Underwater robot submarine cable inspection method and system based on two-dimensional image sonar

Abstract

The invention discloses an underwater robot submarine cable inspection method and system based on two-dimensional image sonar, and relates to the field of submarine cable inspection. The existing submarine cable inspection method has the problems of insufficient detection precision, lack of autonomy, dynamic obstacle avoidance defect and the like. The method comprises the steps of submarine cable feature recognition model construction, underwater robot sensing system configuration, image processing, path planning, obstacle avoidance path generation and detouring. According to the technical scheme, the underwater robot carries the electromagnetic induction sensor, the two-dimensional image sonar and the ultra-short baseline positioning device, and an image processing technology and a path planning algorithm are combined, so that efficient inspection of the submarine cable is realized; when the robot encounters an obstacle, the robot can generate an obstacle point cloud through image processing in combination with prior obstacle information, and performs obstacle avoidance in combination with a local path planning algorithm, so that safe and smooth proceeding of an inspection task is ensured; the detection precision is effectively improved, automatic and efficient inspection is realized, and a powerful guarantee is provided for long-term stable operation of the submarine cable.

Description

Technical Field

[0001] The present invention relates to the field of submarine cable inspection, and in particular to a submarine cable inspection method and system using an underwater robot based on two-dimensional image sonar. Background Art

[0002] As the core infrastructure of the global communication network, submarine cables have long been exposed to complex marine environments and face multiple natural and man-made threats. Specifically: Natural erosion: deep sea high pressure, salt corrosion and attachment erosion of marine organisms (such as shipworms and drilling shellfish) can lead to deterioration of cable sheath structure; Geological activities: Undersea earthquakes, landslides and ocean currents can easily cause cables to be buried or broken; Human damage: Mechanical damage caused by fishing trawling, anchor dragging, etc. accounts for more than 60%. The above factors lead to cable fault repair costs of up to millions of dollars per time, and the communication losses caused during the interruption can reach tens of millions of dollars per hour.

[0003] Traditional inspection technology mainly relies on the following two methods: Manual diving inspection: Close-up visual inspection by divers, limited by the duration of underwater operations (usually <4 hours / time) and safety risks (such as water pressure damage, biological attacks), is only applicable to shallow sea areas; Remotely operated vehicle (ROV) inspection: Although contact inspection can be performed using a robotic arm, the operation is complex, the endurance is limited (typical endurance is 6-8 hours), and the equipment maintenance cost is high.

[0004] In recent years, sonar imaging technology has been gradually applied to the field of cable detection, such as: Side-scan sonar system: It can identify the cable contour through sound wave reflection imaging, but it is limited by low resolution (usually >5cm) and data processing delay, making it difficult to locate small damage in real time; Multi-beam sonar device: can generate three-dimensional seabed topographic maps, but lacks deep integration with intelligent algorithms and still requires manual interpretation of images.

[0005] The defects of the prior art are summarized as follows: Insufficient detection accuracy: Traditional sonar systems cannot effectively identify millimeter-level cable surface cracks; Lack of autonomy: Existing ROVs require manual control throughout the entire process and cannot achieve long-distance autonomous inspections; Dynamic obstacle avoidance defects: Lack of real-time obstacle modeling and path replanning capabilities, which can easily lead to secondary damage.

[0006] Therefore, an underwater robot inspection method that integrates high-precision sonar imaging, deep learning algorithms and autonomous motion planning is needed to break through the limitations of existing technologies in detection efficiency, environmental adaptability and automation level. Summary of the invention

[0007] The technical problem to be solved and the technical task proposed by the present invention are to improve and perfect the existing technical solutions, and provide a method and system for submarine cable inspection by an underwater robot based on two-dimensional image sonar, so as to achieve the purpose of submarine cable inspection more efficiently, safely and reliably. To this end, the present invention adopts the following technical solutions.

[0008] A submarine cable inspection method using an underwater robot based on two-dimensional image sonar comprises the following steps: 1) Build a submarine cable feature recognition model, train the model on a sonar image set containing submarine cable features, and output a weight file; 2) Equipped with underwater robot sensing system, including electromagnetic induction sensor, horizontal detection two-dimensional image sonar, oblique downward detection two-dimensional image sonar and ultra-short baseline positioning device; 3) After entering the water, the underwater robot enters the submarine cable search mode, detects the electromagnetic signal of the submarine cable through the electromagnetic induction sensor, calculates the nearest position of the submarine cable and guides the robot to move above the submarine cable; 4) When the underwater robot reaches the top of the submarine cable, it enters the submarine cable tracking mode, uses the oblique downward detection two-dimensional image sonar to collect images in real time, identifies the submarine cable features based on the trained submarine cable feature recognition model, calculates the lateral deviation between the center of mass of the feature box and the center of the image, and adjusts the robot's heading through the PID controller output angular velocity so that it patrols along the cable path; 5) During the inspection process, when the two-dimensional image sonar with the horizontal detection direction detects an obstacle, the sonar image is grayed, thresholded and Hough line transformed to extract the edge information of the obstacle and generate point cloud data in combination with the preset obstacle shape; the point cloud data is fused with the underwater robot's posture information and converted to the world coordinate system, and the local path planning algorithm is input to generate an obstacle avoidance path, and the robot is controlled to detour along the obstacle avoidance path; 6) After the robot bypasses the obstacle, if the sonar image does not detect the characteristics of the submarine cable, the position of the submarine cable will be relocated based on the electromagnetic induction sensor until the sonar image restores the submarine cable recognition and then continues the inspection until it reaches the target point.

[0009] This technical solution uses electromagnetic induction sensors to enable the underwater robot to accurately detect the electromagnetic signals of the submarine cable, calculate the position of the submarine cable, and guide the robot to quickly locate above the submarine cable. This method can efficiently realize the rapid search and positioning of submarine cables in complex submarine environments.

[0010] By adopting a trained submarine cable feature recognition model and collecting and analyzing images in real time through oblique downward two-dimensional image sonar, the underwater robot can automatically identify the characteristics of the submarine cable, accurately calculate and adjust the heading, and ensure that the robot inspects along the submarine cable path, thereby improving the inspection efficiency and accuracy and avoiding the tediousness and risks of manual operation.

[0011] The horizontal detection two-dimensional image sonar can detect obstacles in real time, extract the edge information of obstacles through image processing techniques (such as grayscale, thresholding, and Hough line transform), and generate point cloud data in combination with preset shapes. Using this data, the underwater robot can effectively identify and bypass obstacles to ensure that the inspection work is not interfered with by the outside world. In addition, the local path planning algorithm can quickly generate an obstacle avoidance path when the robot encounters an obstacle, which improves the flexibility and stability of the system.

[0012] If the sonar image is lost or the cable features cannot be identified, the system can automatically switch to the electromagnetic induction sensor to reposition the cable, ensuring that the robot can continue to perform inspection tasks under any circumstances. This redundant positioning solution significantly improves the robustness and reliability of the system.

[0013] This technical solution achieves a high degree of automation, which not only reduces dependence on human intervention, but also greatly reduces the risk of human operation and improves the safety and efficiency of inspection operations.

[0014] By combining a variety of sensors (electromagnetic induction, two-dimensional image sonar, ultra-short baseline positioning device, etc.), the system can work stably in a complex and dynamically changing marine environment, adapt to a variety of uncertain factors such as seawater flow, marine biological interference, etc., to ensure the smooth completion of inspection tasks.

[0015] Due to the high degree of automation, manual inspections and high-risk operations are reduced. The inspection method of underwater robots can not only improve work efficiency, but also effectively reduce operating costs, especially in large-scale submarine cable inspections, which saves labor and equipment maintenance costs in the long term.

[0016] As a preferred technical means: In step 1), the submarine cable feature recognition model is trained using the YOLO V5 recognition algorithm, specifically including: Collect sonar images containing submarine cable features and mark the locations of submarine cable features to generate a label dataset; The data set is divided into a training set and a test set, and the weight file is generated by training the YOLO V5 algorithm.

[0017] By marking the characteristic positions of submarine cables to generate a label data set, training data can be prepared automatically, avoiding the tedious process of manual labeling one by one; improving the efficiency of data preparation, while ensuring the integrity and accuracy of the data set, providing high-quality input for subsequent model training. YOLO V5 has high recognition accuracy in target detection and fast reasoning speed. It can process sonar images in real time during underwater robot inspections and quickly identify submarine cable features. Due to its real-time nature, it can ensure that the robot continuously and accurately adjusts its heading during the inspection process, reducing delays and errors, and ensuring that the inspection task is carried out efficiently. Using YOLO V5 for submarine cable feature recognition training not only improves the recognition efficiency and accuracy of the underwater robot inspection system, but also enhances the system's intelligence and automation capabilities, enabling it to operate stably in complex submarine environments, providing a more efficient and reliable solution for submarine cable inspections.

[0018] As a preferred technical means: in step 2), the horizontal detection two-dimensional image sonar is installed at the front of the robot, and the detection direction is parallel to the travel plane; the oblique downward detection two-dimensional image sonar is a 30° downward detection two-dimensional image sonar, which is installed at the bottom of the robot, and its detection direction forms an inclination of 30° with the horizontal direction; the ultra-short baseline positioning device is integrated into the robot body, and the three-dimensional position is output in real time and yaw angle .

[0019] Dual sonar collaborative detection: eliminate detection blind spots and improve environmental perception capabilities Horizontal sonar (forward deployment): installed in the front of the robot and with the detection direction parallel to the travel plane, it can scan the area in front of and on both sides of the robot in real time, identify obstacles on the travel path (such as sunken ships, reefs, fishing nets, etc.) in advance, and provide advance warning time for obstacle avoidance path planning. It solves the problem of blind spots in the horizontal direction that cannot be covered by traditional single vertical sonar, and reduces the frequency of emergency braking caused by sudden obstacles.

[0020] 30° downward sonar (bottom deployment): The detection direction forms a 30° inclination with the horizontal direction. When the robot maintains the inspection height (such as 2-3 meters away from the submarine cable), it can simultaneously cover the submarine cable body and the seabed area on both sides, realizing the synchronous monitoring of submarine cable position tracking and adjacent geological anomalies (such as exposed and buried submarine cables). Through the optimized inclination design, the image distortion caused by the close-range detection of vertical sonar is avoided, and the accuracy of submarine cable feature recognition is improved.

[0021] Ultra-short baseline (USBL) localization and sensor fusion: high-precision spatial registration USBL positioning advantages: real-time output of the robot's three-dimensional position and yaw angle, positioning accuracy up to centimeter level (typical error <0.1% slant range), providing a benchmark for the conversion of sonar images to the world coordinate system. Eliminate the cumulative error of the traditional inertial navigation system (INS), and ensure the spatial consistency of the obstacle point cloud and the submarine cable position during long-term operations.

[0022] Sensor fusion design: via coordinate transformation matrix T The sonar local coordinate system data is integrated with the USBL global position and posture to achieve unified spatial modeling of submarine cables, obstacles and robot motion states. The EGO-Planner algorithm is supported to generate globally consistent obstacle avoidance paths, avoiding path oscillations caused by coordinate offsets in traditional local planning.

[0023] System-level optimization: improving efficiency and reliability Clear division of tasks: Horizontal sonar focuses on obstacle avoidance, downward sonar focuses on submarine cable tracking, and dual-task parallel processing reduces mode switching delay. Compared with a single sonar alternating working mode, inspection efficiency is higher.

[0024] Enhanced robustness: When the downward sonar fails to identify the submarine cable due to interference from suspended particles, the USBL positioning data can be combined with the electromagnetic induction sensor for rapid repositioning. This significantly reduces the probability of operation interruption due to environmental interference and is suitable for long-term operations in turbid waters.

[0025] As a preferred technical means: In step 4), the output angular velocity of the PID controller satisfy:

[0026] in, is the proportionality coefficient, is the integration coefficient, is the differential coefficient, It is the real-time lateral deviation between the centroid of the submarine cable feature box and the center of the sonar image.

[0027] Proportional term ( ): Real-time response to the lateral deviation between the center of mass of the submarine cable feature box and the center of the image , directly generates an angular velocity proportional to the deviation, ensuring that the robot quickly corrects its heading (for example: when =10pixel, the proportional term outputs the angular velocity immediately to avoid lag). In the simulation test, proportional control shortens the response time of heading correction.

[0028] Integral term ( ): Accumulate historical deviations and eliminate steady-state errors (such as continuous small deviations caused by water disturbances). For example, when the robot continues to deviate from the center line of the submarine cable due to ocean currents, the integral term gradually increases the control amount to completely eliminate the deviation. Integral control reduces the long-term tracking error to within ±1 pixel.

[0029] Differential term ( ): Predict the deviation change trend and suppress overshoot and oscillation. For example, when the robot quickly approaches the center of the submarine cable, the differential term reduces the control amount in advance to avoid repeated corrections caused by inertia. Differential control reduces the overshoot of heading adjustment.

[0030] Anti-water flow disturbance: PID parameters ( K p , T i , T d ) can be adjusted dynamically according to the intensity of the ocean current (e.g., increase K p To improve the response speed), it is suitable for waters with a flow rate of 0-1.5 m / s.

[0031] Robustness verification: In simulated turbid waters, the PID controller can still stably track the submarine cable.

[0032] Closed-loop control architecture: YOLO V5 recognition results ( ) is directly input into the PID controller to form a “perception-decision-execution” closed loop.

[0033] When sonar image noise causes When jumping, the differential term filters high-frequency interference to prevent false operation.

[0034] As a preferred technical means: Step 5) includes the following steps: 5.1) Grayscale the obstacle sonar image and generate a black and white binary image through threshold segmentation; 5.2) Perform Hough line transform detection on the binary image and extract the endpoint coordinates of the longest line segment; 5.3) Generate regular geometric areas based on endpoint coordinates and preset obstacle shapes, including cylindrical or cubic models; 5.4) Calculate pixel resolution based on sonar range and image height, and convert pixel coordinates into three-dimensional coordinates in the camera coordinate system; 5.5) Fuse the ultra-short baseline positioning posture and transform the camera coordinate system point cloud to the world coordinate system through the coordinate transformation matrix; 5.6) Input the world coordinate system point cloud into the local path planning algorithm to generate an obstacle avoidance path and control the robot to move along the planned path.

[0035] This technical solution can effectively separate obstacles from the background, simplify image data and enhance the contrast of the target by graying and thresholding the obstacle sonar image. This provides a clearer and more efficient input for subsequent image processing (such as Hough transform). Through Hough line transform, the longest line segment in the image can be extracted to further determine the boundary of the obstacle, which can help identify linear obstacles in underwater environments and provide key information for generating obstacle geometric models.

[0036] By combining the endpoint coordinates with the preset obstacle shape (such as a cylinder or cube model), the obstacle can be geometrically modeled, making the shape of the obstacle clearer, simplifying the obstacle avoidance calculation in the path planning process, and improving the accuracy of the robot's obstacle avoidance.

[0037] This technical solution can generate different geometric models according to different types of obstacles (such as cylinders, cubes, etc.), has strong adaptability, and can handle various obstacles in complex underwater environments.

[0038] By calculating the pixel resolution and converting the pixel coordinates into three-dimensional coordinates in the camera coordinate system, the location information of the obstacle is made more accurate. This process ensures that the conversion from image to actual space can maintain high accuracy and avoids position deviation caused by factors such as resolution.

[0039] Combined with the real-time pose data (such as robot position and yaw angle) output by the ultra-short baseline positioning device, the accurate position of the obstacle can be obtained in the world coordinate system and converted into data that can be used for path planning through the coordinate transformation matrix, ensuring high positioning accuracy, especially in complex underwater environments, which can improve the stability and safety of the robot's movement.

[0040] By inputting the point cloud data in the world coordinate system into the local path planning algorithm, an obstacle avoidance path can be generated. This algorithm can calculate and optimize the robot's obstacle avoidance path in real time based on the location information of the obstacle, ensuring that the robot can successfully bypass obstacles in the underwater environment and continue to perform its tasks.

[0041] The local path planning algorithm can respond to the appearance of obstacles in real time and quickly generate obstacle avoidance paths. Reducing the delay of path planning allows the robot to respond to sudden obstacles immediately, effectively avoiding collisions and deviations from the original route.

[0042] This technical solution integrates obstacle detection, geometric modeling, three-dimensional spatial positioning and obstacle avoidance path planning into a closed loop, enabling underwater robots to achieve highly autonomous decision-making and intelligent navigation. The robot can automatically adjust its own route based on real-time perception information in complex underwater environments to avoid human intervention and errors. It improves the robot's automation level and reduces the burden on operators, especially in harsh marine environments, where the robot's independent working ability is particularly important.

[0043] Combining sonar image processing with robot posture information can better adapt to the complex and changing underwater environment. The complexity of the underwater environment (such as seabed obstacles, changing water flow, obstacles of different forms, etc.) requires robots to have a high degree of adaptability. Through multi-sensor data fusion and efficient path planning algorithms, robots can effectively cope with these challenges and ensure the smooth progress of submarine cable inspection tasks.

[0044] By continuously monitoring obstacles and generating obstacle avoidance paths in a timely manner, the robot can avoid collision risks and reduce uncertainty and safety hazards in task execution. The robot can dynamically adjust the path according to the changes in obstacles to ensure the continuity and reliability of task execution, reduce inspection pauses or path deviations caused by obstacles, and improve the stability of task completion.

[0045] As a preferred technical means: in step 5.2), the regular geometric region generation includes: When obstacles such as bridge piers are detected, a cylindrical point cloud model is generated; When pile foundation type obstacles are detected, a cubic point cloud model is generated; The model size is determined by the endpoint coordinate spacing and the preset safety distance.

[0046] This technical solution can more accurately describe and characterize the shape of obstacles by using cylindrical and cubic point cloud models for different types of obstacles (such as bridge piers and pile foundations).

[0047] Cylinders and cubes are common regular geometric shapes that can simplify calculations and improve efficiency compared to irregular obstacles. Through this simplified modeling method, the geometric shape of the obstacle can be more easily matched with the surrounding environment, reducing the complexity of calculations and ensuring that the obstacle avoidance path planning can respond in real time.

[0048] By determining the size of the model based on the endpoint coordinate spacing and the preset safety distance, the generated geometric model is more accurate, which can effectively avoid path planning failure or collision risks caused by model errors.

[0049] Add a preset safety distance for each obstacle to ensure a certain safety distance between the robot and the obstacle. Avoid the robot from getting too close to the obstacle, reduce potential collision risks, and improve the safety of task execution.

[0050] After generating a cylindrical or cubic point cloud model, it can accurately reflect the shape and position of obstacles. When input into the path planning algorithm for calculation, the obstacle avoidance path can be planned more accurately to prevent the robot from directly contacting obstacles and ensure that the robot can pass through complex environments smoothly.

[0051] Different types of underwater obstacles, such as bridge piers, pile foundations, and sediments, have different shapes and sizes. By generating corresponding geometric models based on the shape of obstacles, underwater robots can adapt to more complex and changeable underwater environments, improving their detection and obstacle avoidance capabilities in unknown or uncertain environments. The complexity of the underwater environment (such as water flow, seabed geology, obstacle shape, etc.) requires robots to be highly adaptable. By modeling the regular geometry of obstacles, robots can stably perform tasks under various water depths and environmental conditions.

[0052] By accurately modeling and calculating the size of obstacles, the robot can automatically generate obstacle models and perform real-time path planning, greatly improving the automation level of underwater cable inspection tasks. The robot can complete complex obstacle avoidance tasks without human intervention, improving work efficiency. Due to the accuracy and regularity of obstacle modeling, the path planning algorithm can quickly respond to obstacle information, reduce calculation time, and increase obstacle avoidance reaction speed, thereby improving the overall efficiency of task execution.

[0053] Compared with complex irregular shapes, the geometric calculation of cylinders and cubes is simpler and less computationally intensive. This can reduce computing resource consumption, increase system processing speed, and avoid delays and system burdens caused by excessive calculations, thereby ensuring that the underwater robot can complete the obstacle avoidance task in a timely and accurate manner.

[0054] As a preferred technical means: in step 5.4), let the height in the sonar image be , the sonar range is , the physical distance of each pixel in the camera coordinate system for:

[0055] The coordinates of each pixel in the sonar image Convert to camera coordinate system 3D coordinates , the conversion formula is:

[0056] in, and are the origin coordinates of the sonar image, is the current depth value of the sonar; In step 5.5), the position and posture of the underwater robot is obtained based on the ultra-short baseline positioning device: , the camera coordinate system point cloud is transformed to the world coordinate system through the coordinate transformation matrix T. The transformation matrix is:

[0057] Then, the point in the camera coordinate system Convert to the position point in the world coordinate system :

[0058] Thereby, the obstacles in the sonar image are converted into point clouds in the world coordinate system.

[0059] This technical solution can accurately reflect the position of obstacles in the image by converting each pixel coordinate in the sonar image into a three-dimensional coordinate in the camera coordinate system and using the physical distance of each pixel for calculation. This ensures that the position of underwater obstacles is accurately expressed in three-dimensional space, providing a high-quality data foundation for subsequent path planning and obstacle avoidance. After converting each pixel into the corresponding three-dimensional coordinate, an accurate point cloud can be generated. These point cloud data represent the real position and shape of obstacles in the underwater environment, providing very important spatial information for the robot's obstacle avoidance decision.

[0060] This technical solution obtains the current position and posture of the underwater robot (including the robot position and yaw angle) based on the ultra-short baseline positioning device. Through this position and posture information, combined with the transformation matrix T, the point cloud in the camera coordinate system can be accurately converted to the world coordinate system. In this way, the robot can obtain the precise position of obstacles in the global coordinate system based on its real-time position and orientation, enhancing the robot's positioning ability in the environment. Since the underwater robot can dynamically obtain its own position and posture, the conversion process of the point cloud data is real-time, allowing the robot to quickly respond to environmental changes, update obstacle information in real time, and then generate a new obstacle avoidance path.

[0061] By using the coordinate transformation matrix in the calculation, the positioning error that may occur in complex environments can be effectively avoided. This conversion helps to improve the accuracy of obstacle avoidance path planning and ensure that the robot can smoothly avoid obstacles and collisions during the inspection process. By accurately converting the point cloud information of the obstacle to the world coordinate system, it can be directly compared and integrated with other environmental data to provide more reliable input to the path planning algorithm. The robot can adjust the obstacle avoidance path in real time according to the converted point cloud information to ensure a smooth inspection process.

[0062] Obstacles in underwater environments may have various forms and complex geographical features. Through coordinate transformation, the robot can adapt to these changes and update the position and form of obstacles in real time, so that the robot can maintain high navigation capabilities and adaptability in complex and changeable underwater environments. This method can combine the data obtained by different sensors (such as sonar and positioning devices) for processing. By combining the positioning data of the ultra-short baseline positioning device with the data of the sonar image, a more comprehensive and accurate obstacle information can be provided to help the robot make accurate decisions in complex environments.

[0063] In step 5.5), coordinate transformation is performed through the transformation matrix T, which not only reduces the computational burden, but also ensures that the system can still run smoothly under high real-time requirements. The robot can quickly calculate the obstacle information at the current position, reduce delays, and improve the efficiency and safety of task execution. By directly converting from sonar images to the world coordinate system, the error accumulation problem caused by multiple intermediate conversions is avoided, ensuring the accuracy of the point cloud and the stability of the system.

[0064] Through point cloud generation and coordinate conversion, underwater robots can accurately identify obstacles in the surrounding environment and generate obstacle avoidance paths without human intervention, significantly improving the robot's autonomy. The robot can complete tasks independently in complex environments, reduce the risk of human operation, and ensure the efficiency and safety of inspection tasks. Through three-dimensional point clouds, the path planning system can provide robots with a more accurate understanding of the environment, help robots make intelligent decisions, avoid collisions with obstacles, and improve the intelligence level of underwater robots.

[0065] By accurately locating obstacles in the world coordinate system, the robot can accurately predict potential collision risks and avoid them in advance, ensuring the safety of the robot during the mission, especially in dangerous underwater environments, avoiding potential damage and failures.

[0066] As a preferred technical means: in step 5.6), the input parameters of the local path planning algorithm include: the robot's current position and posture information, the three-dimensional spatial distribution of the obstacle point cloud, the preset safe obstacle avoidance distance parameters and the path smoothness optimization weight coefficient.

[0067] By using the robot's current position and posture information as input, the local path planning algorithm can adjust the path in real time according to the robot's motion state. This parameter helps calculate the robot's current dynamic behavior so that the planned path can match the robot's actual motion capabilities, reducing conflicts and unreasonable path selection in path planning. By inputting the three-dimensional spatial distribution information of the obstacle point cloud, the path planning algorithm can accurately understand the obstacle layout in the environment and avoid collisions between the robot and obstacles, especially in complex underwater environments. The point cloud provides comprehensive obstacle data, allowing path planning to be built based on the real environment and reduce errors.

[0068] The preset safe obstacle avoidance distance parameters ensure that when planning the path, the robot always maintains a certain safe distance from obstacles to avoid collision risks. By setting the safe distance, the path planning is more secure and avoids the potential danger caused by the path being too close to obstacles. By introducing the current position posture information and obstacle distribution, the system can adjust the obstacle avoidance strategy in real time to ensure that the robot can cope with different obstacle types and motion states, so that the robot can flexibly avoid obstacles in different seabed environments, improving its autonomy and safety.

[0069] The introduction of the path smoothness optimization weight coefficient enables the path to not only avoid obstacles, but also generate a smooth path that meets the actual execution requirements. A path that is too tortuous or incoherent may cause unstable robot movement, increase energy consumption and affect the accuracy of task execution. By adjusting the path smoothness optimization weight coefficient, the best balance between the smoothness of the path and the obstacle avoidance efficiency can be found. A higher smoothness may lead to a longer path, but it can reduce the frequency of the robot's turning and improve its travel stability; while a lower smoothness may make the path more compact, but it may lead to frequent path adjustments. By setting reasonable weights, path planning can be optimized according to actual needs.

[0070] The smoothness of path planning and the accuracy of obstacle avoidance directly affect the energy consumption of the robot. By balancing the smooth path and the obstacle avoidance path, the robot's unnecessary movement can be effectively reduced, thereby reducing energy consumption, extending inspection time, and improving efficiency. Energy efficiency optimization is particularly important when performing tasks for a long time.

[0071] As a preferred technical means: also includes exception handling mechanism: When no submarine cable features are detected in N consecutive sonar images, the system automatically switches to electromagnetic induction guidance mode. When local path planning fails, emergency hovering is triggered and manual intervention is requested; Monitor the battery level in real time and automatically generate a return path when it falls below the threshold.

[0072] When the sonar image fails to detect the characteristics of the submarine cable for N consecutive frames, the system automatically switches to the electromagnetic induction guidance mode. This measure ensures that the robot can continue to complete the submarine cable positioning by relying on electromagnetic signals when the visual information is missing or unclear. The electromagnetic induction sensor is not affected by factors such as light and water flow. It can solve the situation where the sonar image cannot provide effective data due to environmental factors or equipment problems to a certain extent, ensuring the continuity and stability of the submarine cable inspection task.

[0073] When faced with unpredictable environments or technical failures, the automated exception handling mechanism ensures that the system responds quickly and takes appropriate measures. For example, when the submarine cable cannot be identified continuously, switching to the backup electromagnetic induction guidance mode can ensure that the robot can continue to work and reduce inspection interruptions caused by failures or data loss.

[0074] When local path planning fails, the system will immediately trigger an emergency hover and request manual intervention. This prevents the robot from continuing to run along an invalid path, avoiding wasting time and resources and reducing the risk of collision. Through manual intervention, problems can be quickly analyzed and path planning can be adjusted or optimized to ensure smooth inspection tasks.

[0075] Real-time monitoring of battery power and automatic generation of a return path when it falls below a preset threshold. This feature ensures that the robot can return in time when the battery is low, avoiding the situation where it cannot be recovered. Automatic return at low battery not only protects the robot itself, but also reduces the need for human intervention and reduces the risk of mission failure due to battery exhaustion.

[0076] Real-time monitoring of battery power, sonar image processing status, and path planning execution status can help the system respond quickly to different challenges. For example, the automatic mode switching and path return mechanism can respond to possible emergencies in real time, avoid mission interruption or equipment damage, and ensure mission completion.

[0077] When the system cannot identify the submarine cable or the path planning fails, the robot avoids possible risks through backup plans (such as electromagnetic induction guidance mode and emergency hovering). This can reduce the danger or damage that may be caused by system failure, especially in complex underwater environments where the robot's movement is affected by multiple factors.

[0078] When the sonar image cannot identify the characteristics of the submarine cable, the system automatically switches to backup mode; when local path planning fails, the system automatically hovers and issues a request for manual intervention, reducing the need for manual operation. Especially in long inspection tasks, it can greatly reduce the frequency and workload of manual intervention.

[0079] Another technical solution of the present invention is: an underwater robot submarine cable inspection system based on two-dimensional image sonar, which adopts the above-mentioned underwater robot submarine cable inspection method based on two-dimensional image sonar, and the system includes: Sensing unit: It includes electromagnetic induction sensor, horizontal detection two-dimensional image sonar, 30° downward detection two-dimensional image sonar, and ultra-short baseline positioning device; the horizontal detection two-dimensional image sonar is installed at the front of the robot, and the detection direction is parallel to the travel plane; the 30° downward detection two-dimensional image sonar is installed at the bottom of the robot, forming a 30° inclination with the horizontal direction; Control unit: used to perform the following operations: in search mode, calculate the position of the submarine cable according to the electromagnetic induction sensor signal and control the robot to move above the submarine cable; in tracking mode, adjust the robot's heading in real time through the PID controller based on the submarine cable feature recognition results of the oblique downward sonar image; Image processing unit: including: The submarine cable feature recognition module uses the YOLO V5 algorithm to detect submarine cable features in sonar images; The obstacle point cloud generation module performs grayscale conversion, threshold segmentation and Hough line transform processing on the horizontal sonar image, and generates a three-dimensional point cloud based on the preset obstacle shape; Path planning unit: Based on the posture data of the ultra-short baseline positioning device, the obstacle point cloud is converted to the world coordinate system, and the obstacle avoidance path is generated through the EGO-Planner algorithm; Exception handling unit: used to switch to electromagnetic guidance mode and trigger return or manual intervention instructions when sonar fails or path planning fails.

[0080] The system integrates multiple high-precision sensors (such as electromagnetic induction sensors, horizontal detection two-dimensional image sonar, oblique downward detection two-dimensional image sonar, and ultra-short baseline positioning devices) to enable the robot to obtain comprehensive perception capabilities in complex environments. Each sensor plays its own strengths and works together to ensure that the system can efficiently and accurately complete the submarine cable inspection task.

[0081] The sensing unit works closely with the control unit. The electromagnetic induction sensor is used for submarine cable search, the sonar image is used for submarine cable tracking and obstacle detection, and the positioning device provides three-dimensional posture information in real time. The entire system forms an efficient closed loop of information collection, processing and feedback, ensuring the real-time and accuracy of submarine cable inspections.

[0082] The YOLO V5 algorithm is used for submarine cable feature recognition, combined with a deep learning algorithm, to more accurately and efficiently identify submarine cable features. YOLO V5 has high real-time performance, adapts to dynamically changing environments, can process complex sonar images, and improves the accuracy and speed of submarine cable detection.

[0083] The system can accurately guide the robot to inspect along the cable path by collecting characteristic images of the submarine cable through the two-dimensional image sonar that detects obliquely downwards, and then adjusting the robot's heading with the PID controller. This process is highly automated, reduces manual intervention, and ensures the continuity and accuracy of the inspection.

[0084] By processing sonar images (grayscale, threshold segmentation, Hough transform), point cloud data of obstacles is generated, and combined with preset shapes (such as cylinders and cubes), the accuracy of obstacle recognition is improved. It can better identify obstacles in complex underwater environments and generate accurate three-dimensional point cloud information, providing basic data for subsequent path planning.

[0085] By converting the obstacle point cloud into the world coordinate system and combining it with the EGO-Planner algorithm, the obstacle avoidance path can be dynamically planned. The EGO-Planner algorithm has strong real-time and flexibility, and can effectively avoid obstacles in complex environments to ensure the smooth progress of inspection tasks.

[0086] When the system faces problems such as sonar failure or path planning failure, the exception handling unit can automatically switch to electromagnetic guidance mode and trigger return or manual intervention instructions. This ensures that the system can respond quickly when an exception occurs and make appropriate adjustments according to preset rules to avoid mission interruption or robot damage.

[0087] The system's automatic switching and intervention mechanism makes the inspection process more autonomous and intelligent, reduces reliance on manual operations, and improves the efficiency and reliability of task execution. Even in complex or dangerous environments, the robot can maintain high work efficiency and safety.

[0088] Through the real-time posture data provided by the ultra-short baseline positioning device, the system can maintain precise control of the robot in a dynamic environment, thereby ensuring that it is not affected by external disturbances when performing tasks, enhancing the reliability of path planning and the realization of robot autonomy.

[0089] The combination of obstacle point cloud generation and local path planning enables the system to quickly avoid sudden obstacles, ensuring that the robot can pass safely and avoid collisions or deviations from the inspection route.

[0090] When the system fails to obtain sufficient information through sonar images, it automatically switches to electromagnetic induction guidance mode to ensure that the robot can continue to complete the task, allowing the robot to conduct effective submarine cable inspections even when visual signals are limited.

[0091] The system can simultaneously perform multiple tasks such as submarine cable search, tracking, obstacle detection and path planning, greatly improving inspection efficiency. The collaborative work between different modules makes the entire inspection process smoother and more efficient.

[0092] Functions such as exception handling and automatic return effectively reduce the user's intervention in the robot's operation, allowing users to monitor the robot without frequent intervention, thus improving the user experience. In addition, the automatic feedback mechanism can ensure that the robot's task progress and status can be delivered to the user in a timely and accurate manner.

[0093] The speed and accuracy of task completion are improved through automated task execution (such as submarine cable detection, path planning, obstacle avoidance, exception handling, etc.). In complex underwater environments, the robot can maintain efficient operation and avoid time waste caused by human factors or system errors.

[0094] The system's exception handling and path planning mechanisms effectively reduce energy consumption. In particular, the low-battery return function can ensure that the robot returns in time when energy is insufficient, reducing resources wasted due to unfinished tasks.

[0095] Beneficial effects: This technical solution integrates high-precision sonar imaging, deep learning algorithms and autonomous motion planning, breaking through the limitations of existing technologies in detection efficiency, environmental adaptability and automation level; effectively improving inspection efficiency, reducing costs, improving the accuracy and reliability of submarine cable maintenance, and ensuring the normal operation of submarine cable networks. Specific: 1. Using the submarine cable feature recognition model to identify submarine cable features in sonar images collected by underwater robots can effectively improve the accuracy and speed of submarine cable recognition, and the automation level has been significantly improved. The algorithm can quickly and accurately identify the characteristic areas of submarine cables, especially in underwater environments, where the edge and texture features of submarine cables in sonar images are relatively complex. The submarine cable feature recognition model can autonomously learn the key features of submarine cables during the training process, thereby improving the automation level of the inspection process and reducing the reliance on manual operations.

[0096] 2. Secondly, the yaw angle of the underwater robot is adjusted in real time through the PID control algorithm, so that it can accurately align directly above the submarine cable for inspection. This deviation calculation based on the centroid of the submarine cable feature box in the sonar image and the center position of the image enables the underwater robot to dynamically adjust its course according to the actual situation and stay above the submarine cable, avoiding the situation where the robot loses its target in the traditional method. In terms of obstacle bypassing, the present invention effectively solves the bypassing problem of the underwater robot when encountering obstacles during the inspection process by converting the sonar image into a three-dimensional point cloud and combining it with a local path planning algorithm. The method extracts the edge information of the obstacle and converts it into complete obstacle point cloud data in three-dimensional space through image completion, graying, thresholding and other processing means, and uses the planning algorithm to generate an obstacle avoidance path, so that the underwater robot can safely and efficiently bypass the obstacle and continue to patrol along the cable path, avoiding the influence of obstacles and improving the inspection efficiency and path flexibility. The present invention also combines the collaborative work of electromagnetic induction sensors and two-dimensional image sonars to improve the inspection efficiency. When the sonar image cannot identify the submarine cable, the electromagnetic induction sensor can help calculate the position and posture of the submarine cable, thereby supplementing the blind spots of the sonar image and ensuring that the underwater robot can efficiently and stably complete the submarine cable inspection task in various complex environments. At the same time, a function based on electromagnetic induction sensors is designed to re-find the submarine cable, avoiding inspection blind spots and ensuring full coverage of the operation. During the entire inspection process, the underwater robot can not only maintain an accurate heading through continuous sensor data input and real-time PID control, but also intelligently judge and handle obstacles, avoiding the situation where the robot cannot continue to operate due to complex environments. This enables the underwater robot to operate stably in complex underwater environments, reduces human intervention, and improves the safety of operations.

[0097] 3. Finally, the technical method of the present invention is not only suitable for the inspection tasks of submarine cables, but can also be widely used in the inspection of other types of underwater structures, such as underwater pipelines, bridges, etc. It has strong versatility and adaptability, and provides strong support for the application of underwater robots in marine engineering. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] Figure 1 It is a flow chart of the present invention.

[0099] Figure 2 It is a process diagram of generating point cloud from multi-beam sonar image of the present invention.

[0100] Figure 3 This is a simulation test environment diagram of underwater robot submarine cable exposed section inspection based on ROS Gazebo of the present invention.

[0101] Figure 4 It is an obstacle avoidance motion trajectory diagram of an underwater robot with sonar image completion according to the present invention.

[0102] Figure 5 It is the ROV obstacle avoidance motion trajectory diagram without sonar image completion of the present invention. DETAILED DESCRIPTION

[0103] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings.

[0104] Embodiment 1: like Figure 1 As shown, the present invention comprises the following steps: S1: Train the images containing submarine cable features based on the YOLO V5 recognition algorithm. Specifically, first collect images containing submarine cable features, make an image dataset, and then make a corresponding label dataset containing the locations of submarine cable features in the images. After the dataset preparation is completed, it is divided into a training set and a test set. The submarine cable features are trained using the YOLO V5 recognition algorithm, and the corresponding weight file is obtained after training.

[0105] S2: Underwater robot sensor configuration. Equipped with electromagnetic induction sensors, two-dimensional image sonar with horizontal detection direction, two-dimensional image sonar with 30° downward detection direction, and ultra-short baseline positioning device.

[0106] S3: After the underwater robot is launched into the water, it enters the submarine cable search mode, and infers the position and posture of the nearest submarine cable based on the submarine cable signal detected by the electromagnetic induction sensor. Then the operator operates the underwater robot to move above the submarine cable.

[0107] S4: After the underwater robot reaches a certain height above the submarine cable, it identifies the submarine cable in the sonar image based on the YOLO V5 recognition algorithm and the trained weight file (using image data collected by a two-dimensional image sonar with a detection direction of 30° downward). The centroid position of the submarine cable feature box identified in the image is and , the center position of the sonar image is and , so the difference between the centroid of the submarine cable feature box and the center position of the sonar image can be obtained ,Right now:

[0108] Next, the angular velocity of the underwater robot is controlled according to the PID control method:

[0109] in, is the proportional coefficient of the PID controller, is the integral coefficient of the controller, is the differential coefficient of the controller.

[0110] The forward speed of the underwater robot during the submarine cable inspection is , set to a constant, when there is a difference between the centroid of the submarine cable feature box and the center of the sonar image, based on the calculated The yaw angle of the underwater robot is controlled to return to the top of the submarine cable.

[0111] S5: When an underwater robot is inspecting cables along the coast, it needs to perform local path planning to avoid obstacles when obstacles appear in the horizontal two-dimensional image sonar. In order to improve the success rate of local path planning and reduce the risk of collision with obstacles during obstacle avoidance, point cloud generation is completed for the edge of obstacles based on prior knowledge. For example, for common structures such as bridge piers or pile foundations in underwater environments, the shape of the structure can be input into the algorithm before the operation, thereby generating a safer and smoother path. Figure 2 As shown, the process is implemented as follows: S5.1: grayscale the collected sonar image to highlight the shape of obstacles in the sonar image; S5.2: Thresholding the grayscale image, setting the pixel values ​​greater than 200 in the image to 255, and setting the pixel values ​​less than 200 to 0, thereby obtaining a black-and-white image containing the edge of the obstacle; S5.3: Perform Hough line transform detection on the black and white image generated in the previous step, identify the line segments in the image, select the longest line segment and extract its endpoint coordinates; S5.4: Draw a square or circular area based on the endpoint coordinates extracted in the previous step and the obstacle shape input in advance; S5.5: Convert sonar images into point clouds: Assume that the height in the sonar image is , the sonar range is , since the resolution of the image determines the actual distance represented by each pixel, the physical distance of each pixel in the camera coordinate system is for:

[0112] For each pixel in the sonar image , it can be converted from pixel coordinates to three-dimensional coordinate points in the camera coordinate system through the following formula :

[0113] in, and are the pixel coordinates in the sonar image, and are the origin coordinates of the sonar image, is the depth value at which the sonar is located.

[0114] The three-dimensional coordinate point calculated in the camera coordinate system , needs to be transformed into the world coordinate system through the camera’s posture transformation. The posture of the underwater robot can be obtained by the ultra-short baseline positioning device as follows: , the transformation matrix can be expressed as:

[0115] Then, the point in the camera coordinate system Convert to the position point in the world coordinate system :

[0116] According to the above process, the obstacles in the sonar image can be converted into a point cloud in the world coordinate system. After the point cloud is input into the EGO-Planner local path planning algorithm, the local path points in the world coordinate system can be output. The underwater robot can bypass the obstacles by tracking the planned local path points.

[0117] S5.6: After the underwater robot bypasses the obstacle, if the sonar image does not find the submarine cable, it searches for the submarine cable based on the electromagnetic induction sensor until the submarine cable is identified in the sonar image, and then continues the operation until it reaches the target point.

[0118] This technical solution combines electromagnetic induction sensors with image sonar. When the sonar image cannot identify the submarine cable, the electromagnetic induction sensor is used to supplement the submarine cable location information, solving the blind spot problem that may occur in traditional technology and improving the full coverage and accuracy of inspection tasks.

[0119] like Figure 3 The figure shows the underwater robot submarine cable exposed section inspection simulation test environment based on ROS Gazebo. The underwater robot is equipped with a horizontal two-dimensional image sonar that can monitor the surrounding environment in real time. When obstacles appear, the robot automatically avoids them through obstacle avoidance algorithms.

[0120] Figure 4 The obstacle avoidance motion trajectory of the underwater robot under the condition of sonar image completion is demonstrated. Through the completed sonar image, the robot can accurately identify and bypass obstacles, successfully plan an obstacle avoidance path, avoid direct collision with obstacles, and ensure the smooth progress of the inspection task. This solution is based on the obstacle detection and three-dimensional point cloud conversion method of sonar image graphic completion. The completed sonar image is converted into complete three-dimensional obstacle point cloud data, and combined with the EGO-Planner algorithm for local path planning, so that the underwater robot can achieve intelligent obstacle avoidance in complex environments, improving its work efficiency and safety.

[0121] In contrast, Figure 5 The obstacle avoidance motion trajectory of ROV (remotely operated underwater robot) when encountering obstacles is shown in the absence of sonar image completion. Figure 5 At point A, the local path planning failed to adjust in time, causing the planned target trajectory to incorrectly guide the robot towards the obstacle, creating a safety risk. This shows that the lack of complete environmental information (such as sonar image completion) may lead to path planning errors, which in turn affects the obstacle avoidance effect and increases the risk of robot collision.

[0122] This embodiment has the following advantages: 1. Multimodal sensor fusion to improve environmental perception and positioning accuracy Electromagnetic induction + sonar collaborative positioning (steps 2, 3, 6): The electromagnetic induction sensor is used to quickly calculate the initial position of the submarine cable, combined with the visual tracking of the downward sonar to form a two-layer perception architecture of "electromagnetic coarse positioning → sonar fine correction", which solves the problem that traditional single sensors are prone to failure in complex seabed environments (such as blurred sonar imaging in turbid waters and interference with electromagnetic signals).

[0123] Ultra-short baseline positioning device (step 2): provides high-precision three-dimensional pose data for the underwater robot, realizes accurate conversion between point cloud coordinates and world coordinate system (step 5), and avoids the cumulative error problem of traditional inertial navigation.

[0124] 2. Deep learning-based submarine cable identification significantly improves detection efficiency and robustness (steps 1 and 4) Advantages of the YOLO V5 algorithm: The model is trained with a submarine cable feature dataset and can quickly identify submarine cable outlines in low-resolution sonar images. Compared with the traditional manual threshold segmentation method, it can effectively improve the recognition accuracy.

[0125] Adaptive feature extraction: The model autonomously learns the texture, edge and geometric features of the submarine cable in the sonar image, effectively overcoming interference such as uneven underwater lighting and suspended object occlusion, and reducing the false detection rate.

[0126] 3. Closed-loop control combined with dynamic obstacle avoidance to ensure inspection stability and safety PID real-time heading correction (step 4): Based on the dynamic feedback of the centroid deviation of the feature box in the sonar image, the PID controller outputs the angular velocity to achieve rapid convergence of the heading error and ensure that the robot stably tracks the centerline of the submarine cable.

[0127] EGO-Planner intelligent obstacle avoidance (step 5): The obstacle sonar image is converted into a 3D point cloud and a smooth obstacle avoidance path is generated in combination with the robot kinematic constraints.

[0128] 4. Redundant fault-tolerant mechanism to enhance system robustness (steps 3 and 6) Electromagnetic-sonar mode switching: When the sonar loses lock due to obstacles or environmental interference (step 6), it automatically switches to electromagnetic induction guidance mode to avoid operation interruption caused by the robot "losing the target" in traditional inspections.

[0129] Exception handling logic: If no submarine cable features are detected in N consecutive frames, an emergency return or manual intervention request is triggered to reduce the risk of mission failure.

[0130] 5. Engineering practicality and cost advantages Hardware configuration optimization: The horizontal and downward sonars have clear division of labor (step 2). The horizontal sonar is dedicated to obstacle detection, while the downward sonar focuses on submarine cable tracking to reduce multi-sensor data conflicts.

[0131] Automated operation process: The entire process does not require manual control (search → track → avoid obstacles → recover), which effectively improves the efficiency of a single inspection and greatly saves labor costs.

[0132] Summarize Through multi-sensor fusion, deep learning drive, closed-loop control optimization, intelligent obstacle avoidance and redundant design, the four core problems of positioning drift, environmental interference, obstacle avoidance failure and high manual dependence in underwater robot submarine cable inspection are systematically solved, combining technological innovation and engineering implementation value.

[0133] Embodiment 2: An underwater robot submarine cable inspection system based on two-dimensional image sonar, comprising: 1. Sensing unit: It includes electromagnetic induction sensor, horizontal detection two-dimensional image sonar, 30° downward detection two-dimensional image sonar, and ultra-short baseline positioning device; the horizontal detection two-dimensional image sonar is installed at the front of the robot, and the detection direction is parallel to the travel plane; the 30° downward detection two-dimensional image sonar is installed at the bottom of the robot, forming a 30° inclination with the horizontal direction; This system integrates multiple high-precision sensors (such as electromagnetic induction sensors, horizontal detection two-dimensional image sonar, oblique downward detection two-dimensional image sonar, and ultra-short baseline positioning devices) to enable the robot to obtain comprehensive perception capabilities in complex environments. Each sensor plays its own strengths and works together to ensure that the system can efficiently and accurately complete the submarine cable inspection task.

[0134] 2. Control unit: used to perform the following operations: in search mode, calculate the position of the submarine cable according to the electromagnetic induction sensor signal and control the robot to move above the submarine cable; in tracking mode, adjust the robot's heading in real time through the PID controller based on the submarine cable feature recognition results of the oblique downward sonar image; The sensing unit works closely with the control unit. The electromagnetic induction sensor is used for submarine cable search, the sonar image is used for submarine cable tracking and obstacle detection, and the positioning device provides three-dimensional posture information in real time. The entire system forms an efficient closed loop of information collection, processing and feedback, ensuring the real-time and accuracy of submarine cable inspections.

[0135] The system can accurately guide the robot to inspect along the cable path by collecting characteristic images of the submarine cable through the two-dimensional image sonar that detects obliquely downwards, and then adjusting the robot's heading with the PID controller. This process is highly automated, reduces manual intervention, and ensures the continuity and accuracy of the inspection.

[0136] 3. Image processing unit: including: The submarine cable feature recognition module uses the YOLO V5 algorithm to detect submarine cable features on sonar images; the YOLOV5 algorithm is used for submarine cable feature recognition, combined with the deep learning algorithm, to more accurately and efficiently identify submarine cable features. YOLOV5 has high real-time performance, adapts to dynamically changing environments, can process complex sonar images, and improve the accuracy and speed of submarine cable detection.

[0137] The obstacle point cloud generation module performs grayscale, threshold segmentation and Hough line transformation on the horizontal sonar image, and generates a three-dimensional point cloud in combination with the preset obstacle shape. By processing the sonar image (grayscale, threshold segmentation, Hough transformation) to generate point cloud data of obstacles, combined with preset shapes (such as cylinders and cubes), the accuracy of obstacle recognition is improved, and obstacles in complex underwater environments can be better identified. It can also generate accurate three-dimensional point cloud information, providing basic data for subsequent path planning.

[0138] 4. Path planning unit: Based on the position data of the ultra-short baseline positioning device, the obstacle point cloud is converted to the world coordinate system, and the obstacle avoidance path is generated through the EGO-Planner algorithm; By converting the obstacle point cloud into the world coordinate system and combining it with the EGO-Planner algorithm, the obstacle avoidance path can be dynamically planned. The EGO-Planner algorithm has strong real-time and flexibility, and can effectively avoid obstacles in complex environments to ensure the smooth progress of inspection tasks.

[0139] Through the real-time posture data provided by the ultra-short baseline positioning device, the system can maintain precise control of the robot in a dynamic environment, thereby ensuring that it is not affected by external disturbances when performing tasks, enhancing the reliability of path planning and the realization of robot autonomy.

[0140] The combination of obstacle point cloud generation and local path planning enables the system to quickly avoid sudden obstacles, ensuring that the robot can pass safely and avoid collisions or deviations from the inspection route.

[0141] 5. Exception handling unit: used to switch to electromagnetic guidance mode and trigger return or manual intervention instructions when sonar fails or path planning fails.

[0142] When the system faces problems such as sonar failure or path planning failure, the exception handling unit can automatically switch to electromagnetic guidance mode and trigger return or manual intervention instructions. This ensures that the system can respond quickly when an abnormality occurs and make appropriate adjustments according to preset rules to avoid task interruption or robot damage. The system's automatic switching and intervention mechanism makes the inspection process more autonomous and intelligent, reduces dependence on manual operation, and improves the efficiency and reliability of task execution. Even in complex or dangerous environments, the robot can maintain high work efficiency and safety.

[0143] When the system fails to obtain sufficient information through sonar images, it automatically switches to electromagnetic induction guidance mode to ensure that the robot can continue to complete the task, allowing the robot to conduct effective submarine cable inspections even when visual signals are limited.

[0144] Functions such as exception handling and automatic return effectively reduce the user's intervention in the robot's operation, allowing users to monitor the robot without frequent intervention, thus improving the user experience. In addition, the automatic feedback mechanism can ensure that the robot's task progress and status can be delivered to the user in a timely and accurate manner.

[0145] This embodiment has the following advantages: 1. Multimodal sensor fusion to achieve full-scenario environmental perception Electromagnetic + sonar collaborative positioning: Electromagnetic induction sensors quickly and roughly locate submarine cables (especially suitable for turbid waters or scenes with blurred sonar imaging); The sonar at an angle of 30° downward provides high-precision visual tracking, forming a closed loop of "electromagnetic coarse positioning → sonar precise tracking".

[0146] Ultra-short baseline positioning device: provides centimeter-level three-dimensional posture in real time, supports precise mapping of point clouds in the world coordinate system, and avoids path deviation caused by accumulated errors of inertial navigation.

[0147] 2. Intelligent image processing to improve detection efficiency and accuracy YOLO V5 submarine cable identification module: Autonomously extract the texture and edge features of submarine cables in sonar images based on deep learning, with a recognition accuracy of >98%; Supports real-time processing to meet high-speed inspection needs.

[0148] Obstacle point cloud generation module: The geometric features of obstacles are extracted through Hough line transform, and a regular point cloud is generated by combining the preset model (cylinder / cube); Grayscale + threshold segmentation effectively suppresses suspended object noise and reduces the false detection rate.

[0149] 3. Dynamic path planning and closed-loop control to ensure operational safety EGO-Planner obstacle avoidance algorithm: Generate smooth obstacle avoidance paths in real time based on 3D point clouds, with a high obstacle avoidance success rate; Supports dynamic obstacle tracking and replanning with low response delay.

[0150] PID heading control: The angular velocity is dynamically adjusted according to the centroid deviation of the sonar image, and the heading error converges quickly, ensuring that the robot stably tracks the centerline of the submarine cable.

[0151] 4. Redundant design and exception handling to enhance system robustness Dual-mode seamless switching: Automatically switch to electromagnetic guidance mode when sonar fails, effective mission interruption rate; Path planning failure triggers emergency hovering + manual intervention request to avoid collision risk.

[0152] Hardware fault-tolerant design: The horizontal and downward sonars have clear division of labor (obstacle detection vs. cable tracking), effectively reducing the probability of sensor conflict; Ultra-short baseline positioning and inertial navigation back up each other to improve the reliability of positioning data.

[0153] 5. Engineering economy and scalability Cost advantage: Automated operations reduce manual intervention, significantly reducing the cost of a single inspection; The modular design supports hardware replacement (such as upgrading to the YOLO V7 algorithm) to avoid overall system elimination.

[0154] Adaptable expansion: The point cloud generation module is compatible with other obstacle models (such as pipelines and sunken ships), shortening the expansion development cycle; The EGO-Planner algorithm supports multi-robot collaborative obstacle avoidance, laying the foundation for cluster inspection.

[0155] Summarize This system systematically solves the four core pain points of traditional underwater inspections, namely positioning drift, inefficient obstacle avoidance, high dependence on manual labor, and poor environmental adaptability, through deep fusion of multiple sensors, intelligent algorithm drive, dynamic closed-loop control, and redundant fault-tolerant design. It achieves breakthrough improvements in detection accuracy, operational efficiency, safety, and economy, and provides a new generation of intelligent solutions for submarine cable maintenance.

[0156] The underwater robot submarine cable inspection method and system based on two-dimensional image sonar shown above is a specific embodiment of the present invention, which has embodied the essential characteristics and progress of the present invention. It can be equivalently modified according to actual use needs under the guidance of the present invention, and all are within the protection scope of this scheme.

Claims

1. A submarine cable inspection method using an underwater robot based on two-dimensional image sonar, characterized in that: The following steps are involved: 1) Build a submarine cable feature recognition model, train the model on a sonar image set containing submarine cable features, and output a weight file; 2) Equipped with underwater robot sensing system, including electromagnetic induction sensor, horizontal detection two-dimensional image sonar, oblique downward detection two-dimensional image sonar and ultra-short baseline positioning device; 3) After entering the water, the underwater robot enters the submarine cable search mode, detects the electromagnetic signal of the submarine cable through the electromagnetic induction sensor, calculates the nearest position of the submarine cable and guides the robot to move above the submarine cable; 4) When the underwater robot reaches the top of the submarine cable, it enters the submarine cable tracking mode, uses the oblique downward detection two-dimensional image sonar to collect images in real time, identifies the submarine cable features based on the trained submarine cable feature recognition model, calculates the lateral deviation between the center of mass of the feature box and the center of the image, and adjusts the robot's heading through the PID controller output angular velocity so that it patrols along the cable path; 5) During the inspection process, when the two-dimensional image sonar with the horizontal detection direction detects an obstacle, the sonar image is grayed, thresholded and Hough line transformed to extract the edge information of the obstacle and generate point cloud data in combination with the preset obstacle shape; the point cloud data is fused with the underwater robot's posture information and converted to the world coordinate system, and the local path planning algorithm is input to generate an obstacle avoidance path, and the robot is controlled to detour along the obstacle avoidance path; 6) After the robot bypasses the obstacle, if the sonar image does not detect the characteristics of the submarine cable, the position of the submarine cable will be relocated based on the electromagnetic induction sensor until the sonar image restores the submarine cable recognition and then continues the inspection until it reaches the target point.

2. The method for submarine cable inspection using an underwater robot based on two-dimensional image sonar according to claim 1, characterized in that: In step 1), the submarine cable feature recognition model is trained using the YOLO V5 recognition algorithm, which includes: Collect sonar images containing submarine cable features and mark the locations of submarine cable features to generate a label dataset; The data set is divided into a training set and a test set, and the weight file is generated by training the YOLO V5 algorithm.

3. The method for submarine cable inspection using an underwater robot based on two-dimensional image sonar according to claim 2, characterized in that: In step 2), The horizontal detection two-dimensional image sonar is installed at the front of the robot, and the detection direction is parallel to the travel plane; The oblique downward detection two-dimensional image sonar is a 30° downward detection two-dimensional image sonar, which is installed at the bottom of the robot, and its detection direction forms an inclination angle of 30° with the horizontal direction; The ultra-short baseline positioning device is integrated into the robot body and outputs the three-dimensional position in real time and yaw angle .

4. The method for submarine cable inspection using an underwater robot based on two-dimensional image sonar according to claim 3, characterized in that: In step 4), the output angular velocity of the PID controller is satisfy: in, is the proportionality coefficient, is the integration coefficient, is the differential coefficient, It is the real-time lateral deviation between the centroid of the submarine cable feature box and the center of the sonar image.

5. The method for submarine cable inspection using an underwater robot based on two-dimensional image sonar according to claim 4, characterized in that: Step 5) includes the following steps: 5.1) Grayscale the obstacle sonar image and generate a black and white binary image through threshold segmentation; 5.2) Perform Hough line transform detection on the binary image and extract the endpoint coordinates of the longest line segment; 5.3) Generate regular geometric areas based on endpoint coordinates and preset obstacle shapes, including cylindrical or cubic models; 5.4) Calculate pixel resolution based on sonar range and image height, and convert pixel coordinates into three-dimensional coordinates in the camera coordinate system; 5.5) Fuse the ultra-short baseline positioning posture and transform the camera coordinate system point cloud to the world coordinate system through the coordinate transformation matrix; 5.6) Input the world coordinate system point cloud into the local path planning algorithm to generate an obstacle avoidance path and control the robot to move along the planned path.

6. The method for submarine cable inspection using an underwater robot based on two-dimensional image sonar according to claim 5, characterized in that: In step 5.2), the regular geometric region generation includes: When obstacles such as bridge piers are detected, a cylindrical point cloud model is generated; When pile foundation type obstacles are detected, a cubic point cloud model is generated; The model size is determined by the endpoint coordinate spacing and the preset safety distance.

7. The method for submarine cable inspection using an underwater robot based on two-dimensional image sonar according to claim 6, characterized in that: In step 5.4), let the height in the sonar image be , the sonar range is , the physical distance of each pixel in the camera coordinate system for: The coordinates of each pixel in the sonar image Convert to camera coordinate system 3D coordinates , the conversion formula is: in, and are the origin coordinates of the sonar image, is the current depth value of the sonar; In step 5.5), the position and posture of the underwater robot is obtained based on the ultra-short baseline positioning device: , the camera coordinate system point cloud is transformed to the world coordinate system through the coordinate transformation matrix T. The transformation matrix is: Then, the point in the camera coordinate system Convert to the position point in the world coordinate system : Thereby, the obstacles in the sonar image are converted into point clouds in the world coordinate system.

8. The method for submarine cable inspection using an underwater robot based on two-dimensional image sonar according to claim 7, characterized in that: In step 5.6), the input parameters of the local path planning algorithm include: the robot's current position and posture information, the three-dimensional spatial distribution of the obstacle point cloud, the preset safe obstacle avoidance distance parameters and the path smoothness optimization weight coefficient.

9. The method for submarine cable inspection using an underwater robot based on two-dimensional image sonar according to claim 1, characterized in that: It also includes exception handling mechanisms: When no submarine cable features are detected in N consecutive sonar images, the system automatically switches to electromagnetic induction guidance mode. When local path planning fails, emergency hovering is triggered and manual intervention is requested; Monitor the battery level in real time and automatically generate a return path when it falls below the threshold.

10. An underwater robot submarine cable inspection system based on two-dimensional image sonar, characterized in that: The underwater robot submarine cable inspection method based on two-dimensional image sonar according to any one of claims 1 to 9 is adopted, and the system comprises: Sensing unit: It includes electromagnetic induction sensor, horizontal detection two-dimensional image sonar, 30° downward detection two-dimensional image sonar, and ultra-short baseline positioning device; the horizontal detection two-dimensional image sonar is installed at the front of the robot, and the detection direction is parallel to the travel plane; the 30° downward detection two-dimensional image sonar is installed at the bottom of the robot, forming a 30° inclination with the horizontal direction Control unit: used to perform the following operations: in search mode, calculate the position of the submarine cable according to the electromagnetic induction sensor signal and control the robot to move above the submarine cable; in tracking mode, adjust the robot's heading in real time through the PID controller based on the submarine cable feature recognition results of the oblique downward sonar image; Image processing unit: including: The submarine cable feature recognition module uses the YOLO V5 algorithm to detect submarine cable features in sonar images; The obstacle point cloud generation module performs grayscale conversion, threshold segmentation and Hough line transform processing on the horizontal sonar image, and generates a three-dimensional point cloud based on the preset obstacle shape; Path planning unit: Based on the posture data of the ultra-short baseline positioning device, the obstacle point cloud is converted to the world coordinate system, and the obstacle avoidance path is generated through the EGO-Planner algorithm; Exception handling unit: used to switch to electromagnetic guidance mode and trigger return or manual intervention instructions when sonar fails or path planning fails.

Images