Ship bottom fouling curve monitoring method based on three-dimensional reconstruction and adaptive scanning

Abstract

The application provides a ship bottom fouling organism curved surface monitoring method based on three-dimensional reconstruction and adaptive scanning, comprising the following steps: collecting environmental parameters around the ship bottom; enhancing the recovered underwater image based on a multi-modal large model, and adaptively adjusting the strategy according to the environmental parameters; planning a path based on an environmental change prediction model, adjusting the priority, and supplementing a blocked area bypass path to generate an initial model; analyzing the image and the model, coarsely segmenting the fouling organisms and determining the type, combining the environmental parameters to output optimal scanning parameters, and generating optimized data; updating the three-dimensional model, preliminarily classifying the fouling organisms by using a large model, and then finely identifying by using a retrieval enhancement generation technology combined with a multi-modal large model. The method combines multi-dimensional environmental parameters to improve image quality, dynamically adjusts the path to avoid data loss, constructs a complete processing chain, and realizes efficient scanning and accurate identification.

Description

Technical Field

[0001] This invention relates to the field of ship bottom fouling monitoring technology, specifically providing a method for monitoring bio-surface fouling on the ship bottom based on three-dimensional reconstruction and adaptive scanning. Background Technology

[0002] When ships navigate or anchor in the marine environment, their hulls are prone to fouling organisms such as barnacles, algae, and shellfish. These organisms increase hull drag, corrode the hull, and affect the accuracy of navigation equipment. Traditional methods for monitoring hull fouling have many limitations, such as low efficiency and high risk of manual inspection, difficulty in adapting simple two-dimensional scanning to the complex curved surface of the hull leading to data distortion, and poor image quality and difficulty in biometric identification in complex underwater environments such as low light and high turbidity, which cannot meet the needs of accurate monitoring.

[0003] Therefore, there is an urgent need for a method for monitoring biofouling on the bottom of ships based on three-dimensional reconstruction and adaptive scanning to solve the above problems. Summary of the Invention

[0004] In order to overcome the above-mentioned defects, the present invention is proposed to provide a solution or partial solution to the aforementioned problems.

[0005] This invention provides a method for monitoring biofouling on the bottom of a ship based on 3D reconstruction and adaptive scanning, comprising: collecting environmental parameters around the ship's bottom, including image contrast, noise intensity, laser reflectivity fluctuation, water flow velocity, water turbidity, and chlorophyll content; enhancing and restoring underwater images using a multimodal large model based on the environmental parameters, wherein the multimodal large model adaptively adjusts the image processing strategy according to real-time environmental parameters; performing path planning based on a constructed environmental change prediction model, adjusting path priority or order according to regional environmental change trends; and simultaneously supplementing and bypassing data-missing areas caused by occlusion by incorporating dynamic information from the 3D reconstruction process. The process involves: 1) completing the scan and generating an initial model; 2) analyzing the enhanced underwater image and the initial 3D model to output coarse segmentation results of fouling organisms, determining the type of fouling organisms in the current scanned area; 3) learning environmental change patterns based on the type of fouling organisms and environmental parameters in the current scanned area, outputting the optimal combination of scanning parameters in real time, and generating optimized scan data; 4) updating the 3D model based on the optimized scan data, and simultaneously performing preliminary classification of fouling organisms in the underwater image associated with the updated 3D model based on a large model with general segmentation capabilities or zero-sample / few-sample learning capabilities; 5) using retrieval enhancement generation technology combined with a multimodal large model to perform fine recognition based on the preliminary classification results.

[0006] In one technical solution of the above-mentioned method for monitoring biofouling on the bottom of ships based on three-dimensional reconstruction and adaptive scanning, the process of enhancing and restoring underwater images using a multimodal large model based on the environmental parameters includes: the multimodal large model receiving underwater images and the environmental parameters, and analyzing the image distortion features caused by low illumination and high turbidity in the environmental parameters; based on the analysis results, autonomously selecting appropriate image processing strategies, including defogging, color correction, brightness optimization, and super-resolution reconstruction; during the processing, receiving updated environmental parameters in real time, dynamically adjusting the intensity and parameter ratio of each processing strategy, and finally outputting the enhanced and restored underwater image.

[0007] In one of the above-mentioned technical solutions for the monitoring method of biofouling on the bottom of ships based on three-dimensional reconstruction and adaptive scanning, the method further includes constructing an environmental change prediction model. Specifically, based on historical ship navigation data, marine environment database, real-time received marine forecasts and the environmental parameters, a time-series prediction algorithm is used to construct the environmental change prediction model.

[0008] In one technical solution of the above-mentioned method for monitoring biofouling surfaces on the bottom of ships based on 3D reconstruction and adaptive scanning, the process of performing path planning based on the constructed environmental change prediction model and adjusting the path priority or order according to the regional environmental change trend, and simultaneously combining the dynamic information in the 3D reconstruction process to supplement detour paths for data-missing areas caused by occlusion, and completing the scanning and generating the initial model includes: outputting the regional environmental change trend based on the environmental change prediction model, analyzing the regional environmental change trend, clarifying the differences in environmental stability of each region, setting regions with poor environmental stability and prone to drastic changes as high priority, prioritizing the planning of scanning paths for these regions, and reducing the priority and adjusting the scanning order for environmentally stable regions; during the scanning process, acquiring the dynamic information of 3D reconstruction in real time, identifying data-missing areas caused by occlusion through the dynamic information, and planning supplementary scanning paths that can bypass the occlusions based on the spatial location of the missing areas and the shape of the occlusions; completing the scanning of all regions according to the adjusted priority paths and supplementary detour paths, integrating all scanning data, constructing and outputting the initial model.

[0009] In one of the technical solutions of the above-mentioned method for monitoring biofouling on the bottom of a ship based on three-dimensional reconstruction and adaptive scanning, the process of learning the environmental change pattern based on the biofouling type and environmental parameters of the current scanning area and outputting the optimal combination of scanning parameters in real time includes: taking the environmental parameters and biofouling type as environmental state inputs, training the agent with three-dimensional reconstruction accuracy, biofouling identification accuracy and scanning efficiency as joint reward functions, and using the agent to output the optimal combination of scanning parameters in real time to perform targeted adjustments and generate optimized scanning data.

[0010] In one technical solution of the above-mentioned method for monitoring biofouling surfaces on the bottom of ships based on 3D reconstruction and adaptive scanning, during the scanning process, dynamic information of 3D reconstruction is acquired in real time. The process of identifying data gaps caused by occlusion using this dynamic information and planning supplementary scanning paths that bypass the occlusions based on the spatial location of the gaps and the shape of the occluders includes: real-time acquisition and analysis of the dynamic data from the 3D reconstruction to extract spatial structure information; identification of data gaps caused by occlusions by comparing the expected shape of the complete scene with the current reconstruction results; spatial positioning of the gaps while analyzing the geometric shape and distribution characteristics of the occluders; and generation of supplementary scanning paths that bypass the occluders using a path planning algorithm based on the coordinate information of the gaps and the physical properties of the occluders.

[0011] In one technical solution of the above-mentioned method for monitoring biofouling on the bottom of a ship based on 3D reconstruction and adaptive scanning, the process of generating a supplementary scanning path that can bypass the obstructions based on the coordinate information of the missing area and the physical properties of the obstructions using a path planning algorithm includes: converting the coordinates of the missing area and the physical properties of the obstructions into a 3D spatial model, clarifying the area to be scanned and the obstacles that must be avoided; setting path constraints according to the motion capability of the scanning equipment; using a path planning algorithm to search for an initial path from the current position to the missing area that can bypass the obstructions; optimizing the path to make it smoother and more efficient, and verifying whether it can completely cover the missing area, and finally generating an executable supplementary scanning path.

[0012] In one of the technical solutions of the above-mentioned method for monitoring biological surfaces of ship bottom fouling based on three-dimensional reconstruction and adaptive scanning, after the path planning, parameter initialization and scanning are completed, the method further includes: dynamically adjusting the scanning parameters, data acquisition method and data processing algorithm in response to real-time fluctuations in equipment status.

[0013] In one technical solution of the above-mentioned method for monitoring biofouling on the bottom of ships based on 3D reconstruction and adaptive scanning, the process of dynamically adjusting scanning parameters, data acquisition methods, and data processing algorithms in response to real-time fluctuations in equipment status includes: establishing a collaborative calibration mechanism to incorporate equipment status into the additional dimension of environmental interference; real-time monitoring of equipment status parameters to identify abnormal equipment conditions; initiating corresponding dynamic compensation strategies according to different abnormality types: when laser power attenuation is detected, the scanning frequency is automatically increased to increase data redundancy, while the point cloud denoising algorithm is strengthened to reduce noise caused by insufficient power; when lens fogging is detected, the image is switched to a laser + structured light dual-mode scanning mode to use laser data to compensate for the lack of visual data; and the compensation effect is continuously tracked, and compensation parameters are dynamically adjusted to adapt to changes in equipment status.

[0014] In one technical solution of the above-mentioned method for monitoring biofouling surfaces on the bottom of ships based on 3D reconstruction and adaptive scanning, the process of performing fine identification using retrieval-enhanced generation technology combined with a multimodal large model for the preliminary classification results includes: using a biofouling feature database as a knowledge base; based on the range of biological categories locked by the preliminary classification, retrieving reference images, morphological feature descriptions, and classification criteria under that category from the knowledge base through a retrieval mechanism, and performing targeted comparisons with the images of the organisms to be identified to select the most similar candidate information; using a multimodal large model as the core inference engine, receiving the images of the organisms to be identified and the retrieved candidate information, analyzing the detailed features of the organisms in the images through visual understanding capabilities, including the branching morphology of algae and the shell texture structure of barnacles, and combining logical reasoning capabilities to match and verify the candidate information, finally outputting the specific species identification result of the biofouling organism.

[0015] The beneficial effects of the method for monitoring fouling organisms on the hull surface based on 3D reconstruction and adaptive scanning provided by this invention are as follows: This method fully integrates the multi-dimensional parameter characteristics of complex underwater environments, including key indicators such as image contrast, noise intensity, and laser reflectivity fluctuations. Through the adaptive adjustment strategy of a multimodal large model, it can more efficiently enhance and restore underwater images, significantly improving the quality and usability of image data. In addition, by dynamically adjusting the scanning path priority through an environmental change prediction model and supplementing detour paths for occluded areas, it ensures that the scanning process can adapt to environmental dynamics and completely cover the target area, effectively avoiding data loss problems caused by sudden environmental changes or occlusion. Finally, a complete processing chain is constructed, from coarse segmentation of fouling organisms and optimization of scanning parameters to 3D model updates and fine identification, forming an identification system suitable for different underwater environments and fouling types, which can accurately achieve efficient scanning and precise identification of fouling organisms on the hull. Attached Figure Description

[0016] The disclosure of this invention will become more readily understood with reference to the accompanying drawings. It will be readily understood by those skilled in the art that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. Furthermore, similar numbers in the drawings are used to denote similar components, wherein:

[0017] Figure 1 This is a method for monitoring biofouling on the bottom of a ship based on three-dimensional reconstruction and adaptive scanning, according to an embodiment of the present invention. Detailed Implementation

[0018] Some embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0019] like Figure 1As shown, an embodiment of the present invention provides a method for monitoring biofouling on the bottom of a ship based on three-dimensional reconstruction and adaptive scanning, which mainly includes the following steps S1-S5.

[0020] Step S1: Collect environmental parameters around the bottom of the ship, including image contrast, noise intensity, laser reflectivity fluctuation, water flow velocity, water turbidity, and chlorophyll content.

[0021] Step S2: Based on the environmental parameters, a multimodal large model is used to enhance and restore the underwater image. The multimodal large model adaptively adjusts the image processing strategy according to the real-time environmental parameters.

[0022] Step S3: Based on the constructed environmental change prediction model, perform path planning and adjust the path priority or order according to the regional environmental change trend; at the same time, combine the dynamic information in the 3D reconstruction process to supplement detour paths for data missing areas caused by occlusion, complete the scan and generate the initial model.

[0023] Step S4: Analyze the enhanced underwater image and the initial 3D model, output coarse segmentation results of fouling organisms, and determine the type of fouling organisms in the current scanning area; based on the type of fouling organisms and environmental parameters in the current scanning area, learn the environmental change pattern, output the optimal combination of scanning parameters in real time, and generate optimized scanning data;

[0024] Step S5: Update the 3D model based on the optimized scan data. At the same time, based on a large model with general segmentation capabilities or zero-sample / few-sample learning capabilities, perform preliminary classification of fouling organisms in the underwater images associated with the updated 3D model. Based on the preliminary classification results, use retrieval enhancement generation technology combined with a multimodal large model to perform fine recognition.

[0025] In one embodiment, step S2, the process of enhancing and restoring the underwater image using a multimodal large model based on the environmental parameters, includes:

[0026] Step S21: The multimodal large model receives underwater images and environmental parameters, and analyzes the image distortion features caused by low illumination and high turbidity in the environmental parameters;

[0027] Step S22: Based on the analysis results, autonomously select an appropriate image processing strategy, including dehazing, color correction, brightness optimization, and super-resolution reconstruction;

[0028] Step S23: During the processing, the updated environmental parameters are received in real time, and the intensity and parameter ratio of various processing strategies are dynamically adjusted to finally output the enhanced and restored underwater image.

[0029] In this embodiment, the image distortion features analyzed in step S21 may include not only those caused by low illumination and high turbidity, but also image blurring caused by water flow disturbance and light spot interference caused by uneven laser reflection; the image processing strategy in step S22 may also include noise reduction processing, edge enhancement, etc., to deal with different image quality problems; the parameter ratio dynamically adjusted in step S23 can be flexibly set according to the specific environmental parameter change range. For example, when the turbidity of the water increases sharply, the intensity ratio of the defogging strategy can be increased.

[0030] In one embodiment, the method further includes constructing an environmental change prediction model. Specifically, based on historical ship navigation data, a marine environment database, real-time received marine forecasts, and the environmental parameters, an environmental change prediction model is constructed using a time-series prediction algorithm.

[0031] In this embodiment, the ship's historical navigation data includes past routes, speeds, and environmental conditions corresponding to the navigation periods; the marine environment database covers long-term monitoring data such as water temperature, salinity, and tidal currents in different sea areas; the real-time received marine forecasts include information such as wind force, wind direction, and wave size for a future period; the environmental parameters, as mentioned above, include image contrast and water flow velocity; the temporal prediction algorithms used include Long Short-Term Memory (LSTM), Temporal Convolutional Network (TCN), and the Prophet algorithm. Of course, these are not limited to the scenarios listed above. For example, the ship's historical navigation data may also include information affecting navigation status such as the ship's cargo load; the marine environment database may also include marine biological distribution data; and the temporal prediction algorithm may be optimized by incorporating attention mechanisms. Those skilled in the art can adjust and expand upon these features according to actual needs.

[0032] In one embodiment, step S3, performing path planning based on the constructed environmental change prediction model, adjusting path priority or order according to regional environmental change trends; and simultaneously combining dynamic information from the 3D reconstruction process to supplement detour paths for data-missing areas caused by occlusion, completing the scanning and generating the initial model, includes:

[0033] Step S31: Based on the environmental change prediction model, output the regional environmental change trend, analyze the regional environmental change trend, clarify the differences in environmental stability of each region, set the regions with poor environmental stability and prone to drastic changes as high priority, prioritize the planning of scanning paths, reduce the priority of environmentally stable regions and adjust the scanning order.

[0034] Step S32: During the scanning process, acquire dynamic information of 3D reconstruction in real time, identify data missing areas caused by occlusion through the dynamic information, and plan supplementary scanning paths that can bypass the occlusions based on the spatial location of the missing areas and the shape of the occlusions.

[0035] Step S33: Complete the scanning of the entire area according to the adjusted priority path and the supplementary detour path, integrate all scan data, and construct and output the initial model.

[0036] In this embodiment, the criteria for judging the stability of the regional environment in step S31 may include the magnitude and frequency of changes in environmental parameters, and high-priority areas such as areas of sudden changes in water flow velocity and areas of rapid increase in water turbidity; the dynamic information of the 3D reconstruction in step S32 includes real-time point cloud data and model reconstruction progress, and the shape of occlusion objects can be divided into regular shapes (such as protruding ship bottom structures) and irregular shapes (such as floating objects); the integration method of the scanned data in step S33 includes point cloud registration, data deduplication, etc., and the output format of the initial model may include 3D mesh model, point cloud model, etc. Of course, these contents are not limited to the situations listed above. For example, priority division can also be combined with the importance of the scanned area, and the detour path planning can also consider the equipment motion constraints. Those skilled in the art can flexibly adjust according to the actual scenario.

[0037] In one embodiment, step S4, learning the environmental change patterns based on the type of contaminated organisms and environmental parameters in the current scanning area, and outputting the optimal combination of scanning parameters in real time, includes: taking the environmental parameters and type of contaminated organisms as environmental state inputs, training the agent with 3D reconstruction accuracy, contamination identification accuracy, and scanning efficiency as joint reward functions, and using the agent to output the optimal combination of scanning parameters in real time to perform targeted adjustments and generate optimized scanning data.

[0038] In one embodiment, step S32, during the scanning process, involves acquiring dynamic information of the 3D reconstruction in real time, identifying data loss areas caused by occlusion using the dynamic information, and planning supplementary scanning paths that can bypass the occlusions based on the spatial location of the loss areas and the shape of the occlusions.

[0039] Step S321: Collect and analyze the dynamic data of the 3D reconstruction in real time, and extract spatial structure information from it;

[0040] Step S322: By comparing the expected shape of the complete scene with the current reconstruction result, identify the data missing areas caused by occlusion;

[0041] Step S323: Spatial localization of the missing area, and analysis of the geometric shape and distribution characteristics of the obstructing objects;

[0042] Step S324: Based on the coordinate information of the missing region and the physical properties of the obstruction, a path planning algorithm is used to generate a supplementary scanning path that can bypass the obstruction.

[0043] In this embodiment, the dynamic data for 3D reconstruction in step S321 includes real-time point cloud streams, image sequences, etc., and the extracted spatial structure information can cover the object's position coordinates, size, surface texture, etc.; the expected shape of the complete scene in step S322 can be constructed based on ship design drawings, historical scanning models, etc., and the identification of missing data areas can be achieved through interpolation calculation, feature matching, etc.; the geometric shape of the occluder in step S323 can be divided into spherical, cylindrical, irregular, etc., and the distribution characteristics include density, spatial arrangement, etc.; the path planning algorithm in step S324 can adopt A* algorithm, RRT algorithm, etc., and the physical properties of the occluder include hardness, surface smoothness, etc. Of course, these contents are not limited to the above-listed situations. For example, the spatial structure information can also include the object's motion trend, and the path planning algorithm can also be optimized by combining the device's motion performance parameters. Those skilled in the art can flexibly adjust according to the actual scenario.

[0044] In one embodiment, step S324, the process of generating a supplementary scanning path that can bypass the obstruction based on the coordinate information of the missing area and the physical properties of the obstruction using a path planning algorithm, includes: converting the coordinates of the missing area and the physical properties of the obstruction into a three-dimensional spatial model, clarifying the area to be scanned and the obstacles that must be avoided; setting path constraints according to the motion capability of the scanning device; using a path planning algorithm to search for an initial path from the current position to the missing area that can bypass the obstruction; optimizing the path to make it smoother and more efficient, and verifying whether it can completely cover the missing area, and finally generating an executable supplementary scanning path.

[0045] In this embodiment, the transformation of the 3D spatial model can be achieved through point cloud modeling, mesh generation, and other methods, clearly presenting the boundary range of the missing area and the spatial occupancy of obstructions. The motion capabilities of the scanning device include maximum moving speed, turning angle limits, scanning radius, etc., and path constraints can be set accordingly, such as safety distance and upper limit of path length. Path optimization can be achieved through methods such as removing redundant nodes and curve fitting. The verification process can be combined with simulation to check the coverage integrity and obstacle avoidance effectiveness of the path. Of course, these contents are not limited to the above-listed situations. For example, the 3D spatial model can also incorporate the influence weight of environmental parameters, and path constraints can also consider energy consumption factors. Those skilled in the art can flexibly adjust according to actual needs.

[0046] In one embodiment, after path planning, parameter initialization, and scanning are completed, the method further includes: dynamically adjusting the scanning parameters, data acquisition method, and data processing algorithm in response to real-time fluctuations in device status.

[0047] In this embodiment, device status fluctuations include laser emitter power attenuation, lens fogging or contamination, sensor response delay, and abnormal motor operation. Adjustments to scanning parameters may involve laser emission frequency, scanning resolution, and exposure time; for example, increasing the scanning frequency to increase data redundancy when laser power attenuates. Adjustments to data acquisition methods include switching between single-mode and multi-mode (e.g., laser + structured light), such as switching to dual-mode scanning when lens fogging causes image blurring. Adjustments to data processing algorithms cover improving the strength of point cloud denoising algorithms and optimizing image enhancement strategies; for example, strengthening noise reduction processing to address increased sensor noise. Of course, these are not limited to the situations listed above. For example, device status fluctuations may also include battery power depletion, and scanning parameter adjustments may involve dynamic scaling of the scanning range. Those skilled in the art can flexibly set these parameters according to the actual scenario.

[0048] In one embodiment, the process of dynamically adjusting scanning parameters, data acquisition methods, and data processing algorithms in response to real-time fluctuations in device status includes: establishing a collaborative calibration mechanism to incorporate device status into the additional dimension of environmental interference; monitoring device status parameters in real time and identifying abnormal device conditions; initiating corresponding dynamic compensation strategies based on different abnormality types: when laser power attenuation is detected, automatically increasing the scanning frequency to increase data redundancy, while strengthening the point cloud denoising algorithm to reduce noise caused by insufficient power; when lens fogging is detected, switching to a laser + structured light dual-mode scanning mode to compensate for the lack of visual data with laser data; continuously tracking the compensation effect and dynamically adjusting compensation parameters to adapt to changes in device status.

[0049] In this embodiment, the collaborative calibration mechanism can be established by constructing a correlation model between equipment status and environmental parameters, ensuring that the two work synergistically during adjustment. Real-time monitored equipment status parameters include laser power, lens transmittance, and sensor frame rate. Anomaly identification can employ methods such as threshold judgment and trend analysis. In addition to the above-mentioned examples, dynamic compensation strategies may include reducing the scanning speed to ensure stability when the motor speed is abnormal, and activating the calibration algorithm to correct deviations when sensor data drifts. Tracking the compensation effect can be achieved by comparing data quality indicators (such as point cloud accuracy and image clarity) before and after adjustment. The adjustment range of the compensation parameters can be flexibly set according to the severity of equipment status fluctuations. Of course, these are not limited to the scenarios listed above. For example, the collaborative calibration mechanism can also incorporate historical equipment fault data, and anomaly identification can be combined with artificial intelligence algorithms to improve accuracy. Those skilled in the art can adjust these features according to actual needs.

[0050] In one embodiment, step S5, the process of performing fine identification using retrieval enhancement generation technology combined with a multimodal large model based on the preliminary classification results, includes: using a database of contaminated biological features as a knowledge base; based on the range of biological categories locked by the preliminary classification, retrieving reference images, morphological feature descriptions, and classification standards under that category from the knowledge base through a retrieval mechanism; comparing these with the images of the organisms to be identified to select the most similar candidate information; using a multimodal large model as the core inference engine, receiving the images of the organisms to be identified and the retrieved candidate information; analyzing the detailed features of the organisms in the images through visual understanding capabilities, including the branching morphology of algae and the shell texture structure of barnacles; and combining logical reasoning capabilities to match and verify the candidate information, finally outputting the specific species identification result of the contaminated organism.

[0051] In this embodiment, the biomarker database contains a massive amount of biological sample data, including reference images and information such as the organism's growth environment preferences and common attachment areas. The retrieval mechanism can employ feature vector matching and semantic similarity calculation to improve the accuracy of candidate information retrieval. Detailed features analyzed by the multimodal large model can also include mussel shell textures and hydroid tentacle morphology. The logical reasoning process can combine the correlation between the organism's local and overall features for comprehensive judgment. The output recognition result can include a confidence score to facilitate the assessment of recognition reliability. Of course, these features are not limited to those listed above. For example, the knowledge base can be updated in real time with newly added species data, and the retrieval mechanism can incorporate user feedback to optimize the retrieval strategy. Those skilled in the art can flexibly adjust these features according to actual needs.

[0052] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the original technical features, and the technical solutions resulting from these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for monitoring biofouling on the bottom of ships based on three-dimensional reconstruction and adaptive scanning, characterized in that, include: Collect environmental parameters around the bottom of the ship, including image contrast, noise intensity, laser reflectivity fluctuation, water flow velocity, water turbidity, and chlorophyll content. Based on the environmental parameters, a multimodal large model is used to enhance and restore underwater images. The multimodal large model adaptively adjusts the image processing strategy according to the real-time environmental parameters. Path planning is performed based on the constructed environmental change prediction model, and the priority or order of paths is adjusted according to the regional environmental change trend. At the same time, combined with the dynamic information in the 3D reconstruction process, detour paths are supplemented for data missing areas caused by occlusion, and the scanning is completed and the initial model is generated. Analysis of the enhanced underwater images and the initial 3D model outputs coarse segmentation results of fouling organisms, determining the type of fouling organisms in the current scanning area; based on the type of fouling organisms and environmental parameters in the current scanning area, the system learns the environmental change patterns, outputs the optimal combination of scanning parameters in real time, and generates optimized scanning data; The 3D model is updated based on the optimized scan data. At the same time, based on a large model with general segmentation capabilities or zero-shot / few-shot learning capabilities, the fouling organisms in the underwater images associated with the updated 3D model are initially classified. Based on the initial classification results, retrieval enhancement generation technology is used in combination with a multimodal large model to perform fine recognition.

2. The method according to claim 1, characterized in that, Based on the aforementioned environmental parameters, the process of enhancing and restoring underwater images using a multimodal large model includes: The multimodal large model receives underwater images and environmental parameters, and analyzes the image distortion features caused by low illumination and high turbidity in the environmental parameters; Based on the analysis results, the system can independently select appropriate image processing strategies, including dehazing, color correction, brightness optimization, and super-resolution reconstruction. During processing, the system receives updated environmental parameters in real time, dynamically adjusts the intensity and parameter ratio of various processing strategies, and finally outputs enhanced and restored underwater images.

3. The method according to claim 1, characterized in that, The method also includes constructing an environmental change prediction model. Specifically, based on historical ship navigation data, marine environment database, real-time received marine forecasts, and the environmental parameters, an environmental change prediction model is constructed using a time-series prediction algorithm.

4. The method according to claim 1, characterized in that, Based on the constructed environmental change prediction model, path planning is performed, and the path priority or order is adjusted according to the regional environmental change trend. Simultaneously, by incorporating dynamic information from the 3D reconstruction process, detour paths are supplemented for data-missing areas caused by occlusion. The process of completing the scan and generating the initial model includes: Based on the environmental change prediction model, the regional environmental change trend is output, the regional environmental change trend is analyzed, the differences in environmental stability of each region are clarified, and regions with poor environmental stability and prone to drastic changes are set as high priority and scan paths are planned in advance. Regions with stable environment are reduced in priority and scan order is adjusted. During the scanning process, dynamic information of 3D reconstruction is acquired in real time. The data missing areas caused by occlusion are identified through the dynamic information. Based on the spatial location of the missing areas and the shape of the occlusion, a supplementary scanning path that can bypass the occlusion is planned. Complete the scanning of the entire area according to the adjusted priority path and supplementary detour path, integrate all scan data, and build and output the initial model.

5. The method according to claim 4, characterized in that, The process of learning environmental change patterns based on the types of fouling organisms and environmental parameters in the current scanned area, and outputting the optimal combination of scanning parameters in real time, includes: The environmental parameters and the type of contaminated organisms are used as environmental state inputs. The agent is trained using the 3D reconstruction accuracy, contamination identification accuracy, and scanning efficiency as joint reward functions. The agent outputs the optimal combination of scanning parameters in real time to perform targeted adjustments and generate optimized scanning data.

6. The method according to claim 4, characterized in that, During the scanning process, dynamic information of the 3D reconstruction is acquired in real time. This dynamic information is used to identify data gaps caused by occlusion. Based on the spatial location of the gaps and the shape of the occlusions, a supplementary scanning path that bypasses the occlusions is planned. This process includes: Real-time acquisition and analysis of dynamic data from 3D reconstruction, extracting spatial structure information from it; By comparing the expected shape of the complete scene with the current reconstruction results, the data missing areas caused by occlusion are identified; Spatial location of the missing area, and analysis of the geometric shape and distribution characteristics of the obstructing objects; Based on the coordinate information of the missing region and the physical properties of the obstructions, a path planning algorithm is used to generate a supplementary scanning path that can bypass the obstructions.

7. The method according to claim 6, characterized in that, Based on the coordinate information of the missing region and the physical properties of the obstructions, the process of generating a supplementary scanning path that can bypass the obstructions using a path planning algorithm includes: The missing area coordinates and the physical properties of the occluded objects are converted into a three-dimensional spatial model to clarify the area that needs to be scanned and the obstacles that must be avoided. Path constraints are set based on the motion capabilities of the scanning equipment; A path planning algorithm is used to search for an initial path from the current position to the missing area that avoids obstructions. The path is optimized to be smoother and more efficient, and it is verified that it can fully cover the missing areas, ultimately generating an executable supplementary scan path.

8. The method according to claim 2, characterized in that, After path planning, parameter initialization, and scanning are completed, the method further includes: dynamically adjusting the scanning parameters, data acquisition method, and data processing algorithm in response to real-time fluctuations in device status.

9. The method according to claim 8, characterized in that, The process of dynamically adjusting scanning parameters, data acquisition methods, and data processing algorithms in response to real-time fluctuations in device status includes: Establish a collaborative calibration mechanism to incorporate equipment status into the additional dimension of environmental interference; Real-time monitoring of equipment status parameters to identify abnormal equipment conditions; Based on different anomaly types, corresponding dynamic compensation strategies are activated: when laser power attenuation is detected, the scanning frequency is automatically increased to increase data redundancy, while the point cloud denoising algorithm is strengthened to reduce noise caused by insufficient power; when lens fogging is detected, causing image blurring, the laser + structured light dual-mode scanning mode is switched to use laser data to make up for the lack of visual data. Continuously track the compensation effect and dynamically adjust the compensation parameters to adapt to changes in equipment status.

10. The method according to claim 1, characterized in that, Based on the preliminary classification results, the process of performing fine-grained identification using retrieval enhancement generation techniques combined with a multimodal large model includes: Using the biomarker database as a knowledge base, and based on the biological categories initially identified through classification, a retrieval mechanism is used to retrieve reference images, morphological descriptions, and classification criteria for each category from the knowledge base. These are then compared with the images of the organisms to be identified to select the most similar candidate information. Using a multimodal large model as the core inference engine, it receives images of organisms to be identified and retrieves candidate information. Through visual understanding, it analyzes the detailed features of organisms in the images, including the branching morphology of algae and the shell texture of barnacles. Combined with logical reasoning, it matches and verifies the candidate information, and finally outputs the specific species identification result of the contaminated organism.

Images