative to the sea surface measured by the airborne downward-looking lidar recorded by the UAV flight control system; By combining the camera's intrinsic parameter model and distortion parameters, a spatial back intersection ray projection model is constructed. By using the spatial resection ray projection model, the geometric center point in the pixel coordinate system is restored to a three-dimensional spatial ray in the camera coordinate system, and the geometric intersection point of the three-dimensional spatial ray and the sea level is solved by the pose matrix to obtain the geodetic coordinates of the geometric center point of the garbage patch.

[0015] Furthermore, the air-sea integrated method for the low-disturbance removal and control of nearshore marine debris also includes the following steps: Based on the DBSCAN algorithm, density clustering analysis is performed on the geographic coordinates of multiple garbage patches to generate multiple independent garbage aggregation areas. For each of the aforementioned waste accumulation areas, the coordinates of all discrete points within it are extracted to construct a convex hull. The minimum area bounding rectangle of the convex hull is then calculated based on the rotating caliper algorithm. The ratio of the physical length of the major axis to the minor axis of the minimum area bounding rectangle is defined as the aspect ratio of the waste accumulation area. If the aspect ratio is greater than or equal to a preset aspect ratio threshold, the corresponding garbage accumulation area is determined to be a garbage-dense zone. A centerline coordinate sequence that runs through the garbage-dense zone is generated by principal component analysis, and its average physical width is calculated along the normal of the centerline coordinate sequence. If the aspect ratio is less than the preset aspect ratio threshold, the corresponding garbage accumulation area is determined to be a dense garbage cluster. The convex hull algorithm is called to extract the discrete point set that encloses the outer edge of the dense garbage cluster, generate a closed polygon boundary, and calculate the actual coverage area of ​​the closed polygon boundary.

[0016] Furthermore, in the air-sea integrated method for the low-disturbance removal and control of nearshore marine debris, the cleaning effect of surface debris sweepers is evaluated, and the surface debris sweepers are dynamically adjusted based on the evaluation results. Specifically: The system identifies the cleaned garbage areas, calculates the amount of garbage removed and the area of ​​the cleaned areas, compares the coverage area before and after cleaning, and calculates the garbage removal rate. If the garbage removal rate is greater than or equal to the preset removal rate threshold, the cleaning task is confirmed to be completed, a work report is generated, and the drone returns to base autonomously. If the garbage removal rate is less than the preset removal rate threshold, a remediation route will be generated for the remaining garbage, and a secondary cleaning route will be sent to the surface cleaner. The cleaner will then perform secondary remediation cleaning, and after the secondary cleaning is completed, the drone will be triggered for a follow-up inspection. This cycle continues until the garbage removal rate is greater than or equal to the preset removal rate threshold.

[0017] Furthermore, in the air-sea integrated method for the low-disturbance removal and control of nearshore marine debris, operational data and logs are collected to construct a database of the spatiotemporal evolution of marine debris, predict debris enrichment trends, and optimize debris identification models. Specifically, this includes: After receiving the mission termination command from the drone, the water surface cleaner autonomously returns to its home port to perform equipment self-checks and recharge. After completing the mission, the drone autonomously returns to the airport. The command center collects operational data and logs, constructs a database of the spatiotemporal evolution of marine debris, predicts the trend of debris accumulation, and continuously optimizes the debris identification model using data from the database of the spatiotemporal evolution of marine debris.

[0018] The second aspect of the present invention provides an air-sea integrated nearshore marine debris removal and control system with low disturbance, including a memory and a processor. The memory includes an air-sea integrated nearshore marine debris removal and control method with low disturbance. When the program of the air-sea integrated nearshore marine debris removal and control method with low disturbance is executed by the processor, it implements the steps of the air-sea integrated nearshore marine debris removal and control method with low disturbance as described in any one of the claims.

[0019] This invention addresses the shortcomings of the prior art and has the following beneficial effects: Traditional marine debris cleanup equipment typically operates on a fixed schedule, lacking responsiveness to the dynamic aggregation patterns of marine debris. This invention's fully automated unmanned aerial vehicle (UAV) airport system integrates wind field vector orthogonal decomposition and tidal phase analysis. It can calculate the overall drift velocity of debris based on the superposition of surface wind current velocity induced by the onshore wind component and the current tidal current velocity, accurately predicting the window of opportunity for marine debris to reach the shore and automatically triggering UAV patrols. This invention deeply integrates meteorological and hydrodynamic dynamics with UAV equipment scheduling, transforming passive retrieval into active interception, significantly improving debris removal efficiency. Existing marine debris surface sweepers are limited by an extremely low first-person field of view, resulting in large blind spots. Furthermore, after cleaning, large-area, blind re-sweeping is required to ensure a high cleanliness rate, significantly increasing ineffective energy consumption and water disturbance. This invention utilizes the computing power of unmanned aerial vehicles (UAVs) for wide-area search of marine debris and accurately distinguishes between strip-shaped and clump-shaped debris clusters using DBSCAN clustering and convex hull algorithms. It then issues adaptive retrieval routes to the debris sweeper and employs in-situ resurvey and quantitative evaluation mechanisms. When the debris removal rate falls short of the target, the UAV dynamically generates a supplementary route to guide the sweeper in precise secondary operations, effectively reducing the debris omission rate. Existing unmanned vessels often rely on high-frequency active sonar obstacle avoidance, high-power radio frequency radar ranging, and traditional propeller shaft propulsion. The secondary underwater burst noise (typically >100 dB), radio frequency pollution, and propeller mechanical shearing forces generated by these technologies cause serious interference and damage to marine life. This invention uses a shaftless rim propulsion system at the power end to eliminate mechanical noise (background noise <60 dB) and blade entanglement damage; it employs fully passive RTK navigation and low-power communication modes to control electromagnetic pollution; and it uses IP68-rated carbon fiber shielding at the electrical end to prevent leakage current and magnetic flux. This invention provides the first feasible engineering solution for managing marine debris in sensitive marine areas such as marine life breeding grounds, coral reefs, and cetacean habitats. Existing marine debris collection equipment mostly uses rigid metal filters or conveyor belts, which can easily damage marine life that accidentally enters the debris. This invention introduces a microporous aeration bubble wall below the debris collection opening, using the white noise generated by the bubbles and the weak updraft as a physical deterrent signal to warn and drive away schools of fish below. Simultaneously, a conveyor net with densely packed soft silicone teeth and mesh openings improves the efficiency of capturing lightweight plastic debris while avoiding damage to any inadvertently entering organisms. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without any creative effort.

[0021] Fig. 1 The diagram illustrates the overall system architecture and workflow. Fig. 2 An isometric view of a low-disturbance marine debris sweeper is shown; Fig. 3 A schematic diagram of a flexible marine debris collector and a bubble water curtain system is shown. Detailed Implementation

[0022] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0024] like Figs. 1 to 3 As shown, this invention first includes an unmanned aerial vehicle (UAV) system for searching marine debris with onboard computing power and a low-disturbance marine debris cleaner. The UAV system for searching marine debris with onboard computing power includes modules such as a fully automated unmanned airport, a UAV flight platform, an onboard multi-source sensing module, an onboard computing unit, and an air-sea collaborative communication link.

[0025] Among them, the fully automated drone airport is equipped with a miniature weather station (which can detect air temperature, humidity, atmospheric pressure, wind speed, wind direction, rainfall, and light intensity) and an interface for acquiring tidal and hydrological data. It has functions such as automatic opening and closing, drone take-off and landing guidance, wireless / contact automatic charging, and two-way synchronization of mission data.

[0026] Among them, the multi-rotor drone equipped with a high-precision real-time dynamic (RTK) positioning module, laser rangefinder and flight control unit in the drone flight platform has a dustproof and waterproof rating of IP67, a wind resistance of 14 m / s, a payload of 6 kg and a cruise time of 50 min.

[0027] The airborne multi-source sensing module includes a visible light high-definition gimbal camera responsible for real-time acquisition of video streams from the sea surface. The camera has 24 million effective pixels and a focal length of 24 mm.

[0028] The onboard computing unit incorporates an edge artificial intelligence (AI) computing chip, DDR5 memory, and SSD storage. The chip boasts a computing power of 100 TOPS, 32 GB of memory, and 256 GB of storage. It is pre-loaded with an AI model for detecting marine debris, supporting real-time local search and identification of marine debris even in environments with no or weak network connectivity.

[0029] In the air-sea collaborative communication link, a dedicated network radio and a 5G dual communication module are used to send target coordinates and cleaning routes to the surface garbage cleaner in real time, and to receive feedback on the cleaner's operation status.

[0030] like Fig. 2 As shown, the low-disturbance marine debris cleaner includes a silent propeller, a flexible debris collector, a wave-piercing bow, and a hull. The silent propeller utilizes a shaftless rim-driven design, eliminating the need for a drive shaft. The motor and propeller are directly integrated and installed within a ring-shaped duct, preventing entanglement with fishing nets and seaweed, and mitigating mechanical damage to marine life from exposed propellers. It also eliminates the sharp underwater burst noise of traditional propellers, ensuring the cleaner operates at a noise level of <60 dB. Equipped with dual-sided rim-driven propellers, the cleaner achieves zero-turn radius maneuvers through differential motor control, enabling efficient execution of high-frequency turning routes.

[0031] like Fig. 3 As shown, in the flexible debris collector, a microporous aeration pipe is installed below the debris collection opening. During operation, a dense wall of bubbles is sprayed out. The buoyancy of the bubbles gently pushes semi-submersible plastic debris to the surface. The white noise and weak water flow generated by the bubbles serve as physical deterrent signals, warning fish below to avoid the collection opening and preventing accidental capture. The debris conveyor belt surface is covered with soft silicone teeth. Compared to metal filters, silicone teeth can easily pick up lightweight debris such as plastic bottles and foam through friction. Its flexible material ensures that it will not scratch the mucous membranes of marine organisms that accidentally enter the water.

[0032] The hull employs a catamaran structure with a wave-piercing bow to prevent the scattering of debris in adjacent waterways, preserving the original aggregate form of marine debris. All electrical wiring is sealed within an IP68-rated carbon fiber shielded enclosure to prevent leakage and avoid misleading or startling cartilaginous fish (such as sharks and rays) that rely on weak electric fields for sensing. Equipped with low-light night vision cameras and infrared thermal imagers, and except for legally mandated signal lights, it avoids high-powered lighting and lidar to minimize light interference with marine life. A milliwatt-level directional microstrip antenna enables two-way communication with the drone, minimizing electromagnetic pollution. An RTK positioning module guides the cleaning device, executing debris-collecting tasks according to the drone's route.

[0033] The workflow includes steps S1-S8. In step S1: the fully automated unmanned aerial vehicle (UAV) airport deployment, the fully automated UAV airport adopts one of the following three deployment modes or a combination thereof: shipborne mobile deployment, shore-based fixed deployment, and offshore node deployment. In shipborne mobile deployment, the UAV airport is directly installed on the water surface garbage cleaner. The UAV takes off from the cleaner platform as a mobile base, greatly expanding the UAV's operating radius, and can land and recharge at any time, achieving rapid accompanying operations.

[0034] In shore-based fixed deployments, drone airports are fixed at high points near the shore, such as elevated structures, port breakwaters, lighthouses, or river estuaries. Deployment sites must be adjacent to areas with high concentrations of marine debris while avoiding ecologically sensitive terrestrial areas. Relying on shore-based deployments provides access to stable municipal power and fiber optic / 5G networks, enabling high-frequency takeoffs and landings around the clock.

[0035] In offshore deployments, drone airports are attached to existing fixed offshore facilities, such as wind turbine tower work platforms, large aquaculture cage work platforms, or dedicated offshore buoys. Drones can take off and land here to enable relay patrols across sea areas.

[0036] In step S2: During the UAV adaptive takeoff, the fully automated UAV airport acquires real-time wind vector and tidal dynamics data of the operational sea area through its built-in meteorological station and marine hydrological interface, and submits this data to the onboard computing unit. The computing unit orthogonally decomposes the wind field vector along the direction perpendicular to the coastline to obtain the onshore wind component. When this component reaches the marine debris displacement threshold of 2 m / s, it confirms that the current wind force has the characteristics of debris transport to the shore. Simultaneously, tidal phase analysis is performed. When it is in the middle of high tide to the high tide-low tide period, it confirms that the hydrological conditions meet the characteristics of marine debris nearshore retention. Then, the surface wind current velocity (calculated at 4% of the wind speed) caused by the onshore wind component is vector-superimposed with the current tidal current velocity to obtain the comprehensive drift velocity vector of marine debris. When the direction of this vector points to the preset nearshore retrieval area, and the estimated arrival time of the debris on shore is ≤2 hours, the computing unit determines that it is time for marine debris search and cleanup. Subsequently, the computing unit performs reciprocating bow-shaped UAV flight path planning. The flight altitude is 30 m above the sea surface, ensuring that the attenuation of the drone's rotor noise reaching the water surface is less than 50 dB, thus preventing significant noise pollution. The flight speed is 10 m / s. The main flight path is at a 90° angle to the prevailing wind direction, ensuring that the drone can detect densely distributed longitudinal debris zones with the highest probability during each lateral flight. The main flight path spacing is 60 m, achieving a 20% overlap rate between images. The camera gimbal pitch angle is set to -45° (-90° is for direct downward view). If sunlight reflection causes overexposure areas to account for ≥25% of the image (which will cause the AI ​​model to fail), the gimbal pitch angle will be adjusted to -60°, and the drone's cruising speed will be reduced to 6 m / s.

[0037] In step S3: Real-time search, identification, and georeferencing of marine debris, the drone's onboard camera continuously captures video of the sea surface during flight. To balance onboard computing power and real-time performance, the video stream specifications sent to the computing unit are dynamically adjusted to a resolution of 1920×1080 and a frame rate of 10 fps. The onboard AI debris detection model detects the video frame by frame. When the target outline of marine debris is identified with a confidence level ≥80%, the coordinates of the geometric center point of the target in the image pixel coordinate system are extracted.

[0038] Simultaneously acquire high-precision GNSS position information, real-time attitude data, and vertical altitude relative to the sea surface measured by the airborne downward-looking lidar recorded by the UAV flight control system. Combine the camera's intrinsic parameter model and distortion parameters to construct a spatial reintersection ray projection model. Using the spatial reintersection ray projection model, the geometric center point in the pixel coordinate system is restored to a three-dimensional spatial ray in the camera coordinate system. The geometric intersection point of the three-dimensional spatial ray with the sea surface is solved by the pose matrix to obtain the geodetic coordinates of the center point of the garbage patch.

[0039] The main steps in building an AI-powered garbage detection model are as follows: We acquire sea surface videos and construct a sea surface video dataset. We build an initial deep learning network and embed a bidirectional routing attention module into the backbone of the deep learning network.

[0040] Images from the ocean surface video dataset are input into the backbone network for feature extraction. A directed graph is constructed based on the extracted features and then input into a bidirectional routing attention module. This module dynamically selects the key-value pairs most relevant to the marine debris target, thereby extracting and generating deep semantic features of the target. This allows for precise focusing on debris targets in complex ocean backgrounds, effectively suppressing interference from background noise such as seabed and algae. An adaptive upsampling module is constructed in the neck of a deep learning network to replace traditional deconvolution or interpolation upsampling. During feature pyramid fusion in the neck network, the adaptive upsampling module can aggregate contextual information within a larger receptive field and reconstruct kernels based on content prediction, generating richer features.

[0041] The deep semantic features output by the bidirectional routing attention module are concatenated with the shallow texture features output by the adaptive upsampling module at multiple levels. A scale-balanced attention module is introduced into the fusion network to dynamically calculate and allocate the fusion weights of feature maps at different levels, outputting high-quality multi-scale fusion features. This completes the iterative training of the deep learning network and obtains an AI detection model for marine debris.

[0042] It should be noted that the model construction method described in this paper can solve the problem that marine debris varies greatly in scale (such as large fishing nets and tiny fragments) and is easily confused with rocks and seaweed in the background, so that the model can balance the positioning accuracy of large targets and the recognition recall rate of small targets.

[0043] In step S4: clustering and boundary extraction of dense marine debris zones and clusters, the airborne computing unit uses the DBSCAN algorithm (physical neighborhood radius set to 5 m; minimum number of core points in the neighborhood set to 8) to perform density clustering analysis on the discrete absolute geographic coordinates of the mapped debris, generating multiple independent debris clustering areas.

[0044] For each of the aforementioned waste accumulation areas, the coordinates of all discrete points within it are extracted to construct a convex hull. The minimum area bounding rectangle of the convex hull is then calculated based on the rotating caliper algorithm. The ratio of the physical length of the major axis to the minor axis of the minimum area bounding rectangle is defined as the aspect ratio of the waste accumulation area. If the aspect ratio of the aggregated area is ≥3.0, it is determined to be a dense garbage belt formed under the influence of wind, ocean currents, or fronts. Principal component analysis (PCA) is used to generate a coordinate sequence of the centerline running through the belt-like area, and its average physical width is calculated along the normal of the centerline. If the aspect ratio of the aggregated area is <3.0, it is determined to be a dense garbage cluster formed in a eddy or stagnant water zone. In this case, the airborne computing unit calls the convex hull algorithm to extract the discrete point set enveloping the outer edge of the cluster, generates a closed polygon boundary (polygon vertex coordinate sequence), and calculates the actual coverage area of ​​the polygon in real time based on the WGS84 ellipsoid geometric projection area formula.

[0045] In step S5, the low-disturbance cleaning route planning and distribution for the surface platform, the airborne computing module plans a garbage cleaning route for the surface garbage cleaner based on the boundary information extracted in S4. For densely packed garbage areas, a reciprocating route is planned using the extracted strip centerline as the reference axis, with a route spacing of 1.8 m, a track overlap rate of 20%, and a travel speed of 1 m / s. For densely packed garbage clusters, because garbage clumps are easily dispersed outwards due to platform bow wave disturbances, an inward spiral-centered contraction route is planned. The outermost boundary of the garbage cluster polygon extracted in S4 is used as the initial retrieval circle, and the route spirals clockwise towards the geometric center point of the polygon. The spiral circle spacing is 1.8 m, and the travel speed is 0.8 m / s. After the route is generated, it is directly distributed to the surface garbage cleaner via a milliwatt-level directional antenna in the form of a track point sequence through the aircraft-sea collaborative communication link. The UAV then returns to base.

[0046] In step S6: During the surface cleaning operation, the surface cleaner receives the route information from the UAV via its directional antenna. Its onboard RTK module provides centimeter-level self-positioning and heading data, controlling the dual-sided shaftless rim propellers to navigate to the starting point of the route. The cleaner then activates its bow microporous aeration pipe to create a fish-repelling bubble wall and starts the flexible silicone toothed conveyor belt, strictly following the planned route to perform marine debris retrieval. Throughout the operation, the cleaner maintains a low noise level of <60 dB and zero radio frequency leakage. The cleaner monitors the weight of the captured debris in real time using a gravity sensor built into the bottom of the collection compartment. Once the full load threshold is reached, the operation stops, and the cleaner autonomously returns to the preset debris unloading point. After completing the route cleaning task, the cleaner sends a completion signal to the UAV.

[0047] In step S7: UAV re-fly inspection and cleaning effect evaluation, the UAV re-flies after receiving the completion signal and performs an in-situ inspection. The onboard model re-identifies and calculates the area of ​​the cleaned zone. The coverage area before and after cleaning is compared to calculate the garbage removal rate. If the garbage removal rate is ≥90%, the cleaning task is confirmed as complete, a work report is generated, and the UAV autonomously returns to base. If the garbage removal rate is <90%, the UAV's onboard model generates a remedial route for the remaining garbage and issues a secondary cleaning route to the surface cleaner, which performs secondary remedial cleaning. After the secondary cleaning is completed, the UAV is triggered for a second inspection, and this cycle continues until the garbage removal rate is ≥90%.

[0048] In step S8, the return to port and the construction of the marine debris feature database, after receiving the mission termination command from the drone, the surface cleaner autonomously returns to its home port to perform equipment self-checks and recharge. After completing its mission, the drone autonomously returns to the airport. The command center collects operational data and logs to construct a database of the spatiotemporal evolution of marine debris, predicting debris accumulation trends, optimizing debris identification models, and improving the effectiveness of subsequent operations.

[0049] It should be noted that this invention overcomes the technical bottlenecks of existing marine debris cleanup platforms, such as weak global perception capabilities and large search blind spots due to low observation viewpoints, and the potential for disturbance and damage to organisms caused by the sound, light, electricity, and mechanical operations generated during equipment operation. It provides a low-disturbance nearshore marine debris removal system and control method for high-value and ecologically sensitive sea areas, integrating air and sea operations. By using unmanned aerial vehicles (UAVs) for wide-area marine debris perception and surface platform scheduling, it achieves efficient search and precise planning of retrieval routes for marine debris under air-sea coordination. Simultaneously, through systematic acoustic, optical, and electrical control and a non-destructive, flexible recovery structure design, it eliminates interference and damage to marine life and their habitats, achieving large-scale, efficient removal and environmentally friendly recycling of nearshore marine debris. This provides equipment and technical support for the ecological governance of high-value and ecologically sensitive sea areas.

[0050] In addition, this method also includes: By simultaneously acquiring video frame data and registering RGB and polarized light images, a two-stream neural network is constructed. One stream processes the RGB image for initial candidate box detection; the other stream processes the polarized light image, introducing a feature fusion and confidence recalibration module. This module maps the candidate boxes detected by the RGB stream onto the polarization feature map, calculates the spectral angle similarity of the candidate box regions, and determines whether the spectral angle similarity of the candidate box regions is greater than a preset similarity threshold. If the spectral angle similarity of the candidate box regions is greater than the preset similarity threshold, its detection confidence is increased; if the spectral angle similarity of the candidate box regions is not greater than the preset similarity threshold, the confidence is decreased or the candidate box is directly filtered out. By combining video temporal information, optical flow is used to track the position of the same target in consecutive frames. For targets that appear only in a single frame and disappear in the frames before and after, a deletion operation is performed.

[0051] It should be noted that this method solves the technical problem that relying solely on RGB visual images is easily affected by water bubbles, sunlight flares (light stripes), and white waves, resulting in false detections, thereby further improving the robustness of garbage identification and detection.

[0052] In addition, this method also includes: Collect target image data of marine debris in open environments, and collect real underwater background images with different turbidity and lighting conditions, and introduce image blending technology; The image blending technique described above uses the Poisson image editing algorithm to seamlessly embed foreground debris into an underwater background image. By solving the Poisson equation, a smooth transition at the boundary is achieved, generating a preliminary synthetic image. Generative adversarial network is constructed, with the initial synthetic image as the source domain and the real underwater debris image as the target domain. Through cycle consistency loss, the generator learns the style of underwater images, including blue-green tones, scattering blur effects, and spectral changes caused by water absorption. The trained generator is used to perform style transfer on the initial synthetic images to generate a high-quality augmented dataset with diversity, and the garbage recognition model is continuously trained using the high-quality augmented dataset.

[0053] It should be noted that this method can solve the problems of high cost and small sample size in acquiring real underwater debris annotation data, as well as color distortion and low contrast in real-world scenarios. Moreover, through transfer learning, the synthetic data can more closely resemble the physical processes of real underwater imaging in terms of visual features, thereby improving the recognition accuracy of the debris identification model.

[0054] In summary, the fully automated UAV airport system of this invention integrates wind field vector orthogonal decomposition and tidal phase analysis. It can calculate the comprehensive drift velocity of marine debris based on the surface wind current velocity induced by the onshore wind component and the current tidal current velocity, accurately predicting the window period for marine debris to reach the shore and automatically triggering UAV patrol routes. This invention deeply integrates meteorological hydrodynamics with UAV equipment scheduling, transforming passive retrieval into active interception, significantly improving debris removal efficiency. Existing marine debris surface cleaners are limited by a very low first-person perspective, resulting in large blind spots. Furthermore, after cleaning, large-area, blind re-sweeping is required to ensure a cleanliness rate, significantly increasing ineffective energy consumption and water disturbance. This invention utilizes the computing power of UAVs for wide-area search of marine debris and accurately distinguishes between strip-shaped and clump-shaped debris clusters through DBSCAN clustering and convex hull algorithms. Adaptive retrieval routes are issued to the debris cleaners, and an in-situ resurvey and quantitative evaluation mechanism is used. When the debris removal rate does not meet the standard, the UAV dynamically generates a supplementary route to guide the cleaners for precise secondary operations, effectively reducing the debris omission rate. Existing unmanned surface vessels (USVs) often rely on high-frequency active sonar obstacle avoidance, high-power radio frequency radar ranging, and traditional propeller shaft propulsion. The secondary underwater burst noise (typically >100 dB), radio frequency pollution, and propeller mechanical shearing forces generated by these systems cause serious interference and damage to marine life. This invention employs a shaftless rim propulsion system at the power end, eliminating mechanical noise (background noise <60 dB) and blade entanglement damage; it uses fully passive RTK navigation and low-power communication modes to control electromagnetic pollution; and the electrical system uses IP68-rated carbon fiber shielding to prevent leakage current and magnetic flux. This provides the first feasible engineering solution for the management of marine debris in sensitive marine areas such as marine breeding grounds, coral reefs, and cetacean habitats. Existing marine debris collection equipment mostly uses rigid metal filters or conveyor belts, which can easily damage marine life that accidentally enters the debris. This invention introduces a microporous aeration bubble wall below the garbage collection opening, using the white noise generated by the bubbles and the weak updraft as a physical deterrent signal to warn and drive away fish below in advance; at the same time, it is combined with a conveyor net with dense soft silicone protrusions and mesh, which improves the capture efficiency of lightweight plastic waste while avoiding damage to accidentally entered organisms.

[0055] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0056] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0057] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0058] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0059] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0060] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for low-disturbance removal and control of nearshore marine debris using an air-sea integrated approach, characterized in that, Includes the following steps: By deploying drone airports in a pre-defined area and utilizing the built-in weather stations and marine hydrological interfaces of the drone airports, real-time wind vector and tidal dynamics data of the operational sea area can be obtained. Flight path planning for the UAV is performed based on wind vector and tidal dynamics data of the aforementioned sea area. The system continuously collects ocean surface videos using drones traveling along the flight path, identifies marine debris based on the videos, and obtains the geographic coordinates of the marine debris. The plan for the garbage cleaning route of the surface garbage sweeper is based on the geographical coordinates of the marine debris. The cleaning effect of the surface garbage sweeper was evaluated, and the sweeper was dynamically adjusted based on the evaluation results. Operation data and logs were collected to build a database of the spatiotemporal evolution of marine debris, predict the trend of garbage accumulation, and optimize the garbage identification model.

2. The method for low-disturbance removal and control of nearshore marine debris using an air-sea integrated approach as described in claim 1, characterized in that, By deploying drone airports in a pre-defined area and utilizing the built-in weather stations and marine hydrological interfaces of the drone airports, real-time wind vector and tidal dynamics data of the operational sea area are obtained. Specifically: The UAV airport is equipped with a miniature weather station and interfaces for acquiring tidal and hydrological data, and is equipped with a multi-rotor UAV with a built-in high-precision real-time dynamic positioning module, laser rangefinder and flight control unit. The multi-rotor drone is equipped with an edge AI computing chip, DDR5 memory and SSD storage, and uses a private network radio and 5G dual communication module to send target coordinates and cleaning routes to the water surface garbage cleaner in real time, and to receive the cleaning status feedback from the cleaner. The UAV airport can be deployed in a ship-borne, shore-based, or offshore node manner, and can use the built-in meteorological station and marine hydrological interface of the UAV airport to obtain wind vector and tidal dynamics data of the operating sea area in real time.

3. The method for low-disturbance removal and control of nearshore marine debris using an air-sea integrated approach as described in claim 1, characterized in that, Based on the wind vector and tidal dynamics data of the aforementioned sea area, the flight path of the UAV is planned as follows: Based on the wind vector of the operating sea area, orthogonal decomposition is performed along the direction perpendicular to the coastline to obtain the onshore wind component, and it is determined whether the onshore wind component reaches the preset threshold for easy displacement of marine debris. When the onshore wind component reaches the preset threshold for easy displacement of marine debris, it is confirmed that the current wind force has the characteristics of transporting debris to the shore, and tidal phase analysis is performed at the same time. When it is in the middle of the high tide to the high tide and low tide period, it is confirmed that the hydrological conditions meet the characteristics of nearshore retention of marine debris. The surface wind current velocity induced by the onshore wind component is vector-superimposed with the current tidal current velocity to obtain the comprehensive drift velocity vector of marine debris. When the direction of the overall drift velocity vector of the marine debris points to the preset nearshore retrieval area, and the estimated time for the debris to reach the shore is less than a threshold, it is determined that the current time is for marine debris search and cleanup, and the drone is used to plan a reciprocating bow-shaped drone flight path.

4. The method for low-disturbance removal and control of nearshore marine debris using an air-sea integrated approach as described in claim 1, characterized in that, The drone continuously collects sea surface video along its flight path, identifies marine debris based on the video, and obtains the geographic coordinates of the marine debris. Specifically: Acquire sea surface videos and construct a sea surface video dataset, build an initial deep learning network, and embed a bidirectional routing attention module into the backbone network of the deep learning network; Images from the sea surface video dataset are input into the backbone network for feature extraction; a directed graph is constructed based on the extracted features, and the directed graph is input into the bidirectional routing attention module; The bidirectional routing attention module dynamically filters out the key-value pairs most relevant to the marine debris target, thereby extracting and generating the deep semantic features of the target. An adaptive upsampling module is constructed in the neck network of the deep learning network to replace the traditional deconvolution or interpolation upsampling. When the neck network performs feature pyramid fusion, the adaptive upsampling module can aggregate contextual information within a larger receptive field and reconstruct the kernel based on content prediction to generate richer features. The deep semantic features output by the bidirectional routing attention module are multi-layeredly concatenated with the shallow texture features output by the adaptive upsampling module; and a scale-balanced attention module is introduced into the fusion network to dynamically calculate and allocate the fusion weights of feature maps at different levels, outputting high-quality multi-scale fusion features, thereby completing the iterative training of the deep learning network and obtaining a marine debris recognition model. Based on the aforementioned garbage identification model, garbage in the sea surface video is identified, and the location information of the garbage is marked, generating the geographic coordinates of the garbage.

5. The method for low-disturbance removal and control of nearshore marine debris using an air-sea integrated approach according to claim 4, characterized in that, Based on the aforementioned litter identification model, litter in marine surface videos is identified, and the location information of the litter is marked to generate the geographic coordinates of marine litter patches, specifically including: The video is detected frame by frame by the garbage identification model. When marine debris is identified and the identification confidence is greater than a preset threshold, the coordinates of the geometric center point of the garbage patch in the image pixel coordinate system are extracted. Simultaneously acquire high-precision GNSS position information, real-time attitude data, and vertical height relative to the sea surface measured by the airborne downward-looking lidar recorded by the UAV flight control system; By combining the camera's intrinsic parameter model and distortion parameters, a spatial back intersection ray projection model is constructed. By using the spatial resection ray projection model, the geometric center point in the pixel coordinate system is restored to a three-dimensional spatial ray in the camera coordinate system, and the geometric intersection point of the three-dimensional spatial ray and the sea level is solved by the pose matrix to obtain the geodetic coordinates of the geometric center point of the garbage patch.

6. The method for low-disturbance removal and control of nearshore marine debris using an air-sea integrated approach according to claim 5, characterized in that, It also includes the following steps: Based on the DBSCAN algorithm, density clustering analysis is performed on the geodetic coordinates of the geometric center points of multiple garbage patches to generate multiple independent garbage aggregation areas. For each of the aforementioned waste accumulation areas, the coordinates of all discrete points within it are extracted to construct a convex hull. The minimum area bounding rectangle of the convex hull is then calculated based on the rotating caliper algorithm. The ratio of the physical length of the major axis to the minor axis of the minimum area bounding rectangle is defined as the aspect ratio of the waste accumulation area. If the aspect ratio is greater than or equal to a preset aspect ratio threshold, the corresponding garbage accumulation area is determined to be a garbage-dense zone. A centerline coordinate sequence that runs through the garbage-dense zone is generated by principal component analysis, and its average physical width is calculated along the normal of the centerline coordinate sequence. If the aspect ratio is less than the preset aspect ratio threshold, the corresponding garbage accumulation area is determined to be a dense garbage cluster. The convex hull algorithm is called to extract the discrete point set that encloses the outer edge of the dense garbage cluster, generate a closed polygon boundary, and calculate the actual coverage area of ​​the closed polygon boundary.

7. The method for low-disturbance removal and control of nearshore marine debris using an air-sea integrated approach as described in claim 6, characterized in that, Based on the geographic coordinates of the marine debris, a debris-collecting route is planned for the surface debris sweeper, specifically including: Based on the boundary information formed by the multiple independent garbage accumulation areas, for the garbage-dense zone, a reciprocating garbage cleaning route is planned with the extracted strip center line as the reference axis. For densely packed waste groups, a spiral-shaped, centripetal contraction route is planned. The outermost boundary of the densely packed waste group is used as the initial retrieval circle, and the route spirals and contracts clockwise layer by layer toward the geometric center of the polygon to generate a waste cleaning route. After the garbage cleaning route is generated, it is transmitted directly to the surface garbage cleaner in the form of a track point sequence via the air-sea collaborative communication link and a milliwatt-level directional antenna, and the surface garbage cleaner is used to carry out the cleaning operation.

8. The method for low-disturbance removal and control of nearshore marine debris using an air-sea integrated approach according to claim 1, characterized in that, The cleaning effect of the water surface garbage sweeper is evaluated, and the sweeper is dynamically adjusted based on the evaluation results. Specifically: The system identifies the cleaned garbage areas, calculates the amount of garbage removed and the area of ​​the cleaned areas, compares the coverage area before and after cleaning, and calculates the garbage removal rate. If the garbage removal rate is greater than or equal to the preset removal rate threshold, the cleaning task is confirmed to be completed, a work report is generated, and the drone returns to base autonomously. If the garbage removal rate is less than the preset removal rate threshold, a replenishment route will be generated for the remaining garbage, and a secondary cleaning route will be sent to the water surface cleaner. The cleaner will perform secondary replenishment cleaning. After the secondary cleaning is completed, the drone will be triggered to re-inspect, and the cycle will continue until the garbage removal rate is greater than or equal to the preset removal rate threshold.

9. A method for low-disturbance removal and control of nearshore marine debris using an air-sea integrated approach according to claim 1, characterized in that, Collect operational data and logs to construct a database of the spatiotemporal evolution of marine debris, predict debris enrichment trends, and optimize debris identification models. Specifically, this includes: After receiving the mission termination command from the UAV, the water surface cleaner autonomously returns to its home port to perform equipment self-checks and recharge. After completing its mission, the UAV autonomously returns to the UAV airport; the command center collects operational data and logs, constructs a spatiotemporal evolution database of marine debris, predicts debris enrichment trends, and continuously optimizes the debris identification model using data from the spatiotemporal evolution database of marine debris.

10. A low-disturbance nearshore marine debris removal and control system integrating air and sea operations, characterized in that, The device includes a memory and a processor. The memory includes a program for a low-disturbance removal and control method for nearshore marine debris that is integrated with air and sea operations. When the processor executes the program for the low-disturbance removal and control method for nearshore marine debris that is integrated with air and sea operations, it implements the steps of the low-disturbance removal and control method for nearshore marine debris that is integrated with air and sea operations as described in any one of claims 1-9.

Images