Robot patrol planning system for offshore platform jacket marine biology monitoring

Abstract

This invention relates to the field of inspection planning technology and discloses a robotic inspection planning system for monitoring marine life on jackets of offshore platforms. The system includes: a shooting mode optimization module: when the quality assessment indicators of marine life monitoring images in risk areas fail to meet standards, it identifies multi-dimensional labels for the risk areas; based on the quality assessment indicators and the identified multi-dimensional labels, it generates a composite response mechanism to determine robot parameters for the risk areas; for non-risk areas, it generates a standard response mechanism based on the quality assessment indicators to determine robot parameters and executes a retake mode to obtain the main image; and an optimized inspection module: used to perform inspections according to the planned path. For non-risk areas, the robot's shooting parameters are automatically adjusted and retaken based on defect types such as sharpness, contrast, and illumination uniformity. For risk areas, features such as the upstream surface, dense attachment, and complex structures are considered, and specific retake strategies are formulated for different features.

Description

Technical Field

[0001] This invention relates to the field of inspection planning technology, and in particular to a robotic inspection planning system for monitoring marine life on the jacket of an offshore platform. Background Technology

[0002] With the deepening development of offshore oil and gas resources, the problem of marine organism attachment on offshore platform jackets, as key structures supporting offshore facilities, is becoming increasingly prominent during their long-term service. However, traditional marine organism monitoring methods have many limitations: diver operations are restricted by weather, sea conditions, and diving depth, posing safety risks and incurring high costs; while remotely operated underwater vehicle (ROV) inspections typically employ fixed paths and static shooting strategies, making it difficult to dynamically adjust shooting parameters based on actual image quality. This results in unstable image quality in environments such as complex currents, low light, or high attachment density, affecting the accuracy of subsequent identification and analysis. Furthermore, existing technologies lack targeted monitoring strategies for risk areas (such as upstream surfaces, structurally complex areas, and densely attached areas), leading to low inspection efficiency and potential omission of critical information.

[0003] Specifically, traditional methods cannot dynamically adjust the robot's posture, light source, distance, and other imaging parameters based on the specific physical characteristics of the jacket (such as whether it is an upstream surface, adhesion density, and structural complexity). This leads to unstable imaging quality, decreased detection accuracy, low work efficiency, and poor data consistency. Furthermore, when the captured image quality is substandard, there is a lack of effective feedback mechanisms to guide the robot to retake the image, resulting in the omission of key information, insufficient decision support, extended inspection cycles, and reduced inspection efficiency. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a robotic inspection and planning system for monitoring marine life on offshore platform jackets, comprising:

[0005] Inspection area division module: used to divide the monitoring area into non-risk areas and risk areas;

[0006] Basic inspection module: connected to the inspection area division module, used to generate marine life monitoring image acquisition points based on the non-risk area and / or the risk area, and generate an initial inspection path based on the marine life monitoring image acquisition points; the robot performs inspection along the initial inspection path and reaches the marine life monitoring image acquisition point to perform marine life monitoring image capture;

[0007] Marine life monitoring image quality assessment module: Connected to the basic inspection module, it is used to assess the quality of captured marine life monitoring images and generate quality assessment indicators;

[0008] Shooting mode optimization module: Connected to the inspection area division module and the marine life monitoring image quality assessment module, when the quality assessment index of the marine life monitoring image in the risk area fails to meet the standard, multi-dimensional labels are identified for the risk area; a composite response mechanism is generated for the risk area based on the quality assessment index and the identified multi-dimensional labels to determine the robot parameters; for non-risk areas, a standard response mechanism is generated based on the quality assessment index to determine the robot parameters, and a reshoot mode is executed to obtain the main image;

[0009] Optimized Inspection Module: Connected to the shooting mode optimization module, it is used to adjust the initial inspection path based on the main image to obtain the planned path, and then perform the inspection according to the planned path.

[0010] In another implementation, the marine life monitoring image quality assessment module includes the following sub-modules:

[0011] Image sharpness determination submodule: used to determine the sharpness of marine life monitoring images through edge detection algorithms;

[0012] Image contrast determination submodule: used to determine the contrast of marine life monitoring images through grayscale histograms;

[0013] Image illumination uniformity determination submodule: used to determine the illumination uniformity of marine life monitoring images through a local brightness difference analysis algorithm;

[0014] The comprehensive evaluation submodule is used to determine quality evaluation indicators based on the clarity, contrast, and illumination uniformity of marine life monitoring images. It presets a quality evaluation target threshold and compares the quality evaluation indicators with the quality evaluation target threshold. If the quality evaluation indicator is greater than or equal to the quality evaluation target threshold, it means that the captured marine life monitoring image meets the standard. If the quality evaluation indicator is less than the quality evaluation target threshold, it means that the captured marine life monitoring image does not meet the standard.

[0015] In another implementation, the standard response mechanism includes:

[0016] Determine the dominant defect type: Based on the sharpness, contrast, and illumination uniformity of the marine organism monitoring image, determine whether the current marine organism monitoring image is a sharpness-dominant defect type, a contrast-dominant defect type, or an illumination uniformity-dominant defect type.

[0017] For the aforementioned clarity-dominant defect type, the robot's pitch angle is adjusted to 10°-20°, and the vertical distance between the robot and the key components of the acquisition point is adjusted to 0.4m±0.1m. Based on the pitch angle and vertical distance, the acquisition point is re-captured to obtain the first re-captured image.

[0018] For the contrast-dominant defect type, turn on the robot's auxiliary light source and set the brightness to level 3-5. Adjust the illumination angle of the auxiliary light source to 30°-45°. Based on the illumination angle, re-capture the acquisition point to obtain a second re-captured image.

[0019] For defects dominated by uniform illumination, the robot is controlled to re-take pictures at four angles: 0°, 90°, 180° and 270° around the key parts of the acquisition point, and the image at the angle with the greatest uniform illumination is used as the third retake image.

[0020] A quality assessment index is determined for the first retaken image and / or the second retaken image and / or the third retaken image. When the quality assessment index meets the standard or the maximum number of retakes is reached, the first retaken image and / or the second retaken image and / or the third retaken image with the highest quality assessment index is used as the main image and recorded in the inspection log.

[0021] In another implementation, the multidimensional labels of the identified risk areas include frontal surface labels, densely attached labels, and complex structure labels. If the current sampling point is a frontal surface, it is assigned a label of 1; if the current sampling point is a non-frontal surface, it is assigned a label of 0. If the current sampling point is a densely attached area, it is assigned a label of 1; if the current sampling point is a non-densely attached area, it is assigned a label of 0. If the current sampling point is a complex structure area, it is assigned a label of 1; if the current sampling point is a non-complex structure area, it is assigned a label of 0.

[0022] In another implementation, a composite response mechanism is generated for the risk area based on the quality assessment indicators and the identified multidimensional labels to determine robot parameters, including:

[0023] Extract the frontal surface label of the current acquisition point. If the extracted frontal surface label of the current acquisition point is 1, then execute the frontal surface re-capture strategy. If the extracted frontal surface label of the current acquisition point is 0, then extract the dense attachment label of the current acquisition point. If the extracted dense attachment label of the current acquisition point is 1, then execute the dense attachment area re-capture strategy. If the extracted dense attachment label of the current acquisition point is 0, then extract the complex structure label of the current acquisition point. If the extracted complex structure label of the current acquisition point is 1, then execute the complex structure area re-capture strategy. If the extracted complex structure label of the current acquisition point is 0, then execute the standard response mechanism.

[0024] If the extracted frontal surface label, densely attached label, and complex structure label of the current acquisition point are all 1, then the optimal shooting position search mechanism is executed;

[0025] Based on the reshooting strategy for the frontal surface and / or the reshooting strategy for densely attached areas and / or the reshooting strategy for complex structure areas and / or the standard response mechanism, the robot parameters are redefined: robot angle, vertical distance between the robot and key components of the acquisition point, and robot posture.

[0026] The data collection points are re-captured based on the newly determined robot parameters, and the quality assessment indicators are redefined until the determined quality assessment indicators meet the standards. The image with the highest quality assessment indicator is then used as the main image and recorded in the inspection log.

[0027] In another implementation, the frontal re-beating strategy includes:

[0028] Adjust the robot's posture to a parallel orientation to the incoming flow; adjust the robot's pitch angle to 15°-25°; adjust the vertical distance b between the robot and the key components of the data collection point to 0.3m-0.5m; re-capture the current data collection point based on the adjusted robot posture and vertical distance, and re-determine the quality assessment indicators until the re-determined quality assessment indicators meet the standards; use the image with the highest quality assessment indicator as the main image.

[0029] In another implementation, the densely attached region re-photographing strategy includes:

[0030] After adjusting the robot's pitch angle to 30°, the robot takes the key component at the current collection point as the origin and patrols around the perimeter with a radius of 0.6m. It takes one marine life monitoring image every 10° of rotation. The quality assessment index is determined for the marine life monitoring images taken during the patrol, and the marine life monitoring image with the highest quality assessment index among the marine life monitoring images taken during the patrol is taken as the main image.

[0031] In another implementation, the optimal shooting location search mechanism includes:

[0032] The current collection point where the frontal surface label, densely attached label, and complex structure label are all 1 is used as the initial collection point;

[0033] Based on the initial acquisition point, several candidate acquisition points and a set of attitude parameters are determined. Based on the set of attitude parameters, the sharpness, contrast, and illumination uniformity of the candidate acquisition points are determined. Combined with the initial attitude parameter set of the initial acquisition point, the optimal shooting position function is determined to obtain the current optimal shooting position.

[0034] The robot is controlled to adjust to the optimal shooting position to capture marine life monitoring images and determine the quality assessment indicators. If the indicators are met, the currently captured marine life monitoring image is used as the main image. If the indicators are not met, a refined candidate posture set is determined. Based on the current refined candidate posture set, marine life monitoring images are captured and the quality assessment indicators are determined until the quality assessment indicators are met or manual intervention is required. The captured marine life monitoring image is then used as the main image.

[0035] In another implementation, the initial inspection path is adjusted according to the main map to obtain the planned path, including: extracting several inspection logs to determine the retake rate and the number of consecutive one-time compliance times for each collection point; taking multiple consecutive collection points greater than or equal to the target retake threshold as continuous segments, and removing collection points greater than or equal to the target number of consecutive one-time compliance times; adding new collection points between adjacent collection points in the continuous segments, and using the trajectory formed by connecting the collection points and the new collection points as the planned path.

[0036] In another implementation, the method further includes generating abnormal events: identifying marine organism species and attachment thickness based on the main maps of risk and non-risk areas; determining the average growth rate of each marine organism species; setting a maintenance threshold; determining the time required to reach the maintenance threshold based on the attachment thickness and average growth rate of each marine organism species; and determining an abnormal event based on the duration.

[0037] The embodiments of the present invention have the following technical effects:

[0038] This invention, building upon the existing risk and non-risk area classifications, further implements a refined re-capture strategy based on quality assessment indicators. Specifically, it introduces multi-dimensional tags (current-facing surface tags, densely attached tags, and complex structure tags) in high-risk areas to achieve more precise image optimization processing. For non-risk areas, when marine life monitoring images fail to meet preset quality standards, the robot's shooting parameters are automatically adjusted and re-captured based on defect types such as sharpness, contrast, and illumination uniformity until a satisfactory main image is obtained. For risk areas, this invention specifically considers features such as current-facing surfaces, densely attached organisms, and complex structures, and formulates dedicated re-capture strategies for each feature.

[0039] When facing an oncoming flow, the robot adjusts its posture to be parallel to the water flow direction and precisely controls the pitch angle and vertical distance to reduce shaking caused by the water flow, ensuring image stability and clarity. For densely attached areas, the robot uses a surround shooting method to capture images from multiple angles and selects the image with the best lighting uniformity as the main image. For complex structural areas, the robot adjusts its posture to the optimal shooting position to overcome visual obstacles caused by the structure and improve detail recognition capabilities. In addition, when a collection point simultaneously possesses the characteristics of an oncoming flow, dense attachment, and complex structure, the system will activate an optimal shooting position search mechanism to comprehensively consider various factors and find the best shooting solution.

[0040] This invention not only significantly improves image quality in high-risk areas but also greatly enhances the accuracy and reliability of subsequent data analysis. By dynamically adjusting robot parameters to adapt to different environmental conditions and target requirements, it achieves effective capture of key information, avoiding information omissions or misjudgments caused by image quality issues.

[0041] This invention analyzes the retake rate and the number of consecutive first-time compliances for each acquisition point in the inspection log to identify continuous segments with unstable imaging, and eliminates redundant points with stable imaging, thereby optimizing the initial inspection path. In these high-risk continuous segments, the number of new acquisition points is calculated based on the distance between adjacent acquisition points and the preset target distance, and their positions are determined through linear interpolation to ensure a smooth path transition. Finally, the original acquisition points and the new points are connected in spatial order to form the optimized planned path. This method not only improves the stability and information integrity of image quality in high-risk areas but also optimizes resource allocation, achieving efficient and accurate monitoring of marine life on the jacket of marine platforms, shortening the inspection cycle and improving inspection efficiency.

[0042] This invention identifies marine organism species and their attachment thickness within risk and non-risk areas by using master images determined through re-enhancing imaging. It then pre-sets maintenance thresholds based on the average growth rate of each species and calculates the time required to reach these thresholds, thereby identifying abnormal events. Multidimensional tags are used for targeted re-enhancing of risk areas to ensure high-quality master images. Subsequently, inspection logs are generated based on these master images, analyzing the re-enhancing rate and the number of consecutive first-time compliances at each collection point to optimize the initial inspection path and form a more reasonable planned path. Regular monitoring based on this optimized path can accurately predict when maintenance is needed and take corresponding measures according to different levels of warnings. This method not only improves the predictability and scientific rigor of maintenance plans but also ensures the safe operation of offshore platform jackets, reduces downtime and costs caused by unexpected maintenance, and improves overall management efficiency and long-term facility stability. Ultimately, the identification of abnormal events is based on high-quality master images and optimized planned paths, realizing an efficient and accurate marine organism monitoring and maintenance strategy and improving decision-making capabilities. Attached Figure Description

[0043] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0044] Figure 1 This is a structural block diagram of a robotic inspection planning system for monitoring marine organisms on a pipe rack of an offshore platform, provided in an embodiment of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0046] like Figure 1 As shown, this embodiment discloses a robotic inspection planning system for monitoring marine life on a jacket structure of an offshore platform, including the following modules:

[0047] Inspection area division module: used to divide the monitoring area into non-risk areas and risk areas.

[0048] Historical attachment data accumulated over a long period was collected from multiple jacket nodes, including marine species diversity, average attachment density, and attachment layer thickness at each node location. Simultaneously, environmental parameters corresponding to each node location, including water depth, current velocity, salinity, and temperature, were collected. Based on this, and combining the jacket design drawings and a 3D point cloud model, key structural feature regions were extracted, including major load-bearing nodes, support structure intersections, and upstream and downstream faces, serving as the basic spatial units for risk analysis. Subsequently, a multi-factor risk assessment model based on fuzzy logic was established: continuous variables such as attachment density, attachment thickness, current velocity, and salinity were transformed into "low, medium, and high" fuzzy linguistic variables through membership functions; a fuzzy rule base was set, such as "if the attachment thickness is high and the current velocity is high, the risk level is high," and "if it is an upstream face and the salinity is high, the risk level is medium-high," comprehensively considering the coupled effects of biological attachment on corrosion, fatigue, and hydrodynamic performance; a comprehensive risk score for each structural unit was calculated using a fuzzy inference system, and a risk index between 0 and 1 was defuzzified and output. Finally, based on preset thresholds, the duct stent is divided into different risk levels: areas with a risk index greater than or equal to 0.7 are classified as risk areas, 0.4 to 0.7 as medium-risk areas (non-risk areas), and less than 0.4 as low-risk areas (non-risk areas). All classification results are displayed on the duct stent 3D model through color-coded overlays. The above-mentioned logic for classifying non-risk and risk areas is a conventional technical method and will not be described in detail in this embodiment.

[0049] Basic inspection module: connected to the inspection area division module, used to generate marine life monitoring image acquisition points based on the non-risk area and / or the risk area, and generate an initial inspection path based on the marine life monitoring image acquisition points; the robot performs inspection along the initial inspection path and arrives at the marine life monitoring image acquisition point to perform marine life monitoring image capture.

[0050] Key structural nodes within the risk area (such as load-bearing connection points, stress concentration zones, and historically corroded areas) are prioritized as image acquisition points and used as priority access points for path planning. The starting point of the path is set at the underwater access point or mooring position closest to the first risk area. All acquisition points are spatially divided into several task clusters, ensuring that the total travel distance and operation time of a single task cluster do not exceed 80% of the robot's endurance, and that each acquisition point is within a preset communication radius to maintain real-time monitoring and emergency control capabilities. For each task cluster, Dijkstra's algorithm is first used to calculate the shortest topological path between acquisition points, generating a complete connected graph. Then, the A* algorithm is introduced, using Euclidean distance as a heuristic function, to accelerate path search while ensuring global optimality, generating an initial inspection path that starts from the starting point, traverses all acquisition points, and returns to the endpoint (or the next task continuation point).

[0051] Furthermore, by incorporating the ocean current vector field as part of the path cost function, a higher weight is given to the countercurrent path, guiding the robot to prioritize traveling with or across the current to reduce energy consumption, improve stability, and minimize drift errors. The path planning engine receives ocean current forecast data in real time and makes local corrections to the initial path to ensure navigation efficiency and safety. Finally, multiple marine life monitoring image acquisition points are clearly set on the generated initial inspection path, each labeled with its three-dimensional coordinates (X, Y, Z) and a recommended shooting angle. The robot will autonomously navigate along this initial path, sequentially reaching each acquisition point to perform image acquisition tasks. The design of the above initial inspection path is also a conventional technical method, and will not be described in detail in this embodiment.

[0052] Marine life monitoring image quality assessment module: Connected to the basic inspection module, it is used to assess the quality of captured marine life monitoring images and generate quality assessment indicators.

[0053] The shooting mode optimization module is connected to the inspection area division module and the marine life monitoring image quality assessment module. When the quality assessment index of the marine life monitoring image in the risk area fails to meet the standard, it identifies multi-dimensional labels for the risk area; it generates a composite response mechanism to determine the robot parameters for the risk area based on the quality assessment index and the identified multi-dimensional labels; for non-risk areas, it generates a standard response mechanism to determine the robot parameters based on the quality assessment index and executes a reshoot mode to obtain the main image.

[0054] Optimized Inspection Module: Connected to the shooting mode optimization module, it is used to adjust the initial inspection path based on the main image to obtain the planned path, and then perform the inspection according to the planned path.

[0055] Next, each of the above modules will be explained in detail:

[0056] The marine life monitoring image quality assessment module includes the following sub-modules:

[0057] Image sharpness determination submodule: Used to determine the sharpness of marine life monitoring images through edge detection algorithms.

[0058] An edge detection algorithm is used to process the acquired images, and the amount of edge information in the image is calculated as a standard for measuring sharpness.

[0059] Image contrast determination submodule: used to determine the contrast of marine life monitoring images through grayscale histograms.

[0060] Image illumination uniformity determination submodule: used to determine the illumination uniformity of marine life monitoring images through a local brightness difference analysis algorithm.

[0061] This algorithm divides the image into several small blocks, calculates the average brightness within each block, and compares the brightness differences between adjacent areas to determine whether the lighting in the entire image is uniform. Good lighting uniformity ensures that all parts of the image receive sufficient and balanced illumination, avoiding information loss or distortion caused by uneven lighting.

[0062] The comprehensive evaluation submodule is used to determine quality evaluation indicators based on the clarity, contrast, and illumination uniformity of marine life monitoring images. It presets a quality evaluation target threshold and compares the quality evaluation indicators with the quality evaluation target threshold. If the quality evaluation indicator is greater than or equal to the quality evaluation target threshold, it means that the captured marine life monitoring image meets the standard. If the quality evaluation indicator is less than the quality evaluation target threshold, it means that the captured marine life monitoring image does not meet the standard.

[0063] The quality assessment index is obtained by weighted summation of the sharpness, contrast, and illumination uniformity of marine life monitoring images. All available marine life monitoring images are collected from historical inspection records. The quality assessment index for these images is calculated, along with its mean and standard deviation. The sum of the mean and standard deviation of the quality assessment index is used as the target threshold for quality assessment. If the quality assessment index is greater than or equal to the target threshold, it indicates that the captured marine life monitoring image meets the predetermined quality standard. Conversely, if the quality assessment index is less than the target threshold, it means that the image has not reached the expected quality level and may have problems such as blurriness, insufficient contrast, or uneven illumination, requiring re-capture or appropriate image enhancement measures. Therefore, the correct understanding is that when the quality assessment index reaches or exceeds the threshold, the image is considered acceptable; otherwise, it is unacceptable. The purpose of this is to ensure that the acquired marine life monitoring images possess sufficient quality and reliability.

[0064] The standard response mechanism includes:

[0065] Determine the dominant defect type: Based on the sharpness, contrast, and illumination uniformity of the marine organism monitoring image, determine whether the current marine organism monitoring image is a sharpness-dominated defect type, a contrast-dominated defect type, or an illumination uniformity-dominated defect type.

[0066] First, determine the co-occurrence deviations of sharpness, contrast, and illumination uniformity:

[0067] △S=max(0,Sth-S), △C=max(0,Cth-C), △U=max(0,Uth-U).

[0068] Where △S represents the sharpness contribution deviation, Sth is the sharpness target threshold, △C represents the contrast contribution deviation, Cth is the contrast target threshold, △U represents the illumination uniformity contribution deviation, Uth is the illumination uniformity target threshold, S represents sharpness, C represents contrast, U represents illumination uniformity, and max represents the maximum value operation. The sharpness target threshold, contrast target threshold, and illumination uniformity target threshold are also obtained based on the mean and standard deviation of the corresponding indicators in the historical inspection records.

[0069] If △S>△C and △S>△U, the current marine life monitoring image is determined to be a sharpness-dominated defect; if △C>△S and △C>△U, the current marine life monitoring image is determined to be a contrast-dominated defect; if △U>△S and △U>△C, the current marine life monitoring image is determined to be an illumination uniformity-dominated defect.

[0070] For the aforementioned clarity-dominant defect type, the robot's pitch angle 'a' is adjusted to 'a' = 10°-20°, and the vertical distance 'b' between the robot and the key components of the acquisition point is adjusted to 'b' = 0.4m ± 0.1m. Based on the pitch angle and vertical distance, the acquisition point is re-captured to obtain the first re-captured image.

[0071] For the contrast-dominant defect type, turn on the robot's auxiliary light source and set the brightness to level 3-5. Adjust the illumination angle c of the auxiliary light source to c=30°-45°. Based on the illumination angle, retake the acquisition point to obtain a second retake image.

[0072] To address the clarity-driven deficiency, the robot's pitch angle was adjusted to 10°-20°, and the vertical distance between the robot and key components at the data acquisition point was set at 0.4m ± 0.1m. This optimization strategy is based on optical imaging principles and underwater vision characteristics. The theoretical basis for this parameter setting is that in the underwater environment, due to significant light scattering and absorption effects, both excessively close and excessively far shooting distances will lead to image blurring. When the robot is too close to the target, insufficient lens focusing range or limited field of view may prevent the acquisition of a complete and clear image; while excessive distance will result in increased scattering by suspended particles in the water, leading to decreased image contrast and loss of detail. A vertical distance of approximately 0.4m falls within the optimal working focal length range of most underwater cameras, effectively balancing field of view coverage and image clarity while reducing interference from the water medium on the optical path. Furthermore, controlling the pitch angle between 10°-20° allows the camera to observe the target surface at a slightly tilted angle. This non-orthogonal perspective helps enhance the three-dimensionality of surface texture and the visibility of edge contours, making it particularly suitable for detecting marine organism attachments on metal structures such as pipe supports. This angle avoids glare caused by specular reflection and magnifies the visual differences of tiny protrusions or corrosion pits by observing them at an angle, thereby improving the spatial resolution and detail of the image. Practice has shown that using this parameter combination for re-shooting significantly improves edge sharpness and the discernibility of structural textures, effectively mitigating blurring caused by inaccurate focusing, water current disturbance, or low light, thus improving image clarity scores and providing high-quality visual data support for subsequent marine organism species identification and attachment analysis. This angle and distance parameter not only has a solid theoretical basis but has also been verified through extensive underwater simulation experiments and actual platform testing, ensuring that the acquired images consistently meet the quality assessment criteria.

[0073] For defects dominated by illumination uniformity, the robot is controlled to re-take pictures at four angles: 0°, 90°, 180°, and 270° around the key parts of the acquisition point. The image at the angle with the greatest illumination uniformity is used as the third retake image.

[0074] In underwater environments, due to severe attenuation of natural light, illumination typically relies on auxiliary light sources provided by the robot. However, single-sided light sources are prone to producing shadows, highlights, or uneven transitions between light and dark areas at structural edges, grooves, or complex geometric surfaces, severely impacting image visual quality and the accuracy of subsequent analysis. By taking panoramic shots from four 90° intervals, the relative position of the light source and the target surface can be changed from different angles, thereby altering the incident direction and reflection path of the light. This effectively avoids shadowed areas caused by local occlusion or overexposed areas caused by specular reflection. For example, the weld seam of a guide frame might be shadowed from one angle, but from another angle after rotating 90°, this area might be directly illuminated, revealing details. By acquiring images from multiple angles and evaluating the illumination uniformity of each image using algorithms such as local brightness variance, illumination gradient consistency, or regional contrast, the optimal image can be selected, significantly improving the visibility and information integrity of critical areas. Practice has shown that this multi-directional re-image strategy can not only effectively improve the overall illumination distribution of the image, but also enhance the visibility of complex surface textures and avoid information loss caused by a single illumination angle.

[0075] A quality assessment index is determined for the first retaken image and / or the second retaken image and / or the third retaken image. When the quality assessment index meets the standard or the maximum number of retakes is reached, the first retaken image and / or the second retaken image and / or the third retaken image with the highest quality assessment index is used as the main image and recorded in the inspection log.

[0076] The log content covers multiple dimensions: recording the specific date and time of the inspection; detailing the area covered by the inspection, clearly distinguishing between high-risk and non-risk areas, and listing the 3D coordinates of all image acquisition points within each area and their logical positions within the jacket structure (e.g., east side of J1 node, middle of L3 support leg). In the image acquisition details section, the initial image number and corresponding shooting parameters for each acquisition point are recorded, along with the specific parameter settings and shooting time for retaken images. Additionally, the quality assessment indicators for each retaken image and whether they meet the standards are included. Furthermore, the final selected main image number for each acquisition point is clearly marked, along with the selection criteria. The log also tracks the number of retakes for each acquisition point and records whether retaking was terminated due to reaching the preset maximum number of retakes.

[0077] The multidimensional labels for the identified risk areas include frontal surface labels, densely attached surface labels, and complex structure labels. If the current sampling point is a frontal surface, it is assigned a label of 1; if the current sampling point is a non-frontal surface, it is assigned a label of 0. If the current sampling point is a densely attached area, it is assigned a label of 1; if the current sampling point is a non-densely attached area, it is assigned a label of 0. If the current sampling point is a complex structure area, it is assigned a label of 1; if the current sampling point is a non-complex structure area, it is assigned a label of 0.

[0078] The upstream facing label identifies areas directly facing the main current. These areas, subjected to strong currents over time, are prone to accumulating more marine organisms and are more susceptible to corrosion and abrasion; therefore, a label value of 1 is assigned to indicate high risk. Off-stream facing areas are relatively stable and have a label value of 0. Secondly, the densely attached label targets areas where large amounts of marine organisms have been observed. These areas indicate higher maintenance needs and potential structural health threats, and are also labeled with a value of 1 to highlight their importance. Conversely, areas with less marine organism attachment are marked with a value of 0, indicating lower current risk. Finally, the complex structure label marks areas with complex geometries or special structures. These locations are often difficult to clean and prone to becoming corrosion hotspots, posing additional challenges to inspection work; therefore, such areas are marked with a value of 1. Simpler structures are marked with a value of 0.

[0079] Specifically, the process of generating a composite response mechanism for the risk area based on the quality assessment indicators and identified multidimensional labels to determine robot parameters includes:

[0080] The system extracts the upstream surface label of the current sampling point using a 3D water flow model. If the extracted upstream surface label is 1, the upstream surface re-shooting strategy is executed. If the extracted upstream surface label is 0, the system extracts the dense attachment label of the current sampling point. If the extracted dense attachment label is 1, the dense attachment region re-shooting strategy is executed. If the extracted dense attachment label is 0, the system extracts the complex structure label of the current sampling point. If the extracted complex structure label is 1, the complex structure region re-shooting strategy is executed. If the extracted complex structure label is 0, the standard response mechanism is executed.

[0081] Collecting water flow data and water volume direction data at different depths requires the use of buoy and satellite remote sensing technologies to collect data on tidal cycles, wave characteristics, and seabed topography. This data is then used as input parameters for specialized hydrodynamic software to construct and run a three-dimensional water flow model, simulating the distribution of the water flow field. Based on the simulation results, the average flow velocity and wall shear stress are calculated. If the wall shear stress in a certain area is greater than 0.5 Pa and the average flow velocity is greater than 0.8 m / s, the area is considered to be under long-term strong water flow scouring and belongs to the upstream region, assigned a upstream label value of 1; conversely, if the shear stress is below the threshold or the flow velocity is low, it is marked as 0. Finally, these labels are integrated into the three-dimensional model of the jacket structure to form a detailed upstream risk map. The method for extracting the upstream labels is a conventional technique, which will not be described in detail in this embodiment.

[0082] To extract density-based attachment labels, high-resolution images of the jacket structure surface accumulated from historical inspections were collected and used to train a lightweight convolutional neural network. The model training objectives included marine species identification and coverage area segmentation, outputting the pixel percentage of each species in each image, which was then quantified as an attachment density index (pixel percentage less than 20% defined as low attachment density, pixel percentage between 20% and 50% as medium attachment density, and pixel percentage greater than 50% as high attachment density). Areas with a density exceeding 50% in two consecutive inspections were assigned a dense attachment label of 1, otherwise 0. This label was pre-written into the jacket structure's 3D model as a static risk attribute for use in subsequent inspection tasks.

[0083] Based on the original design drawings, key areas with typical complex geometric configurations were identified, including: multi-branch connection nodes formed by the intersection of multiple pipe fittings (T-joints, Y-branch, K-support intersection areas), narrow annular gaps with limited space (annular area between the sleeve and the main pile, groove at the flange connection), densely stiffened rib areas, acute-angled spaces formed by the intersection of diagonal braces and main components, and closed cavities or U-shaped structures on the backflow side that have a shielding effect. These key areas were assigned a complex structure label of 1, and otherwise 0.

[0084] If the extracted frontal surface label, densely attached label, and complex structure label of the current acquisition point are all 1, then the optimal shooting position search mechanism is executed.

[0085] Based on the reshooting strategy for the frontal surface and / or the reshooting strategy for densely attached areas and / or the reshooting strategy for complex structure areas and / or the standard response mechanism, the robot parameters are redefined: robot angle, vertical distance between the robot and key components of the acquisition point, and robot posture.

[0086] The upstream surface tags, densely attached tags, and complex structure tags used in this invention are essentially based on three dimensions—three completely independent physical mechanisms affecting imaging quality—to characterize different types of risks in the underwater acquisition environment, demonstrating clear technical division and complementary nature. Collaborative optimization is achieved through strategy combination or upgrade mechanisms, fully reflecting the orthogonality and superposition of the tags.

[0087] First, from a technical perspective, the three tags correspond to different physical challenges and image degradation mechanisms:

[0088] The frontal surface label is used to identify areas that have been subjected to strong water flow for a long time. The core problem is that water flow disturbance causes robot posture instability and image blurring. Therefore, it is necessary to adjust the posture angle and distance to enhance system stability.

[0089] Densely attached tags are used to identify areas with high marine life coverage and uneven surface reflection. The main challenge is that the light reflection is complex, the surface texture is severely interfering, and the image contrast is reduced. It is necessary to take pictures from multiple angles to select the image with the best light uniformity.

[0090] Complex structure tags are used to mark areas with complex geometry, limited space, or visual occlusion. The technical challenge is that robots have difficulty facing the target surface directly or obtaining a complete field of view, and need to search for the optimal shooting position to avoid occlusion.

[0091] The technical problems addressed by these three aspects are non-overlapping and non-inclusive, and they act on three different levels: robot dynamics, optical imaging, and spatial geometry, respectively, thus forming a multi-dimensional and orthogonal description of the risks of underwater collection environment.

[0092] For cases where multiple labels are simultaneously 1 (frontal surface label and / or dense attachment label and / or complex structure label), the corresponding reshooting strategy can be executed simultaneously, or the optimal shooting position search mechanism can be selected based on risk complexity. When the frontal surface, dense attachment, and complex structure labels are not simultaneously 1, i.e., when there are combinations such as 110, 101, 011, 100, 010, 001, etc., where some are 1 and some are 0, the operating mechanisms of the three strategies apply to different control dimensions, are independent of each other, and can be naturally integrated:

[0093] The frontal reshooting strategy primarily constrains the robot's posture, requiring it to be adjusted to be parallel to the frontal surface, controlling the pitch angle between 15° and 25°, maintaining a vertical distance of 0.3m to 0.5m, and emphasizing the stability of the shooting geometry.

[0094] The dense attachment area reshooting strategy, while meeting basic shooting conditions, introduces a surround shooting action and selects the image with the best illumination uniformity from multiple frames, focusing on optimizing illumination quality.

[0095] The reshooting strategy for complex structural areas actively avoids structural occlusion by searching for the optimal viewpoint or path, ensuring the visibility of key areas and emphasizing the integrity of the field of view.

[0096] Because these strategies operate on different levels such as attitude control, shooting sequence planning, and viewpoint optimization, their execution logic does not conflict fundamentally, and they naturally have the ability to be superimposed. For example:

[0097] When the tag combination is 110 (fronting surface = 1, dense attachment = 1, complex structure = 0), under the premise of meeting the attitude and distance requirements of the fronting surface, a surround shooting is performed, and the image with the best lighting is selected to achieve dual guarantee of stability and imaging quality.

[0098] At 101 (fronting surface = 1, dense attachment = 0, complex structure = 1), while maintaining the orientation of the fronting surface, the optimal viewing angle search is initiated to avoid obstruction and balance shooting stability and structural integrity.

[0099] In case 011 (fronting surface = 0, dense attachment = 1, complex structure = 1), although there are no specific attitude requirements, it is still possible to perform surround shooting and search for the best viewing angle to ensure that high-quality images with uniform lighting and no obstruction are obtained in complex structures. Other cases other than 1 will not be described in detail in this embodiment.

[0100] The data collection points are re-captured based on the newly determined robot parameters, and the quality assessment indicators are redefined until the determined quality assessment indicators meet the standards. The image with the highest quality assessment indicator is then used as the main image and recorded in the inspection log.

[0101] The aforementioned frontal re-shooting strategy includes:

[0102] Adjust the robot's posture to a parallel orientation to the incoming flow; adjust the robot's pitch angle d to 15°-25°; adjust the vertical distance b between the robot and the key components of the data collection point to 0.3m-0.5m; re-capture the current data collection point based on the adjusted robot posture and vertical distance; and re-determine the quality assessment indicators until the re-determined quality assessment indicators meet the standards. Use the image with the highest quality assessment indicator as the main image.

[0103] Reshooting strategies for densely attached areas include:

[0104] After adjusting the robot's pitch angle to 30°, take the key component at the current collection point as the origin and inspect it once with a radius of 0.6m. Take one image every 10° of rotation. Determine the quality assessment index for the images taken once during the inspection and take the image with the highest quality assessment index as the main image.

[0105] The so-called parallel orientation to the flow surface refers to a robot's main axis (direction of movement) being basically aligned with the direction of water flow when performing a shooting task. Simultaneously, the camera faces the surface of the flow surface, and the robot's orientation plane is parallel to the normal direction of the flow surface of the jacket structure. This orientation means that the robot is facing the flow surface structure head-on, with its front facing the flow and its rear flowing with it, placing it in a state of minimum resistance streamline. This effectively reduces attitude swaying and positional drift caused by the water flow, improving shooting stability.

[0106] For example, if one facade of the jacket structure faces the direction of the water flow (i.e., the east side is the incoming flow direction and the west side is the outgoing flow direction), this east-west facing facade is the upstream face. When performing a filming task, the underwater robot should adjust its posture to align its forward direction with the water flow direction (from east to west), while maintaining a vertical body and ensuring that the robot's front (i.e., the side where the camera is located) faces the upstream face and remains parallel to it, like two walls facing each other side-by-side. At this time, the robot moves slowly forward with the water flow, filming as it moves. This avoids posture swaying caused by lateral water flow impacts and ensures that the camera is directly facing the target surface, obtaining stable, clear, and distortion-free images. This "flowing with the current, facing the wall" posture is the upstream face parallel posture, the core of which lies in the coordinated control of consistent direction and parallel surfaces.

[0107] Compared to the arbitrary angles or misaligned postures that might be used during initial inspections (such as approaching at an angle or shooting from the back of the current), this embodiment adjusts the robot to a parallel posture facing the current, and further adjusts the pitch angle to 15°–25°, with the vertical distance controlled within the range of 0.3m–0.5m. These parameter settings have clear engineering and imaging basis: a pitch angle of 15°–25° can enhance the three-dimensionality of metal surface textures and marine life protrusions while avoiding specular reflection, thus improving edge sharpness; while the distance range of 0.3m–0.5m is within the optimal working focal length of the underwater camera, ensuring sufficient field of view coverage while avoiding limited field of view due to being too close or blurred details due to being too far. More importantly, compared to the misaligned postures often seen during the original shooting, the parallel posture facing the current significantly reduces the robot's shaking amplitude in strong current environments. Actual measurement data shows that attitude angle fluctuations can be reduced by approximately 40%, thereby greatly improving image stability and clarity.

[0108] The optimal shooting location search mechanism includes:

[0109] The current collection point where the frontal surface label, densely attached label, and complex structure label are all 1 is used as the initial collection point;

[0110] Based on the initial acquisition point, several candidate acquisition points and a set of attitude parameters are determined. The sharpness, contrast, and illumination uniformity of the candidate acquisition points are determined based on the set of attitude parameters. Combined with the initial attitude parameter set of the initial acquisition point, an optimal shooting position function is determined to obtain the current optimal shooting position. Here, a candidate acquisition point refers to the shooting pose formed by changing the robot's attitude parameters at the position of the initial acquisition point. The set of attitude parameters is obtained by enumerating the changed robot attitude parameters, and each set of attitude parameters corresponds to a candidate acquisition point. Both the set of attitude parameters and the initial set of attitude parameters include: pitch angle, vertical distance, attitude direction, and light source brightness, represented by vectors.

[0111] The optimal shooting position function is:

[0112] ;

[0113] Q(p) i S(p) represents the comprehensive score of the i-th candidate sampling point. i C(p) represents the image sharpness of the image captured at the i-th candidate acquisition point. i U(p) represents the contrast of the image captured at the i-th candidate acquisition point. i Let |p| represent the illumination uniformity of the image captured at the i-th candidate acquisition point. i -p0|| represents the distance between the i-th candidate acquisition point and the initial attitude parameter set. , , , These represent the weighting coefficients, , , , .

[0114] In the optimal shooting position function, the sharpness, contrast, illumination uniformity, and distance are all calculated after standardization / normalization. Standardization / normalization is a conventional technique and will not be described in detail in this embodiment.

[0115] The candidate acquisition point with the highest comprehensive score is taken as the best shooting position (i.e., the set of attitude parameters corresponding to the highest candidate acquisition point).

[0116] The discretization range of adjustable parameters is preset at the edge: pitch angle (25°–35°, divided into 5 levels in 2°–3° intervals), vertical distance (0.2m–0.4m, divided into 5 levels in 0.05m intervals), attitude direction (whether the frontal surface is parallel or not, 2 states), and light source brightness (low, medium, high, 3 levels). By enumerating all 5×5×2×3=150 parameter combinations at the edge, a complete set of candidate attitudes is formed.

[0117] The robot is controlled to adjust to the optimal shooting position to capture images and determine the quality evaluation index. If the index is met, the captured image is used as the main image. If the index is not met, a refined candidate pose set is determined. Based on the current refined candidate pose set, images are captured and the quality evaluation index is determined until the quality evaluation index is met or manual intervention is required. The captured image is then used as the main image.

[0118] Among them, the refined candidate posture set is formed by taking the robot's locally preset pitch angle (25°–35°, divided into 11 levels in 1° intervals), vertical distance (0.2m–0.4m, divided into 11 levels in 0.02m intervals), attitude direction (parallelism of the frontal surface, 2 states), and light source brightness (low, medium, high, 3 levels) and enumerating all 11×11×2×3=726 parameter combinations locally to form a complete candidate posture set.

[0119] This embodiment optimizes the initial inspection path by performing multiple inspections and generating inspection logs through an inspection area division module, a marine life monitoring image quality assessment module, and a shooting mode optimization module, in order to solve the technical problems of low efficiency and insufficient coverage of key areas by fixed inspection paths.

[0120] The initial inspection path is adjusted based on the main map to obtain the planned path, including: extracting several inspection logs to determine the retake rate and consecutive one-time compliance count for each collection point; taking multiple consecutive collection points with a retake threshold greater than or equal to the target threshold as continuous segments, and removing collection points with a consecutive one-time compliance count greater than or equal to the target count; adding new collection points between adjacent collection points in the continuous segments, and using the trajectory formed by connecting the existing collection points and the new collection points as the planned path.

[0121] Retake rate measures the proportion of images captured at points where image quality is unstable and repeated shooting adjustments are necessary. .

[0122] If the first shot fails to meet the standard during an inspection, attitude optimization is initiated and the shot is retaken, which is recorded as needing a retake; if the first shot meets the standard, it is recorded as meeting the standard in one go.

[0123] Preferably, the target retake threshold is 60%, and the number of consecutive times the target is met in one go is 3.

[0124] Identifying multiple adjacent sampling points with a retake rate exceeding a set threshold as continuous segments elevates the optimization from isolated points to the level of regional systemic risk identification. In actual inspections, some areas, due to complex structures, strong water flow disturbances, or dense biological attachment, often make it difficult to acquire high-quality images from multiple consecutive sampling points in a single attempt. This imaging difficulty has spatial continuity and common environmental causes. If only a single-point perspective is considered, it may be misjudged as an occasional problem. However, by using the condition of multiple consecutive points, the truly high-risk zones can be effectively identified, guiding the insertion of new sampling points within these segments to achieve regional coverage enhancement and resource focus. Secondly, when a sampling point meets the standard on the first shot in multiple inspections, it indicates that its imaging conditions have stabilized, and no additional resources are needed for encrypted sampling. Indiscriminately retaining all high-retake-rate points would lead to an ever-expanding inspection path, increasing task duration and energy consumption, violating the principle of efficient inspection. By eliminating these stable points, path optimization ensures that it focuses on truly unstable risk segments.

[0125] Adding new collection points between adjacent collection points in the continuous segment is a specific process as follows:

[0126] Calculate the Euclidean distance Dab between adjacent sampling points within a continuous segment, and preset the target distance d between adjacent sampling points. target The number of insertions n is determined based on the Euclidean distance and the target distance:

[0127] , This is to round down to the nearest integer.

[0128] Finally, the coordinates of the newly added data collection point are calculated based on the number of insertions and the coordinates of adjacent data collection points:

[0129] P new,j P represents the coordinates of the j-th newly added data collection point. a and P b These represent the coordinates of adjacent data collection points in a continuous segment.

[0130] Connect all the data collection points that were not removed in the original path with the newly added data collection points in spatial order to form a continuous, smooth, and more comprehensive trajectory, which is the optimized planned path.

[0131] Using the target distance as the maximum permissible interval, the entire original path is divided into several equidistant segments. For example, if two points are 1.2 meters apart and the target distance is 0.5 meters, theoretically, this path needs to be divided into 3 segments, meaning that 2 new points need to be inserted in the middle, thus shortening the actual distance between adjacent points to approximately 0.4 meters. This calculation method naturally achieves a uniform distribution of sampling points, ensuring consistent sampling density throughout the entire segment and avoiding discontinuous phenomena such as dense sections and sparse sections, thereby guaranteeing the spatial continuity and consistency of image data.

[0132] Subsequently, based on the coordinates of adjacent points, precise coordinates of newly added data collection points are generated at proportional intervals. This coordinate calculation method follows the principle of linear interpolation, meaning that new points are evenly distributed along the path between adjacent points. For example, the first new point is located at 1 / 3 of the distance from the first adjacent point, and the second is located at 2 / 3. This geometric construction method not only ensures that the new points are perfectly aligned with the original path but also guarantees the smoothness and executability of the robot's motion trajectory. The robot can naturally transition to the next data collection point without making sharp turns or drastic attitude adjustments, conforming to the motion constraints of underwater vehicles.

[0133] This embodiment also includes abnormal event generation: identifying marine organism species and attachment thickness based on the main map of risk areas and non-risk areas; determining the average growth rate of each marine organism species, setting a maintenance threshold, determining the time required to reach the maintenance threshold based on the attachment thickness and average growth rate of each marine organism species, and determining abnormal events based on the time.

[0134] Inspection work is performed based on the planned path, and images are taken at each sampling point along the planned path to determine whether the quality assessment indicators meet the standards and to determine the main image. Anomalies are then identified based on the main image. Specifically, a pre-trained lightweight CNN model deployed at the edge is used to identify the types of marine life and their attachment thickness in the main image. This identification technique is a conventional method and will not be described in detail in this embodiment.

[0135] The marine organisms included are barnacles, mussels, and algae. The average growth rate of barnacles is 0.8 cm / month, mussels 0.5 cm / month, and algae 1.2 cm / month. An attachment thickness threshold of 1 cm was preset for barnacles and mussels, and 3 cm was preset for algae.

[0136] The time required to reach the maintenance threshold is determined based on the adhesion thickness, adhesion thickness threshold, and average growth rate.

[0137] , t p The time required to reach the maintenance threshold, T h Where T is the adhesion thickness threshold, and v is the adhesion thickness. sThis represents the average growth rate.

[0138] t is obtained from the calculation. p Different warning levels can be set to define abnormal events:

[0139] Emergency Warning (tp<1 month): When the maintenance threshold is expected to be reached in less than a month, it indicates that the situation is very urgent and immediate action is required.

[0140] Handling Strategy: Immediately dispatch professional personnel for on-site inspection and cleanup. Activate the emergency response plan to ensure the safe operation of the facilities. Record and analyze the cause of this emergency to optimize subsequent inspection and maintenance plans.

[0141] Warning (1 month ≤ tp < 3 months): When the maintenance threshold is expected to be reached within 1 to 3 months, it indicates that the situation is relatively urgent, but there is still time for planned handling.

[0142] Treatment Strategy: Incorporate the cleanup work into the near-term maintenance plan to ensure completion within the estimated timeframe. Schedule regular monitoring to track the growth of marine life and adjust the treatment plan accordingly. Prepare the necessary human and material resources in advance to ensure the smooth progress of the treatment work.

[0143] Routine monitoring (tp≥3 months): When the maintenance threshold is expected to be reached after three months or longer, it indicates that the situation is relatively stable and can be addressed in the routine maintenance plan.

[0144] Handling Strategy: Incorporate cleaning work into the routine maintenance plan to ensure the long-term stable operation of the facilities. Conduct regular inspections and monitoring to promptly identify and address new issues. Optimize the maintenance plan by combining historical data and experience to improve work efficiency and effectiveness.

[0145] When a data collection point is identified as an abnormal event, the system will automatically generate an abnormal event report containing the following information:

[0146] Exception event ID: A unique identifier.

[0147] Location description: The specific location where the anomaly occurred.

[0148] Main marine organisms: The main types of marine organisms that cause the abnormality.

[0149] Current adhesion thickness: The detected current adhesion thickness.

[0150] Expected time to reach maintenance threshold: The anticipated time calculated based on the growth rate.

[0151] Recommended measures: Maintenance recommendations based on the above information.

[0152] This information will be recorded and used to notify relevant personnel and update maintenance plans.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A robot patrol planning system for offshore platform jacket marine bio-monitoring, characterized in that, include: Inspection area division module: used to divide the monitoring area into non-risk areas and risk areas; Basic inspection module: connected to the inspection area division module, used to generate marine life monitoring image acquisition points based on the non-risk area and / or the risk area, and generate an initial inspection path based on the marine life monitoring image acquisition points; The robot patrols along the initial inspection path and arrives at the marine life monitoring image acquisition point to capture marine life monitoring images; Marine life monitoring image quality assessment module: Connected to the basic inspection module, it is used to assess the quality of captured marine life monitoring images and generate quality assessment indicators; The shooting mode optimization module is connected to the inspection area division module and the marine life monitoring image quality assessment module. When the quality assessment index of the marine life monitoring image in the risk area fails to meet the standard, it identifies multi-dimensional labels for the risk area: upstream surface label, dense attachment label, and complex structure label. If the current acquisition point is upstream, it is assigned a label of 1; if the current acquisition point is not upstream, it is assigned a label of 0. If the current acquisition point is a dense attachment area, it is assigned a label of 1; if the current acquisition point is not a dense attachment area, it is assigned a label of 0. If the current acquisition point is a complex structure area, it is assigned a label of 1; if the current acquisition point is not a complex structure area, it is assigned a label of 0. Based on the quality assessment indicators and the identified multidimensional labels, a composite response mechanism is generated for the risk areas to determine the robot parameters; for non-risk areas, a standard response mechanism is generated based on the quality assessment indicators to determine the robot parameters, and a retake mode is executed to obtain the main image. Based on the quality assessment indicators and the identified multidimensional labels, a composite response mechanism is generated for the risk area to determine robot parameters, including: Extract the frontal surface label of the current acquisition point. If the extracted frontal surface label of the current acquisition point is 1, then execute the frontal surface re-capture strategy. If the extracted frontal surface label of the current acquisition point is 0, then extract the dense attachment label of the current acquisition point. If the extracted dense attachment label of the current acquisition point is 1, then execute the dense attachment area re-capture strategy. If the extracted dense attachment label of the current acquisition point is 0, then extract the complex structure label of the current acquisition point. If the extracted complex structure label of the current acquisition point is 1, then execute the complex structure area re-capture strategy. If the extracted complex structure label of the current acquisition point is 0, then execute the standard response mechanism. If the extracted frontal surface label, densely attached label, and complex structure label of the current acquisition point are all 1, then the optimal shooting position search mechanism is executed, including: The current collection point where the frontal surface label, densely attached label, and complex structure label are all 1 is used as the initial collection point; Based on the initial acquisition point, several candidate acquisition points and a set of attitude parameters are determined. Based on the set of attitude parameters, the sharpness, contrast, and illumination uniformity of the candidate acquisition points are determined. Combined with the initial attitude parameter set of the initial acquisition point, the optimal shooting position function is determined to obtain the current optimal shooting position. The robot is controlled to adjust to the optimal shooting position to capture marine life monitoring images and determine the quality assessment indicators. If the indicators are met, the currently captured marine life monitoring image is used as the main image; if the indicators are not met, a refined candidate posture set is determined, and marine life monitoring images are captured based on the current refined candidate posture set, and the quality assessment indicators are determined until the quality assessment indicators are met or manual intervention is required; the captured marine life monitoring image is used as the main image. Based on the reshooting strategy for the frontal surface and / or the reshooting strategy for densely attached areas and / or the reshooting strategy for complex structure areas and / or the standard response mechanism, the robot parameters are redefined: robot angle, vertical distance between the robot and key components of the acquisition point, and robot posture. The data collection points are re-captured based on the newly determined robot parameters, and the quality assessment indicators are re-determined until the determined quality assessment indicators meet the standards. The image with the highest quality assessment indicator is then used as the main image and recorded in the inspection log. Standard response mechanisms include: Determine the dominant defect type: Based on the sharpness, contrast, and illumination uniformity of the marine organism monitoring image, determine whether the current marine organism monitoring image is a sharpness-dominant defect type, a contrast-dominant defect type, or an illumination uniformity-dominant defect type. For the aforementioned clarity-dominant defect type, the robot's pitch angle is adjusted to 10°-20°, and the vertical distance between the robot and the key components of the acquisition point is adjusted to 0.4m±0.1m. Based on the pitch angle and vertical distance, the acquisition point is re-captured to obtain the first re-captured image. For the contrast-dominant defect type, turn on the robot's auxiliary light source and set the brightness to level 3-5. Adjust the illumination angle of the auxiliary light source to 30°-45°. Based on the illumination angle, re-capture the acquisition point to obtain a second re-captured image. For defects dominated by uniform illumination, the robot is controlled to re-take pictures at four angles: 0°, 90°, 180° and 270° around the key parts of the acquisition point, and the image at the angle with the greatest uniform illumination is used as the third retake image. A quality assessment index is determined for the first retaken image and / or the second retaken image and / or the third retaken image. When the quality assessment index meets the standard or the maximum number of retakes is reached, the first retaken image and / or the second retaken image and / or the third retaken image with the largest quality assessment index is used as the main image and the inspection log is recorded. Optimized Inspection Module: Connected to the shooting mode optimization module, it is used to adjust the initial inspection path based on the main image to obtain the planned path, and then perform the inspection according to the planned path.

2. The robotic inspection and planning system for monitoring marine life on the jacket of an offshore platform according to claim 1, characterized in that, The marine life monitoring image quality assessment module includes the following sub-modules: Image sharpness determination submodule: used to determine the sharpness of marine life monitoring images through edge detection algorithms; Image contrast determination submodule: used to determine the contrast of marine life monitoring images through grayscale histograms; Image illumination uniformity determination submodule: used to determine the illumination uniformity of marine life monitoring images through a local brightness difference analysis algorithm; Comprehensive evaluation submodule: used to determine quality evaluation indicators based on the sharpness, contrast, and illumination uniformity of marine life monitoring images; A preset quality assessment target threshold is set. The quality assessment index is compared with the quality assessment target threshold. If the quality assessment index is greater than or equal to the quality assessment target threshold, it means that the captured marine life monitoring image meets the standard. If the quality assessment index is less than the quality assessment target threshold, it means that the captured marine life monitoring image does not meet the standard.

3. The robotic inspection and planning system for monitoring marine life on the jacket of an offshore platform according to claim 1, characterized in that, The aforementioned frontal re-shooting strategy includes: Adjust the robot's posture to a parallel orientation to the incoming flow; adjust the robot's pitch angle to 15°-25°; adjust the vertical distance b between the robot and the key components of the data collection point to 0.3m-0.5m; re-capture the current data collection point based on the adjusted robot posture and vertical distance, and re-determine the quality assessment indicators until the re-determined quality assessment indicators meet the standards; use the image with the highest quality assessment indicator as the main image.

4. The robotic inspection and planning system for monitoring marine life on the jacket of an offshore platform according to claim 3, characterized in that, Re-shooting strategies for densely attached areas include: After adjusting the robot's pitch angle to 30°, the robot takes the key component at the current collection point as the origin and patrols around the perimeter with a radius of 0.6m. It takes one marine life monitoring image every 10° of rotation. The quality assessment index is determined for the marine life monitoring images taken during the patrol, and the marine life monitoring image with the highest quality assessment index among the marine life monitoring images taken during the patrol is taken as the main image.

5. The robotic inspection and planning system for monitoring marine life on pipe racks of offshore platforms according to claim 1, characterized in that, The initial inspection path is adjusted based on the main map to obtain the planned path, including: extracting several inspection logs to determine the retake rate and consecutive one-time compliance count for each collection point; taking multiple consecutive collection points with a retake threshold greater than or equal to the target threshold as continuous segments, and removing collection points with a consecutive one-time compliance count greater than or equal to the target count; adding new collection points between adjacent collection points in the continuous segments, and using the trajectory formed by connecting the existing collection points and the new collection points as the planned path.

6. The robotic inspection and planning system for monitoring marine life on the jacket of an offshore platform according to claim 1, characterized in that, It also includes abnormal event generation: identifying marine organism species and attachment thickness based on the main map of risk and non-risk areas; determining the average growth rate of each marine organism species, setting a maintenance threshold, determining the time required to reach the maintenance threshold based on the attachment thickness and average growth rate of each marine organism species, and determining abnormal events based on the time.

Images