Hull rust defect laser cleaning method and wall-climbing robot

Abstract

The invention provides a hull rust defect laser cleaning method and a wall-climbing robot, and relates to the technical field of laser cleaning maintenance, and the method comprises the following steps: obtaining laser radar data and depth camera data; through a visual SLAM (Simultaneous Localization and Mapping) fusion mapping navigation algorithm, and in combination with ship wall environment information, pose information is obtained; according to the pose information, the moving path of the laser wall-climbing cleaning robot is determined; obtaining an oblique view of each target area in the moving path; constructing a hull rust defect positioning and segmenting model; positioning and segmenting rust in the oblique view through a ship body rust defect positioning and segmenting model to obtain a positioning and segmenting result; and according to a positioning segmentation result, removing rust of the ship body through laser. The robot utilizes deep learning and laser radar, and combines a multi-source data fusion algorithm and a synchronous positioning and visual SLAM fusion mapping navigation algorithm to realize ship outer wall state and surrounding environment perception and ship rust removal cleaning full-coverage path planning.

Description

Technical Field

[0001] The present application relates to the field of laser cleaning and maintenance technology, and in particular to a laser cleaning method for rust defects on a ship hull and a wall-climbing robot. Background Art

[0002] Since ships operate in seawater for a long time, their hull plates will be corroded by seawater and form an oxide layer, namely rust, under the oxidation of air. If not cleaned in time, it will further corrode the hull, reduce the structural strength of the hull, and affect the service life of the ship; it will also affect the adhesion of the coating when the hull is sprayed, so ship rust removal and cleaning work is very important. However, traditional ship rust removal and cleaning methods mainly rely on manual operation, which is inefficient, time-consuming and labor-intensive, and the rust removal and cleaning working environment is complex and risky, posing a threat to the health of operators. Therefore, it is critical to develop a robot that can walk freely on the outer wall of the ship and complete the rust removal and cleaning process.

[0003] In recent years, with the continuous development of laser technology, laser technology has been applied to the field of cleaning, and its effect has been found to have significant advantages. Compared with traditional cleaning methods, the advantages of laser cleaning are mainly concentrated in: the cleaning parameters are adjustable, which can achieve precise cleaning and a wide range of cleaning; cleaning causes little damage to the substrate, ensuring that the performance of the substrate remains unchanged after cleaning; "green" cleaning methods have less environmental pollution; high cleaning efficiency, low cost and easier to automate. The many advantages of laser cleaning technology make it more suitable for the development of today's cleaning industry, and its huge potential will inevitably replace traditional cleaning methods. Summary of the invention

[0004] The purpose of the present invention is to provide a laser cleaning method for rust defects on a hull and a wall-climbing robot in order to solve the problems of poor automation and low precision in the existing laser cleaning method for rust on a ship.

[0005] The above-mentioned purpose of the present application is achieved through the following technical solutions: S1: Get lidar data and depth camera data; S2: The position information is obtained by integrating the visual SLAM mapping and navigation algorithm with the ship wall environment information; the moving path of the laser wall-climbing cleaning robot is determined based on the position information; S3: Obtain oblique views of each target area in the moving path; S4: Construct a hull rust defect location and segmentation model; S5: Using the hull rust defect positioning and segmentation model, the rust in the oblique view is positioned and segmented to obtain the positioning and segmentation results; S6: Based on the positioning and segmentation results, the rust on the hull is removed by laser.

[0006] Optionally, step S2 includes: S21: preprocessing the lidar data and the depth camera data; S22: The Transformer-based COTR network performs feature extraction on each frame of the LiDAR data and the depth camera data and completes inter-frame feature matching; the matched feature point data and the COTR feature points that restore the depth information are obtained; the feature point data includes: edge feature point data and plane feature data; S23: Obtaining inter-frame pose change information of the laser wall climbing cleaning robot through feature point data and COTR feature points; S24: using an adaptive Monte Carlo algorithm to obtain initial posture data of the laser wall-climbing cleaning robot; S25: By using the visual word bag model, combining the feature point data and COTR feature points, the inter-frame pose estimation information and the initial pose data are optimized to determine the final pose information of the laser wall-climbing cleaning robot; S26: Determine the moving path of the laser wall-climbing cleaning robot according to the position and posture information.

[0007] Optionally, step S4 includes: The hull rust defect positioning and segmentation model includes: YOLOv8 target detection and positioning model and MFAU-net image segmentation model; The U-net image segmentation model is improved through the MFA module to construct the MFAU-net image segmentation model.

[0008] Optionally, step S5 includes: S51: Processing the oblique view through affine transformation to convert it into a top view; S52: Input the top view into the YOLOv8 target detection and positioning model to locate the rust defects on the hull and the area to be cleaned; S53: Input the top view into the MFAU-net image segmentation model to quickly and finely segment the rust defects on the hull and the area to be cleaned; S54: According to the positioning result and the fast and fine segmentation, a rust positioning and segmentation result is obtained.

[0009] Optionally, step S52 includes: The training of the YOLOv8 target detection and positioning model is based on the mixed loss function. The parameters of the model are optimized in iterative training. The calculation formula of the mixed loss function is as follows:

[0010] in, represents the rectangular box loss; is the confidence loss; is the classification loss; , , is the corresponding weight of each loss function.

[0011] Optionally, step S53 includes: Using binary cross entropy loss With Dice loss The mixed loss function is used as the loss function of the MFAU-net image segmentation model and is defined as follows:

[0012]

[0013]

[0014] In the formula, Representing pixels True label; Represents pixels Predicted probability of belonging to the defective class.

[0015] A laser cleaning system for rust defects on a ship hull, the system comprising: a laser radar module, a depth camera module, a processor, and a laser cleaning module; The LiDAR module, depth camera module, and laser cleaning module are all connected to the processor; The LiDAR module is used to obtain LiDAR data; The depth camera module is used to obtain depth camera data and oblique views of each target area in the moving path; The processor is used to obtain the position information by integrating the visual SLAM mapping navigation algorithm with the ship wall environment information; according to the position information, the moving path of the laser wall climbing cleaning robot is determined; The processor is also used to obtain the position information by integrating the visual SLAM mapping navigation algorithm with the ship wall environment information; according to the position information, the moving path of the laser wall-climbing cleaning robot is determined; The processor is also used to construct a hull rust defect location segmentation model; The processor is also used to locate and segment the rust in the oblique view through the hull rust defect location and segmentation model to obtain a location and segmentation result; The laser cleaning module is used to remove rust from the hull by laser according to the positioning and segmentation results.

[0016] A wall-climbing robot for laser cleaning of rust defects on a ship hull, the robot comprising: a robot body, a lifting mechanism, a depth camera, a laser radar, a laser, a driving wheel, a magnetic force combined with a vacuum adsorption unit, a strip light source, a detection camera and a laser transmitter head; The laser transmitter head is connected to the laser; the laser transmitter head is arranged on the lifting mechanism, and the lifting mechanism is embedded in the robot body; The lifting mechanism is used to control the distance between the laser transmitter head and the ship wall; The strip light source is arranged below the detection camera to provide light source for collecting images; The magnetic force combined with the vacuum adsorption unit is arranged at the bottom of the robot body; the driving wheel is arranged on the outside of the robot body; The depth camera, lidar, laser, bar light source and detection camera are arranged on the robot body.

[0017] The beneficial effects of the technical solution provided by this application are: 1. From the method, the moving path of the laser wall-climbing cleaning robot is determined by integrating visual SLAM with mapping and navigation algorithm. From the structural design, a magnetic force combined with vacuum adsorption unit is designed to realize the function of automatic obstacle avoidance and movement of the laser wall-climbing cleaning robot on the ship wall.

[0018] 2. By constructing a hull rust defect positioning and segmentation model, the function of real-time positioning and segmentation of the rust on the hull is realized. Combined with laser cleaning equipment, the function of accurate rust removal is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present application will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 is a step diagram in an embodiment of the present application; Figure 2 is a schematic diagram of a process of cleaning a hull surface by a robot in an embodiment of the present application; Figure 3 It is a flow chart of the visual SLAM fusion mapping and navigation algorithm in the embodiment of the present application; Figure 4 It is a sequential rust removal diagram between the block rust removal and cleaning areas in the embodiment of the present application; Figure 5 It is a flow chart of the rust removal principle in the embodiment of the present application; Figure 6 is a schematic diagram of the camera field of view of the visual recognition system in an embodiment of the present application; Figure 7 is the acquisition image after affine transformation in the embodiment of the present application; Figure 8 It is a structure diagram of the YOLOv8 target detection and positioning model in the embodiment of the present application; Fig. 9 It is a diagram of the MFAU net rust defect fast segmentation algorithm in an embodiment of the present application; Fig.10 It is a diagram of the MFA module in the embodiment of the present application; Fig.11 is an output result diagram in an embodiment of the present application; Fig.12 is a structural diagram of a robot in an embodiment of the present application; Fig.13 is a structural diagram of a control system in an embodiment of the present application; Fig.14 It is a structural diagram of the control system in the embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to have a clearer understanding of the technical features, purposes and effects of the present application, the specific implementation methods of the present application are now described in detail with reference to the accompanying drawings.

[0021] An embodiment of the present application provides a method for laser cleaning of rust defects on a ship hull.

[0022] Please refer to Figure 1 , Figure 1 This is a step diagram of a laser cleaning method for rust defects on a hull according to an embodiment of the present application, comprising: S1: Get lidar data and depth camera data; S2: The position information is obtained by integrating the visual SLAM mapping and navigation algorithm with the ship wall environment information; the moving path of the laser wall-climbing cleaning robot is determined based on the position information; S3: Obtain oblique views of each target area in the moving path; S4: Construct a hull rust defect location and segmentation model; S5: Using the hull rust defect positioning and segmentation model, the rust in the oblique view is positioned and segmented to obtain the positioning and segmentation results; S6: Based on the positioning and segmentation results, the rust on the hull is removed by laser.

[0023] As an embodiment, sequential rust removal between block rust removal and cleaning areas: In order to prevent the wall-climbing robot from sliding relative to the wall due to gravity when crawling horizontally along the vertical surface of the ship to perform rust removal and cleaning operations, the wall-climbing robot is configured to crawl along the vertical direction of the outer wall of the ship as much as possible to perform full coverage rust removal and cleaning, and to avoid crawling in the horizontal direction as much as possible. In an unknown cleaning working environment, the boundaries of the local square block areas are manually set, and the boundary length must be much longer than the side length of the grid map cell, and the lower left corner of each block area is set as the block origin, which is convenient for path planning in the block rust removal and cleaning area. See the diagram of sequential rust removal between block rust removal and cleaning areas. Figure 4 The wall-climbing robot moves in a vertical direction, removing rust and cleaning each block area in order from bottom to top and from left to right until it reaches the boundary.

[0024] This application adopts the above technical solution, and the laser cleaning wall-climbing robot adopts magnetic force combined with vacuum adsorption, AC servo motor wheel drive, and high-power laser cleaning and rust removal. The magnetic force combined with vacuum adsorption can make the robot firmly adsorbed on the surface of the hull, realize flexible forward, backward, turning and other actions, and has a certain ability to cross obstacles on curved surfaces, and can flexibly deal with welds and riveted seams on the outer surface of the hull. During operation, a safety sling mechanism can be connected to avoid high-altitude falling incidents during construction and ensure the safety of robot construction.

[0025] Step S2 includes: S21: preprocessing the lidar data and the depth camera data; S22: The Transformer-based COTR network performs feature extraction on each frame of the LiDAR data and the depth camera data and completes inter-frame feature matching; the matched feature point data and the COTR feature points that restore the depth information are obtained; the feature point data includes: edge feature point data and plane feature data; S23: Obtaining inter-frame pose change information of the laser wall climbing cleaning robot through feature point data and COTR feature points; S24: using an adaptive Monte Carlo algorithm to obtain initial posture data of the laser wall-climbing cleaning robot; S25: By using the visual word bag model, combining the feature point data and COTR feature points, the inter-frame pose estimation information and the initial pose data are optimized to determine the final pose information of the laser wall-climbing cleaning robot; S26: Determine the moving path of the laser wall-climbing cleaning robot according to the position and posture information.

[0026] This application provides an embodiment as follows: Figure 2 The flowchart of the autonomous navigation system of the laser cleaning wall-climbing robot is shown in the figure. The robot uses deep learning and laser radar, combined with multi-source data fusion algorithm, synchronous positioning and visual SLAM fusion mapping navigation algorithm to realize the perception of the state of the ship's outer wall and the surrounding environment and the full coverage path planning of the ship's rust removal and cleaning. The full coverage path planning includes two parts: global and local path planning. The role of global path planning is to perform basic path planning, including sequential rust removal between block rust removal and cleaning areas and path planning within block rust removal and cleaning areas; local path planning is mainly used to avoid obstacles when encountering obstacles or boundaries. After successful obstacle avoidance, the rust removal and cleaning work continues to be performed according to the global planning path until the full coverage cleaning of the entire ship's rust removal area is completed.

[0027] The present application provides an embodiment as follows: a visual SLAM fusion mapping and navigation algorithm: This method uses a lidar as an auxiliary device to complete tasks such as position solution and obstacle detection in the robot's autonomous navigation.

[0028] This application provides an embodiment as follows: the overall process of the algorithm is as follows Figure 3 As shown in the figure. The overall architecture of the algorithm first pre-processes the data sent back by the laser radar and the depth camera. Next, it performs feature extraction on each frame of the laser radar and the depth camera and completes inter-frame feature matching. The matched feature point data and the COTR feature points that restore the depth information are input into the pose estimation module to obtain the inter-frame pose change of the laser wall-climbing cleaning robot. Finally, the loop detection module is designed using the visual word bag model, and the pose estimation data obtained by the front end is optimized to obtain the final pose and point cloud map.

[0029] The framework of the visual SLAM fusion mapping and navigation algorithm is divided into two parts: front-end odometer and back-end nonlinear optimization. In the front-end feature extraction and matching process, a Transformer-based COTR network is used to extract feature points. The depth information of COTR feature points is restored using lidar point cloud data. In addition, in order to reduce the error of pose estimation, the laser vision feature point fusion pose estimation technology based on Lie algebra derivative is also adopted to obtain more accurate pose data.

[0030] In the overall framework of the algorithm, the nonlinear optimization of the backend occupies an important position. The global optimization strategy based on BA (Bundle Adjustment) is used in the nonlinear optimization part of the backend. When executing the loop detection module, the visual word bag model in the ORB-SLAM2 framework is used for reference. Visual data contains rich feature information, so the loop detection module is designed based on the visual word bag model, which effectively corrects the drift caused by the cumulative error of the posture, and finally outputs the posture data and point cloud map required for laser cleaning wall climbing positioning.

[0031] Step S4 includes: The hull rust defect positioning and segmentation model includes: YOLOv8 target detection and positioning model and MFAU-net image segmentation model; The U-net image segmentation model is improved through the MFA module to construct the MFAU-net image segmentation model.

[0032] Step S5 includes: S51: Processing the oblique view through affine transformation to convert it into a top view; S52: Input the top view into the YOLOv8 target detection and positioning model to locate the rust defects on the hull and the area to be cleaned; S53: Input the top view into the MFAU-net image segmentation model to quickly and finely segment the rust defects on the hull and the area to be cleaned; S54: According to the positioning result and the fast and fine segmentation, a rust positioning and segmentation result is obtained.

[0033] As an example, the principle flow chart of the laser cleaning method for rust removal and cleaning area of ​​the present application is as follows: Figure 5 The camera field of view of the wall-climbing cleaning robot is shown in Figure 6 As shown. The image captured by the camera is an oblique view, which needs to be converted into a top view for easy observation and analysis. According to the installation position of the camera, the short side length and height of the trapezoidal field of view can be calculated. The hull surface image obtained by the visual recognition system needs to be cropped and deformed through affine transformation to obtain Figure 7 The image after affine transformation is input into the hull rust defect positioning and segmentation model. The positioning and segmentation model includes: YOLOv8 target detection and positioning model, and MFAU-net image segmentation model.

[0034] Step S52 includes: The training of the YOLOv8 target detection and positioning model is based on the mixed loss function. The parameters of the model are optimized in iterative training. The calculation formula of the mixed loss function is as follows:

[0035] in, represents the rectangular box loss; is the confidence loss; is the classification loss; , , is the corresponding weight of each loss function.

[0036] As an example, the YOLOv8 target detection and positioning model: The YOLOv8 target detection and positioning model locates the rust defects on the hull and the area to be cleaned. The overall framework is as follows Figure 8 shown.

[0037] The quality of the YOLOv8 network detection results can be measured by three indicators: rectangular box, confidence and classification probability. The rectangular box represents the size of the object and its exact location; the confidence indicates the confidence of the predicted rectangle, ranging from 0 to 1. The higher the value, the greater the possibility of finding the object in the rectangle; the classification probability indicates the category of the object. Therefore, in order to enable the network to achieve better detection results, the above three indicators, namely, rectangular box, confidence and classification probability, need to be gradually optimized. Therefore, the network's loss function needs to measure the deviation between the predicted result and the expected label for all three aspects. Therefore, the mixed loss function of the model contains three parts of loss.

[0038] Confidence loss and classification loss are measured using the binary cross entropy loss function (BCEWithLogitsLoss).

[0039] The rectangular box loss is calculated using It is measured by the loss function. is On this basis, the distance between the predicted target box and the true label box, the overlap ratio, the aspect ratio of the box, and the penalty term are additionally considered to optimize the regression accuracy and speed of the rectangular box, which can make the predicted rectangular box regression more stable. The loss is shown as follows:

[0040] In the formula, It is the intersection-union ratio, which is defined as the ratio of the intersection to the union between the predicted box and the actual label box. The calculation formula is: , where A represents the predicted box, B represents the actual label box; c is the diagonal distance of the minimum enclosing rectangle containing the actual label box and the predicted box; and Are loss parameters, the formula is defined as:

[0041]

[0042] in, Indicates the width of the actual label box. Indicates the actual height of the label box; Indicates the width of the prediction box. Indicates the height of the predicted box.

[0043] Step S53 includes: Using binary cross entropy loss With Dice loss The mixed loss function is used as the loss function of the MFAU-net image segmentation model and is defined as follows:

[0044]

[0045]

[0046] In the formula, Representing pixels True label; Represents pixels Predicted probability of belonging to the defective class.

[0047] As an example, the specific structure of the MFA module is as follows: Fig.10 As shown. MFAU-net image segmentation model, as Fig. 9As shown in the figure. The MFAU-net rust defect fine segmentation model quickly and finely segments the rust defects on the hull and the area to be cleaned. The MFAU-net model adds the MFA module to the original U-net to capture rich local features, making the feature expression of rust defects more obvious. The output results of the MFAU-net model are shown in the figure. Fig.11 shown.

[0048] A laser cleaning system for rust defects on a ship hull, the system comprising: a laser radar module, a depth camera module, a processor, and a laser cleaning module; The LiDAR module, depth camera module, and laser cleaning module are all connected to the processor; The LiDAR module is used to obtain LiDAR data; The depth camera module is used to obtain depth camera data and oblique views of each target area in the moving path; The processor is used to obtain the position information by integrating the visual SLAM mapping navigation algorithm with the ship wall environment information; according to the position information, the moving path of the laser wall climbing cleaning robot is determined; The processor is also used to obtain the position information by integrating the visual SLAM mapping navigation algorithm with the ship wall environment information; according to the position information, the moving path of the laser wall-climbing cleaning robot is determined; The processor is also used to construct a hull rust defect location segmentation model; The processor is also used to locate and segment the rust in the oblique view through the hull rust defect location and segmentation model to obtain a location and segmentation result; The laser cleaning module is used to remove rust from the hull by laser according to the positioning and segmentation results.

[0049] As an embodiment, the hull rust defect laser cleaning wall-climbing robot is equipped with Nvidia Jetson Tx2 processor as the central processing unit; Velodyne VLP-16 line laser radar for positioning and environmental mapping; LED strip light source and CCD industrial camera as image acquisition module for defect detection and analysis; servo motor and dedicated motor control board control the movement and posture of the robot.

[0050] like Fig.12 As shown, a wall-climbing robot for laser cleaning of rust defects on a ship hull comprises: a robot body 201, a lifting mechanism 202, a depth camera 203, a laser radar 204, a laser 205, a driving wheel 206, a magnetically combined vacuum adsorption unit 207, a strip light source 208, a detection camera 209 and a laser transmitter head 210; The laser emitting head 210 is connected to the laser 205; the laser emitting head 210 is arranged on the lifting mechanism 202, and the lifting mechanism 202 is embedded in the robot body 201; The lifting mechanism 202 is used to control the distance between the laser emitting head 210 and the ship wall; The strip light source 208 is disposed below the detection camera 209 and is used to provide a light source to collect images; The magnetic force combined with the vacuum adsorption unit 207 is arranged at the bottom of the robot body 201; the driving wheel 206 is arranged on the outside of the robot body 201; The depth camera 203 , the laser radar 204 , the laser 205 , the bar light source 208 and the detection camera 209 are arranged on the robot body 201 .

[0051] As an example, the laser cleaning wall climbing robot control system is as follows: Fig.13 , the hierarchical structure of the control system is as follows Fig.14 The control system 510 is provided with functions, including moving operation 511, laser parameter setting 512, laser emission control 513, lifting control 514, visual detection 515 and multi-source data fusion navigation 516. The driving wheels are the four wheels of the robot.

[0052] Specific implementation of mobile operation: There is a visual interface on the software of the control system. The wall-climbing robot is controlled by the keyboard, and the direction keys are pressed to send corresponding signals to the built-in control unit of the wall-climbing robot.

[0053] Specific implementation of laser parameter setting: There is a visual interface on the control system software. Enter the laser parameter value at the corresponding marked position through the keyboard and click the confirmation button on the interface to complete the setting of the laser parameters.

[0054] Specific implementation of laser emission control: There is a visual interface on the control system software, and laser generation and laser emission can be started by clicking the laser generation confirmation button and laser emission confirmation button on the interface.

[0055] Specific implementation of visual detection: There is a visual interface on the control system software, and there is real-time imaging sent by the detection camera in the secondary visual detection frame, and the scaling accuracy of the detection distance can be achieved by entering the scaling step.

[0056] The specific implementation of multi-source data fusion navigation is as follows: there is a visual interface on the control system software, and in the main visual detection frame there are real-time constructed maps sent by the depth camera and lidar as well as full coverage path planning results.

[0057] The above are only exemplary embodiments of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure.

[0058] This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art not described in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A laser cleaning method for rust defects on a ship hull, characterized in that: The method comprises the following steps: S1: Get lidar data and depth camera data; S2: The position information is obtained by integrating the visual SLAM mapping and navigation algorithm with the ship wall environment information; the moving path of the laser wall-climbing cleaning robot is determined based on the position information; S3: Obtain oblique views of each target area in the moving path; S4: Construct a hull rust defect location and segmentation model; S5: Using the hull rust defect positioning and segmentation model, the rust in the oblique view is positioned and segmented to obtain the positioning and segmentation results; S6: Based on the positioning and segmentation results, the rust on the hull is removed by laser.

2. A laser cleaning method for rust defects on a ship hull as claimed in claim 1, characterized in that: Step S2 includes: S21: preprocessing the lidar data and the depth camera data; S22: The Transformer-based COTR network performs feature extraction on each frame of the LiDAR data and the depth camera data and completes inter-frame feature matching; the matched feature point data and the COTR feature points that restore the depth information are obtained; the feature point data includes: edge feature point data and plane feature data; S23: Obtaining inter-frame pose change information of the laser wall climbing cleaning robot through feature point data and COTR feature points; S24: using an adaptive Monte Carlo algorithm to obtain initial posture data of the laser wall-climbing cleaning robot; S25: By using the visual word bag model, combining the feature point data and COTR feature points, the inter-frame pose estimation information and the initial pose data are optimized to determine the final pose information of the laser wall-climbing cleaning robot; S26: Determine the moving path of the laser wall-climbing cleaning robot according to the position and posture information.

3. The laser cleaning method for rust defects on a ship hull as claimed in claim 1, characterized in that: Step S4 includes: The hull rust defect positioning and segmentation model includes: YOLOv8 target detection and positioning model and MFAU-net image segmentation model; The U-net image segmentation model is improved through the MFA module to construct the MFAU-net image segmentation model.

4. The laser cleaning method for rust defects on a ship hull as claimed in claim 1, characterized in that: Step S5 includes: S51: Processing the oblique view through affine transformation to convert it into a top view; S52: Input the top view into the YOLOv8 target detection and positioning model to locate the rust defects on the hull and the area to be cleaned; S53: Input the top view into the MFAU-net image segmentation model to quickly and finely segment the rust defects on the hull and the area to be cleaned; S54: According to the positioning result and the fast and fine segmentation, a rust positioning and segmentation result is obtained.

5. The laser cleaning method for rust defects on a ship hull as claimed in claim 1, characterized in that: Step S52 includes: The training of the YOLOv8 target detection and positioning model is based on the mixed loss function. The parameters of the model are optimized in iterative training. The calculation formula of the mixed loss function is as follows: in, represents the rectangular box loss; is the confidence loss; is the classification loss; , , is the corresponding weight of each loss function.

6. The laser cleaning method for rust defects on a ship hull as claimed in claim 1, characterized in that: Step S53 includes: Using binary cross entropy loss With Dice loss The mixed loss function is used as the loss function of the MFAU-net image segmentation model and is defined as follows: In the formula, Representing pixels True label; Represents pixels Predicted probability of belonging to the defective class.

7. A laser cleaning system for rust defects on a ship hull, used to implement a laser cleaning method for rust defects on a ship hull as claimed in any one of claims 1 to 6, characterized in that: The system includes: a laser radar module, a depth camera module, a processor, and a laser cleaning module; The LiDAR module, depth camera module, and laser cleaning module are all connected to the processor; The LiDAR module is used to obtain LiDAR data; The depth camera module is used to obtain depth camera data and oblique views of each target area in the moving path; The processor is used to obtain the position information by integrating the visual SLAM mapping navigation algorithm with the ship wall environment information; according to the position information, the moving path of the laser wall climbing cleaning robot is determined; The processor is also used to obtain the position information by integrating the visual SLAM mapping navigation algorithm with the ship wall environment information; according to the position information, the moving path of the laser wall-climbing cleaning robot is determined; The processor is also used to construct a hull rust defect location and segmentation model; The processor is also used to locate and segment the rust in the oblique view through the hull rust defect location and segmentation model to obtain a location and segmentation result; The laser cleaning module is used to remove rust from the hull by laser according to the positioning and segmentation results.

8. A wall-climbing robot for laser cleaning of rust defects on a ship hull, used to implement a laser cleaning method for rust defects on a ship hull as claimed in any one of claims 1 to 6, characterized in that: The robot comprises: a robot body (201), a lifting mechanism (202), a depth camera (203), a laser radar (204), a laser (205), a driving wheel (206), a magnetic force combined with a vacuum adsorption unit (207), a strip light source (208), a detection camera (209) and a laser emission head (210); The laser emitting head (210) is connected to the laser (205); the laser emitting head (210) is arranged on the lifting mechanism (202), and the lifting mechanism (202) is embedded in the robot body (201); The lifting mechanism (202) is used to control the distance between the laser emitting head (210) and the ship wall; The strip light source (208) is arranged below the detection camera (209) and is used to provide light source for collecting images; A magnetic force combined vacuum adsorption unit (207) is arranged at the bottom of the robot body (201); a driving wheel (206) is arranged on the outside of the robot body (201); A depth camera (203), a laser radar (204), a laser (205), a bar light source (208) and a detection camera (209) are arranged on the robot body (201).

Images