Hoop diameter adjusting and operating method for detecting apparent defects of circular pile by underwater robot

Abstract

The invention discloses a hoop diameter adjustment and operation method for an underwater robot to detect apparent defects of a circular pile, and the method comprises the steps: obtaining image data and sonar data, and obtaining an initial surface information data set; performing image enhancement and sonar filtering to obtain a clear surface information set; according to the clear surface information set, correcting an imaging error to obtain a corrected feature information base; according to the feature information base, identifying the defect area to obtain a defect distribution mapping graph; according to the defect distribution mapping graph, combining a preset sonar positioning result to obtain defect distribution information; according to the defect distribution information, the preset circular pile curvature and the preset defect density are combined, the adjustment amplitude of the hoop diameter is calculated, and an adjustment interval is obtained; according to the adjustment interval, the diameter configuration is iteratively optimized in combination with the preset geometric parameters and the preset fitness simulation result, and the optimal hoop diameter value is obtained; and according to the optimal hoop diameter value, an underwater installation control instruction is generated, and a final installation parameter scheme is output.

Description

Technical Field

[0001] This invention relates to the field of underwater engineering inspection and structural installation technology, and in particular to a method for adjusting the diameter of the clamp for underwater robot inspection of apparent defects in round piles. Background Technology

[0002] Currently, underwater engineering inspection, as a crucial component of marine infrastructure operation and maintenance, is gradually evolving from traditional manual inspection to intelligent systems. Particularly in visual inspection, the introduction of industrial vision technology offers a new path to improve operational efficiency and inspection accuracy. In the inspection of structures such as underwater piles, achieving high-quality image acquisition and accurate surface defect identification using intelligent visual equipment has become one of the core objectives of current visual inspection systems. However, due to the unique underwater environment, industrial vision systems face challenges such as light refraction, turbidity interference, and complex background noise, leading to easily distorted visual data and directly limiting the inspection effectiveness of AI visual algorithms in underwater scenarios.

[0003] In one existing technology, underwater cameras or sonar sensors are typically installed to collect image information through manual operation or fixed-path movement, which is then combined with shallow image processing algorithms for defect identification. This method, based on traditional image analysis and rule-based judgment, performs reasonably well in controlled environments, but faces serious challenges in natural water bodies. On the one hand, existing systems lack the ability to fuse image and sonar data in multiple modes, making it unable to cope with signal interference in complex underwater backgrounds. On the other hand, while AI vision technology performs excellently in terrestrial scenarios, its adaptability underwater is limited, making it difficult to accurately extract the surface texture features of circular piles. This results in a lack of reliable basis for subsequent calculation of hoist dimensions, hindering the engineering application of visual intelligence in this field.

[0004] In summary, existing technologies suffer from inaccurate identification of defects in circular piles in complex underwater environments and the inability to adaptively adjust the diameter of the clamps. Summary of the Invention

[0005] This invention provides a method and system for adjusting the clamp diameter and operating procedures for detecting apparent defects in round piles using an underwater robot, in order to solve the problems of inaccurate identification of defects in round piles and the inability to adaptively adjust the clamp diameter in complex underwater environments.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a method for adjusting the diameter of a clamp used by an underwater robot to detect apparent defects in a circular pile, comprising: Acquire image and sonar data of the underwater circular pile surface to obtain an initial surface information dataset; Based on the initial surface information dataset, image enhancement and sonar filtering are performed to remove interference noise and obtain a clear surface information set. Based on the set of clear surface information, surface texture and structural features are extracted, imaging errors are corrected, and a corrected feature information library is obtained. Based on the feature information database, a convolutional neural network is used to identify defect regions and obtain a defect distribution mapping map; Based on the defect distribution map, the location and shape characteristics of the defects are analyzed, and combined with the preset sonar positioning results, the defect distribution information of the defect-affected area is obtained. Based on the defect distribution information, combined with the preset circular pile curvature and preset defect density, the adjustment range of the clamp diameter is calculated to obtain the adjustment range; Based on the adjustment range, and combined with the simulation results of preset geometric parameters and preset fit, the diameter configuration is iteratively optimized to obtain the optimal clamp diameter value; Based on the optimal clamp diameter value, underwater installation control commands are generated, parameter stability is verified, and the final installation parameter scheme is output.

[0007] Preferably, the step of performing image enhancement and sonar filtering based on the initial surface information dataset to remove interference noise and obtain a clear surface information set includes: Based on the initial surface information dataset, the image data is denoised using a wavelet transform algorithm to remove high-frequency noise generated by water flow disturbance, resulting in denoised image data. Based on the denoised image data, histogram equalization is used to enhance the image, thereby improving image contrast and texture clarity, resulting in enhanced image data. Based on the initial surface information dataset, the sonar data is time-domain filtered using an adaptive Kalman filter method to suppress background noise interference, resulting in filtered sonar data. Based on the sonar filtering data, the main frequency signal is extracted using the short-time Fourier transform method, and invalid frequency bands are removed to obtain enhanced sonar data; Based on the enhanced image data and enhanced sonar data, alignment is performed according to timestamps, and multi-source data fusion is completed to obtain a clear surface information set.

[0008] Preferably, the step of extracting surface texture and structural features based on the set of clear surface information, correcting imaging errors, and obtaining a corrected feature information library includes: The set of clear surface information is decomposed into multiple scales using the curvature scale analysis method to extract local curvature abrupt change regions and obtain the texture feature dataset. Based on the texture feature dataset, a spatial registration algorithm is used to align the multi-scale texture features, and the structural deformation deviation value is extracted by combining the preset sonar point cloud reconstruction results to obtain the aligned texture and deformation feature dataset. Based on the alignment texture and the deformation feature dataset, an eigenvalue correction algorithm based on the light field refraction model is applied to correct the pixel offset error caused by imaging distortion, and the correction feature data is obtained. Based on the corrected feature data, texture anomalies and deformation deviation values ​​are fused, and time alignment and consistency checks are performed to obtain the corrected feature information library.

[0009] Preferably, the step of identifying defect regions using a convolutional neural network based on the feature information database to obtain a defect distribution map includes: The feature information database is divided into regions, and a set of candidate abnormal regions is extracted to obtain the initial abnormal distribution range; Based on the initial anomaly distribution range, perform texture feature edge detection to obtain the defect region; The defect region is depth-quantized by combining the gray-level co-occurrence matrix analysis method to obtain the numerical distribution data of the defect depth; Based on the numerical distribution data, a preset convolutional neural network model is used to extract and identify features of abnormal regions to obtain defect region determination results. Based on the defect region determination results, the defect boundary contour and depth values ​​are fused to generate a two-dimensional defect location matrix. A visualization mapping method is used to render a heat map of the defect region in the defect region determination results to obtain a defect distribution mapping map.

[0010] Preferably, the step of analyzing the defect location and shape characteristics based on the defect distribution map, and combining this with preset sonar positioning results to obtain defect distribution information in the defect-affected area includes: Based on the defect distribution map, image region segmentation is performed, and the coordinate records of the defect regions are extracted to obtain preliminary defect location distribution results; Based on the defect location distribution results, geometric feature extraction is performed to analyze the defect location contour structure and generate a defect shape feature set. Based on the set of defect shape features, and by comparing it with a preset defect classification model, defect type labeling information is obtained. Based on the defect type labeling information, the preset sonar signal anomaly location results are matched, and the suspected anomaly points are jointly analyzed to obtain the spatial distribution range of the defect anomaly points. Based on the spatial distribution range, defect type and abnormal point information are fused, boundary division and contour extraction are performed to obtain defect distribution information of the defect-affected area.

[0011] Preferably, the step of calculating the adjustment range of the clamp diameter based on the defect distribution information, combined with the preset circular pile curvature and preset defect density, to obtain the adjustment range includes: Based on the defect distribution information, the surface of the circular pile is analyzed using a curvature measurement method to identify areas of abnormal curvature and obtain a circular pile curvature dataset. Based on the defect distribution information, defect distribution density estimation is performed to identify high-density defect regions and obtain a defect density dataset. Based on the circular pile curvature dataset and the defect density dataset, a spatial mapping relationship between the defect location and the clamp installation stress point is established, and the correlation parameters between the location and the stress point are extracted to obtain the parameter mapping result. Based on the parameter mapping results, the adjustment range of the clamp diameter is calculated using regression analysis, and the range is divided into intervals to obtain the adjustment interval.

[0012] Preferably, the diameter configuration is iteratively optimized based on the adjustment range, combined with the preset geometric parameters and the preset fit simulation results, to obtain the optimal clamp diameter value; Based on the adjustment range, the actual geometric parameters of the circular pile are collected and compared with the preset geometric benchmark to identify structural areas with differences and obtain the geometric calibration requirement distribution. Based on the geometric calibration requirement distribution, dynamic calibration processing of geometric parameters is performed. Combined with the preset fit simulation results, an optimized distribution scheme of the force points for clamp installation is constructed to obtain the optimized configuration result of the force points. Based on the optimized configuration of the force points, the fit error is calculated, the error exceeding the limit area is identified, the error adjustment requirement distribution is formed, and the error evaluation result set is obtained. Based on the error evaluation result set, the diameter parameter is adjusted in multiple rounds using an iterative optimization algorithm, and the optimal clamp diameter value is obtained by comparison and verification according to the clamp diameter configuration rules.

[0013] Preferably, the underwater installation control command is generated based on the optimal clamp diameter value, the parameter stability is verified, and the final installation parameter scheme is output.

[0014] Based on the optimal clamp diameter value, an initial installation instruction set for the underwater equipment is generated, and consistency verification is performed against the preset installation parameter standard to obtain the initial control instruction dataset. Based on the initial control command dataset, a simulation analysis of fit error and installation force point distribution is performed, and the results are judged against a preset stability standard to obtain the matching verification results. When the matching verification result does not meet the preset stability standard, the fitting error value and the force point distribution scheme are adjusted and optimized to generate an adjustment control dataset; Based on the adjusted control dataset, combined with the control requirements of the underwater equipment, parameter integration and path mapping are performed to output the final installation parameter scheme.

[0015] Secondly, the present invention provides a clamp diameter adjustment and operating system for underwater robots to detect apparent defects in circular piles, including: The data acquisition module is used to acquire image data and sonar data of the underwater circular pile surface to obtain an initial surface information dataset; The data cleaning module is used to perform image enhancement and sonar filtering based on the initial surface information dataset, remove interference noise, and obtain a clear surface information set. The correction feature module is used to extract surface texture and structural features based on the set of clear surface information, correct imaging errors, and obtain a corrected feature information library. The defect mapping module is used to identify defect regions using a convolutional neural network based on the feature information database, and obtain a defect distribution mapping map. The defect distribution module is used to analyze the location and shape characteristics of defects based on the defect distribution map, and combine the preset sonar positioning results to obtain the defect distribution information of the defect-affected area. The adjustment range module is used to calculate the adjustment range of the clamp diameter based on the defect distribution information, combined with the preset circular pile curvature and preset defect density. The optimal diameter module is used to iteratively optimize the diameter configuration based on the adjustment range, combined with preset geometric parameters and preset fit simulation results, to obtain the optimal clamp diameter value. The installation control module is used to generate underwater installation control commands based on the optimal clamp diameter value, verify parameter stability, and output the final installation parameter scheme.

[0016] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the method for adjusting the clamp diameter and operating the underwater robot for detecting apparent defects in a circular pile as described in any one of the above.

[0017] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the above-described method for adjusting and operating the clamp diameter of an underwater robot to detect apparent defects in a round pile.

[0018] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention effectively solves the problem of observation distortion caused by high turbidity, light refraction and signal interference in the underwater environment by constructing an integrated image and sonar data acquisition system and combining image enhancement, sonar filtering and multi-source fusion processing. The surface information set after the sharpening process not only improves the perception accuracy of the surface texture and structural features of the round pile, but also establishes a unified and accurate data foundation for subsequent defect identification and parameter evaluation, which greatly improves the robustness and reliability of underwater defect detection.

[0019] (2) In the defect identification process, this invention introduces a region discrimination mechanism based on deep convolutional networks and combines it with image segmentation, texture feature quantization and deformation analysis algorithms to achieve accurate classification of defect types and visualization of spatial distribution through heat maps. This process effectively avoids the problems of blurred defect boundaries and inaccurate depth assessment in traditional methods through feature fusion and model inference, improves the system's response sensitivity to potentially high-risk areas, and provides a clear basis for diameter adjustment.

[0020] (3) This invention constructs a parameter mapping and diameter optimization mechanism with multi-source factors as input. Taking into account constraints such as defect distribution density, circular pile geometry, and fit error, a full-link control strategy from adjustment range estimation to path planning output is designed. Through simulation evaluation and multiple rounds of iterative optimization, the system can automatically derive the optimal clamp diameter configuration and output an installation parameter scheme that has passed stability verification, which significantly improves the accuracy, safety, and engineering adaptability of underwater clamp installation and has good practical application value. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the process for adjusting the clamp diameter and operating method of underwater robot for detecting apparent defects in round piles, provided in the first embodiment of the present invention. Figure 2 This is a schematic diagram of the clamp diameter adjustment and operating system structure for underwater robot detection of apparent defects in round piles, provided in the second embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Reference Figure 1 The first embodiment of the present invention provides a method for adjusting the diameter of the clamp for underwater robot detection of apparent defects in round piles, including the following steps: S11, acquire image data and sonar data of the underwater circular pile surface to obtain the initial surface information dataset; S12, based on the initial surface information dataset, perform image enhancement and sonar filtering to remove interference noise and obtain a clear surface information set; S13, Based on the set of clear surface information, extract surface texture and structural features, correct imaging errors, and obtain a corrected feature information library; S14, Based on the feature information database, use a convolutional neural network to identify defect regions and obtain a defect distribution mapping map; S15. Based on the defect distribution map, analyze the defect location and shape characteristics, and combine the preset sonar positioning results to obtain the defect distribution information of the defect-affected area. S16. Based on the defect distribution information, combined with the preset circular pile curvature and preset defect density, calculate the adjustment range of the clamp diameter to obtain the adjustment range; S17. Based on the adjustment range, and combined with the simulation results of preset geometric parameters and preset fit, the diameter configuration is iteratively optimized to obtain the optimal clamp diameter value. S18. Based on the optimal clamp diameter value, generate underwater installation control commands, verify parameter stability, and output the final installation parameter scheme.

[0024] In step S11, it is necessary to acquire image data and sonar data of the underwater circular pile surface to obtain an initial surface information dataset, including: In one specific embodiment, to acquire image and sonar data of the underwater cylindrical pile surface, the system is mounted on an underwater robot platform with hovering capabilities. A multi-source acquisition link is established through the coordinated scheduling of a high-resolution industrial vision camera and a multi-beam imaging sonar device. The image acquisition utilizes a BlueROV2-CAM-HD underwater camera, set to a resolution of 1920×1080, a frame rate of 30fps, and a lens angle of 120°. A gimbal fixed to the robot's front end ensures the camera always faces the outer wall of the cylindrical pile, acquiring a continuous image sequence of surface texture. To avoid image blurring due to insufficient underwater lighting, an LED light source array is used for supplemental lighting during acquisition, with the brightness adjusted to 500 lumens to ensure the target surface is clearly visible.

[0025] In one specific embodiment, the sonar acquisition section uses a Teledyne BlueView M900 series sonar device, set to an operating frequency of 900kHz, a scanning range of 360°, and a maximum imaging radius of 5 meters. The acquisition method is active sector scanning; as the robot moves at a constant speed around the circular pile, a sonar frame is acquired every 1° of rotation, obtaining point cloud cross-sectional data including distance, azimuth, and echo intensity. An inertial navigation and attitude system (IMU) records real-time pose parameters during the acquisition process, ensuring that the sonar data can be spatially registered with the image frames later. For example, in a complete operation cycle, the system acquires 500 image frames, corresponding to 500 sets of sonar frames, lasting approximately 15 minutes, covering the entire sidewall of the circular pile.

[0026] In one specific embodiment, to ensure consistency between image data and sonar data, the system timestamps and synchronizes the two data sources. The camera and sonar acquisition modules share a high-precision real-time clock (RTC) on the main control board, with a fixed sampling interval of 33ms. The data synchronization module aligns image frames and corresponding sonar frames along the time axis and stores them in a structured dataset. Image data is compressed and saved in PNG format, while sonar data is stored in XYZ point cloud format (including amplitude attributes). All data is archived by task number and time label for easy access during subsequent feature extraction and model processing.

[0027] In one specific embodiment, the system performs preprocessing operations on the acquired raw data to improve its usability and accuracy for subsequent analysis. For the image data, a 3×3 kernel median filter is used to remove high-frequency noise caused by floating objects or bubbles, followed by histogram equalization to enhance contrast and preserve subtle surface textures. For the sonar data, background scattering points are first filtered out based on an intensity threshold of -30dB, and then mean clustering is performed to remove anomalous point cloud fragments located outside the projection boundaries of the circular piles. After this processing, both the image and sonar datasets exhibit clear structure and distinct boundaries, making them suitable for subsequent texture analysis and deformation feature modeling.

[0028] In one specific embodiment, the final generated initial surface information dataset consists of three parts: first, a set of image sequences (including original frames and preprocessed frames); second, a set of sonar point cloud frames under the corresponding timestamp; and third, a metadata index table that records auxiliary information such as the acquisition time, pose information, data path, and processing status of each image and sonar frame.

[0029] In step S12, image enhancement and sonar filtering are performed based on the initial surface information dataset to remove interference noise and obtain a clear surface information set, including: The step of performing image enhancement and sonar filtering based on the initial surface information dataset to remove interference noise and obtain a clear surface information set includes: Based on the initial surface information dataset, the image data is denoised using a wavelet transform algorithm to remove high-frequency noise generated by water flow disturbance, resulting in denoised image data. Based on the denoised image data, histogram equalization is used to enhance the image, thereby improving image contrast and texture clarity, resulting in enhanced image data. Based on the initial surface information dataset, the sonar data is time-domain filtered using an adaptive Kalman filter method to suppress background noise interference, resulting in filtered sonar data. Based on the sonar filtering data, the main frequency signal is extracted using the short-time Fourier transform method, and invalid frequency bands are removed to obtain enhanced sonar data; Based on the enhanced image data and enhanced sonar data, alignment is performed according to timestamps, and multi-source data fusion is completed to obtain a clear surface information set.

[0030] In one specific embodiment, to improve the usability and resolution of the initial surface information dataset, wavelet transform denoising is first performed on the image data. The system selects the Daubechies wavelet (db4) as the mother wavelet, performs a three-level decomposition on each image frame, retains the low-frequency components, and uses a threshold soft thresholding function to compress and reconstruct the high-frequency components of the first and second levels, removing high-frequency artifacts and interference edges caused by water flow disturbance. For example, with an image frame size of 1920×1080, the PSNR is improved to 38.7dB after processing, significantly reducing the jagged edges and blurry bands of the image while preserving the original structural texture.

[0031] In one specific embodiment, based on the denoised image described above, histogram equalization is further applied for image enhancement. This step aims to improve the overall uniformity of brightness distribution and the separability of texture details, making it suitable for low-light or dark underwater image environments. Histogram equalization employs the CLAHE (Contrast-Limited Adaptive Histogram Equalization) method, setting the local window size to 64×64 pixels and limiting the contrast enhancement threshold to 4.0 to avoid excessive amplification of local noise. The enhanced image shows a contrast improvement of over 20%, with surface cracks, attachments, and seam boundaries clearly visible, providing a reliable foundation for subsequent texture analysis.

[0032] In one specific embodiment, an adaptive Kalman filter is used to perform time-domain filtering on sonar data to address random background disturbances and high-frequency echo interference. The system constructs a state variable as the moving average of the sonar echo intensity sequence, and the measurement noise covariance matrix is ​​dynamically adjusted based on historical sample statistics, initially set to 0.001, with a process noise covariance of 0.005. This method can effectively suppress abnormal intensity fluctuations caused by multipath echoes or background fish swarms. The processing results show that the fluctuation range of the echo curve is reduced by approximately 35% after filtering, improving the accuracy of subsequent signal spectrum analysis.

[0033] In one specific embodiment, based on the filtered sonar data, the system further employs Short-Time Fourier Transform (STFT) to extract the dominant frequency signal to enhance feature stability. This method segments the sonar signal and performs a Fast Fourier Transform (FFT) using a Hamming window with a window length of 256 points and an overlap rate of 50%. Frequency bands with a signal-to-noise ratio below -10dB are automatically designated as invalid frequency domains, and bandpass filtering is performed, retaining only the effective echoes within the dominant frequency band (800–920kHz). After this processing, the spatial density uniformity of the point cloud features is increased to 1.6 times the original, significantly improving imaging coherence and boundary stability.

[0034] In one specific embodiment, to achieve unified analysis and processing of image and sonar data, the system performs timestamp-based registration and fusion on the enhanced image and sonar data. Each pair of image and sonar frames is accompanied by a GPS timestamp and IMU attitude record during acquisition, and the system performs pairing matching with a tolerance window of ±50ms. After successful pairing, the system packages the image and sonar frames according to a unified data structure and marks them with a fusion identifier. Finally, the system outputs a set of clear surface information containing time-consistent image enhancement frames, filtered sonar point cloud frames, and pairing metadata indexes, forming a structured fusion dataset.

[0035] In step S13, surface texture and structural features need to be extracted based on the set of clear surface information to correct imaging errors and obtain a corrected feature information library, including: The step of extracting surface texture and structural features based on the set of clear surface information, correcting imaging errors, and obtaining a corrected feature information library includes: The set of clear surface information is decomposed into multiple scales using the curvature scale analysis method to extract local curvature abrupt change regions and obtain the texture feature dataset. Based on the texture feature dataset, a spatial registration algorithm is used to align the multi-scale texture features, and the structural deformation deviation value is extracted by combining the preset sonar point cloud reconstruction results to obtain the aligned texture and deformation feature dataset. Based on the alignment texture and the deformation feature dataset, an eigenvalue correction algorithm based on the light field refraction model is applied to correct the pixel offset error caused by imaging distortion, and the correction feature data is obtained. Based on the corrected feature data, texture anomalies and deformation deviation values ​​are fused, and time alignment and consistency checks are performed to obtain the corrected feature information library.

[0036] First, the set of clear surface information is decomposed into multiple scales using Curvature Scale Space (CSS) analysis. This method effectively extracts local geometric abrupt change regions in the image, such as crack edges or surface erosion points, by calculating the curvature changes of the image contour at different scales. For example, the scale parameter σ of the curvature analysis can be set to vary between 0.5 and 2.5 to construct a multi-layer convolutional kernel sequence, thereby obtaining texture change images with different spatial resolutions. After processing, the extracted curvature abrupt change regions constitute a preliminary texture feature dataset, which includes information such as the center position, contour boundary, and intensity of change of each texture unit.

[0037] In one specific embodiment, based on the aforementioned texture feature dataset, a spatial registration algorithm (such as a RANSAC-based rigid transformation matching method) is used to align texture features at multiple scales, ensuring consistency of multi-source information in spatial dimensions. Simultaneously, to enhance the reliability of structural hierarchy information, a preset sonar point cloud reconstruction result is used as a 3D structural reference. For example, this sonar point cloud can be acquired by an underwater multibeam imaging system with a sampling interval of 0.02 meters and a reconstruction accuracy of ±0.01 meters. By establishing a spatial mapping relationship between the point cloud and texture data, structural deformation deviation values, such as local depression depth and protrusion height, are extracted, ultimately resulting in an aligned texture and deformation feature dataset.

[0038] It's important to note that spatial registration algorithms are techniques used to align data from different sources, at different times, or at different scales into a unified spatial coordinate system. The core principle is to identify corresponding feature points, edges, or structural contours in multiple sets of data. By calculating the geometric transformation relationships between these features (such as translation, rotation, scale changes, or affine transformations), an optimal set of transformation parameters is determined to ensure that the target data overlaps with the reference data as accurately as possible in space. In practice, this typically involves two stages: initial coarse registration and subsequent fine registration. The coarse registration stage uses global information about feature points or regions for initial alignment, while the fine registration stage further improves registration accuracy through iterative optimization (such as ICP, mutual information maximization, etc.), thereby achieving structural alignment and minimizing differences between the data.

[0039] In one specific embodiment, to address potential imaging distortions in underwater images (such as light field bending, viewpoint distortion, and edge stretching), an eigenvalue correction algorithm based on a light field refraction model is employed to correct pixel position errors. This model primarily utilizes Snell's Law to construct the imaging ray path, thereby deducing the pixel mapping position of the target in real space. Specifically, the model inputs include the refractive index of the underwater medium, the refractive index of air, camera position parameters, imaging plane size, incident angle distribution, and target depth information. It calculates the angle of refraction and offset distance of light rays passing through the water surface, and combines this with inverse ray tracing to complete image coordinate correction.

[0040] Specifically, the system first sets the basic physical parameters of the light field model, including a water refractive index of 1.33, an air refractive index of 1.0, a water depth of 8m, a camera field of view of 60°, and an imaging sensor resolution of 1920×1080 pixels. The system discretizes the field of view into multiple incident angle units, traverses all incident rays in 0.1° steps, and calculates the propagation direction of each ray after its deflection in the water using Snell's law. Each pixel is mapped to a spatial ray, and a pixel-coordinate transformation relationship from the image plane to the target surface is established by combining the camera's extrinsic parameter matrix (position, orientation).

[0041] Specifically, the algorithm is implemented in Python, using OpenCV and NumPy for optical path reconstruction and feature point mapping. The least squares method is used to regress and optimize the offset error of all pixels, resulting in a corrected set of feature points. For example, for an edge feature point with an offset of 15 pixels in an image, after refraction model correction, its error can be reduced to within 3 pixels, improving the correction accuracy by approximately 80%, effectively enhancing the accuracy of subsequent texture and structural feature extraction.

[0042] In one specific embodiment, based on the corrected feature data, the identified texture anomalies and deformation deviation values ​​are uniformly integrated, and a time alignment operation based on timestamps and spatial indices is performed. Redundant and inconsistent data are then eliminated using a consistency verification algorithm (such as a comparison strategy based on feature similarity thresholds). After this step, the system outputs a complete, accurate, and corrected feature information library that can be used for subsequent analysis. This library structurally includes fields such as texture type, spatial location, deviation magnitude, and time stamp, and supports standard structured storage and interface calls.

[0043] In step S14, based on the feature information database, a convolutional neural network is used to identify defect regions and obtain a defect distribution mapping map, including: The step of identifying defect regions using a convolutional neural network based on the feature information database to obtain a defect distribution map includes: The feature information database is divided into regions, and a set of candidate abnormal regions is extracted to obtain the initial abnormal distribution range; Based on the initial anomaly distribution range, perform texture feature edge detection to obtain the defect region; The defect region is depth-quantized by combining the gray-level co-occurrence matrix analysis method to obtain the numerical distribution data of the defect depth; Based on the numerical distribution data, a preset convolutional neural network model is used to extract and identify features of abnormal regions to obtain defect region determination results. Based on the defect region determination results, the defect boundary contour and depth values ​​are fused to generate a two-dimensional defect location matrix. A visualization mapping method is used to render a heat map of the defect region in the defect region determination results to obtain a defect distribution mapping map.

[0044] In one specific embodiment, to identify regions exhibiting apparent anomalies from the feature information database, the system first employs an image segmentation method to divide the input image into regions. The image segmentation algorithm uses a superpixel-based SLIC (Simple Linear Iterative Clustering) method, dividing the image into structurally consistent small regions, each containing several pixels, thereby extracting an initial set of candidate anomaly regions. For example, the number of superpixels can be set to 500, and the region compactness factor can be set to 10 to ensure good fit to texture boundaries. Based on this result, the system further analyzes the differences between these regions and surrounding regions in terms of grayscale, edge texture, and brightness abrupt changes, preliminarily determining the distribution range of the anomalies.

[0045] Specifically, the system performs edge detection on texture features in the initial anomalous region. First, it employs the Canny edge detection algorithm, smoothing the image with Gaussian filtering to remove noise. Then, it calculates the pixel grayscale changes in the horizontal and vertical directions to determine the location of potential edges in the image. This method uses a pair of high and low thresholds (e.g., a low threshold of 50 and a high threshold of 150) to filter detected edges. Edges below the low threshold are excluded, while those above the high threshold are retained. Intermediate regions are connected to form valid edges based on context, thus accurately depicting the contours of the defective region. The system further utilizes local gradient analysis to enhance boundary contrast, specifically by calculating the intensity of pixel grayscale changes in the horizontal and vertical directions using the Sobel operator. This operation helps distinguish textured anomalous regions from background regions, enhancing edge continuity and clarity, and ensuring accurate localization during subsequent feature extraction.

[0046] After identifying the boundaries of suspected defective regions, the system employs gray-level co-occurrence matrix (GLCM) analysis to quantitatively evaluate the spatial consistency of texture within the region. This method examines the frequency of gray-level combinations of adjacent pixels at different directions and distances to assess the degree of texture variation within the region. The system primarily extracts three types of statistical features: Energy index: Reflects the overall uniformity of image texture; a higher value indicates a smoother image. If the energy is low (e.g., below 0.1), it indicates that there are many local fluctuations within the area, which may be anomalies caused by corrosion or cracks.

[0047] Contrast ratio: Used to measure the intensity of pixel grayscale changes within a region. High contrast ratio (e.g., above 50) usually indicates the presence of obvious edges or defect contours.

[0048] Entropy index: Represents the complexity of the texture in a region. The higher the entropy value, the more chaotic the texture distribution in the region, which is often related to complex defects such as damage and deformation.

[0049] For example, when processing a set of images of underwater cylindrical piles at a depth of 12 meters, if a certain region has an energy value of 0.08, a contrast value of 72, and an entropy value of 5.6, the system determines that this region has a high-confidence defect feature and uses the depth value of this region as the input to the subsequent convolutional neural network recognition model. This processing flow can significantly improve the accuracy and stability of subsequent recognition stages.

[0050] In a specific embodiment, for the numerical distribution data of the depth features of the defect region obtained after processing based on the gray-level co-occurrence matrix, the system introduces a preset convolutional neural network model to further extract features and classify the candidate regions. This model adopts a bottom-up structural design, containing three layers of convolutional operation units and two layers of fully connected structures, used to progressively extract texture patterns and deformation features from the region image.

[0051] Specifically, in the convolutional part, the first layer of the model has 32 convolutional channels, with each channel having a 3×3 kernel size, used for initial extraction of edges and texture directions in image patches; the second layer adds 64 convolutional channels to enhance the ability to recognize local repetitive textures; the third layer expands to 128 channels to extract higher-order deformation features or complex structural information. Each convolutional layer is followed by a ReLU activation function to improve non-linear expressiveness, and a max-pooling layer is inserted for size compression and feature denoising, with a pooling window size of 2×2.

[0052] Specifically, the convolutional feature map is flattened and then input into a fully connected layer. The first fully connected layer is used for feature vector reconstruction and feature selection, with 256 nodes. The second fully connected layer is used as the output layer, and the probability distribution of the two categories, namely "defect region" and "non-defect region", is calculated through the Softmax function.

[0053] Specifically, during the model training phase, the system used nearly 8,000 underwater circular pile image samples collected from actual engineering projects. These samples included various types of defects such as corrosion, cracks, and spalling, and were manually labeled frame by frame to form the training set. During training, the Adam optimizer was used to control parameter updates, with the learning rate fixed at 0.001 to ensure stable convergence. The cross-entropy loss function was chosen to measure classification error, and the number of training epochs was set to 120, with each epoch containing 256 sample batches.

[0054] Specifically, during the inference phase, the system inputs each candidate anomaly region image patch into the convolutional model and outputs the probability value of the corresponding region belonging to the "defect" category. To ensure the reliability of detection and control of false alarms in actual engineering operations, the system sets the defect confidence threshold to 0.7. That is, when the model outputs a probability value greater than or equal to 0.7, the region is judged to have a defect; otherwise, it is considered a non-defect region.

[0055] For example, after the model identifies a region extracted from an image of a circular pile, the output probability is 0.83. Based on this, the system classifies it as a high-confidence defect region, marks it in the final result, and participates in subsequent defect distribution mapping and clamp diameter adjustment calculations. The model has a compact overall structure and fast inference speed, making it suitable for deployment in underwater robot terminal equipment for real-time identification and feedback. Subsequently, the system fuses the boundary contours and depth values ​​from the above defect region determination results to construct a two-dimensional defect location matrix. This matrix contains parameters such as the center coordinates, circumscribed contour boundary, average depth value, and area of ​​each defect region. To facilitate human-computer interaction and subsequent parameter adjustment, this matrix is ​​further rendered using a heatmap visualization method, employing color gradients to represent the severity of defects, thereby generating a defect distribution mapping map. The final output result contains both quantified coordinate information and supports graphical interface display, and can be directly used by the clamp diameter adjustment module.

[0056] In step S15, based on the defect distribution mapping map, the location and shape characteristics of the defects need to be analyzed, and combined with the preset sonar positioning results, the defect distribution information of the defect-affected area needs to be obtained, including: The step of analyzing the defect location and shape characteristics based on the defect distribution map, and combining this with preset sonar positioning results, to obtain defect distribution information in the defect-affected area includes: Based on the defect distribution map, image region segmentation is performed, and the coordinate records of the defect regions are extracted to obtain preliminary defect location distribution results; Based on the defect location distribution results, geometric feature extraction is performed to analyze the defect location contour structure and generate a defect shape feature set. Based on the set of defect shape features, and by comparing it with a preset defect classification model, defect type labeling information is obtained. Based on the defect type labeling information, the preset sonar signal anomaly location results are matched, and the suspected anomaly points are jointly analyzed to obtain the spatial distribution range of the defect anomaly points. Based on the spatial distribution range, defect type and abnormal point information are fused, boundary division and contour extraction are performed to obtain defect distribution information of the defect-affected area.

[0057] In a specific embodiment, the system first performs region segmentation on the image based on the defect distribution map. It uses a pixel-based clustering segmentation method (such as K-means or region growing algorithm) to divide the image into regions with different colors and depth features. It then extracts information such as the boundary coordinates and center position of the defect region and records its spatial position in the image to generate a preliminary defect location distribution result.

[0058] Subsequently, the system performs geometric feature extraction on the location results to analyze the defect contour structure. Specifically, by calculating parameters such as the area, perimeter, aspect ratio, and rate of change of curvature of each defect region, the system can generate a set of shape features containing each defect. Taking exemplary parameters as an example, a crack-type defect typically has a high aspect ratio (greater than 3) and linear edges, while corrosion-type defects exhibit an approximately circular or irregular shape, an aspect ratio close to 1, and large fluctuations in edge curvature.

[0059] For the aforementioned set of geometric features, the system introduces a pre-defined defect classification model. This model is built based on decision trees or lightweight support vector machines and is mainly used to compare the extracted shape features with known types (such as corrosion, cracks, erosion, etc.). The model loads pre-labeled feature templates, inputs feature parameters of the current region, and outputs the corresponding defect type labels and matching confidence scores, ultimately forming a defect type distribution map with labeled information.

[0060] After type identification, the system further incorporates pre-defined sonar signal anomaly localization results and employs a point-matching algorithm to spatially align the defect coordinates in the image with the anomalous signal nodes in the sonar point cloud. This matching process considers timestamp synchronization and ranging error compensation, making it particularly suitable for scenarios with low image resolution or strong illumination interference. The system marks the matched anomalous points as high-confidence defect points and performs joint analysis to determine the spatial distribution range of these defects in the sonar data.

[0061] Finally, based on the identified defect types, image coordinates, and sonar spatial point information, the system performs boundary reconstruction, using the α-shape boundary envelope extraction method to construct polygonal influence regions around all defect points. This step ensures that the identified defect influence regions not only cover the defect itself but also its potentially affected surrounding structures. The final output is the defect distribution information within the influence regions, which supports subsequent clamp diameter calculations and installation scheme optimization.

[0062] In one specific embodiment, when the system performs image region segmentation to determine the distribution of defect locations, it employs color clustering based on the K-means algorithm, with a preset cluster center number of 5. This primarily distinguishes between the background, areas with slight texture disturbances, and three main types of defect regions. The segmentation scale reference image resolution is set to 512×512 pixels, and the minimum defect region area is set to 20 pixels to exclude falsely detected tiny isolated pixel clusters.

[0063] During the geometric feature extraction process, the system uses the following preset thresholds: if the aspect ratio of the defect area is greater than 3 and the boundary curvature change is less than 0.1, it is initially identified as a crack-type defect; if the area is greater than 800 pixels, the edge closure is greater than 0.9, and the aspect ratio is close to 1 (1±0.2), it is identified as a corrosion pit-type defect. These parameters are derived from statistical analysis of historical fault images to ensure that the type differentiation criteria are practically representative.

[0064] Regarding the defect classification model, the system pre-selects Support Vector Machine (SVM) as the basic classifier, with input features including four items: area, aspect ratio, edge irregularity, and boundary curvature variance. The model was trained on 1200 defect image samples during the experimental phase, with defect type labels including five categories: "crack," "corrosion," "pitting," "exfoliation," and "non-defect." Cross-validation accuracy reached over 92%.

[0065] When matching sonar anomaly locations, the system presets a spatial error tolerance of ±10 cm and sets a timestamp matching window of ±1 s to ensure the reliability of synchronization between image and sonar data under the influence of water flow. Anomaly sonar point identification is based on strong echo anomalies and distance change thresholds. For example, if the echo intensity of a point is greater than 80 dB and the distance change with adjacent points exceeds 30%, it is marked as an anomaly location.

[0066] When reconstructing the defect-affected area, the α-shape boundary extraction algorithm introduced by the system is preset to 1.5 times the average point spacing to avoid overfitting of boundary depressions or irregular envelopes, and finally generates a closed polygonal region containing the defect body and its possible influence range.

[0067] In one specific embodiment, the system introduces a pre-defined defect classification model to achieve type labeling of defect areas. This model uses a Support Vector Machine (SVM) as the classifier structure, constructing an input vector based on structural and grayscale features to classify different types of surface defects on round piles. Specifically, the model's input features include: the pixel area of ​​the defect area, the aspect ratio of the bounding rectangle, the standard deviation of the edge curvature, the boundary closure index, and the grayscale contrast index calculated based on the grayscale co-occurrence matrix. These five features together constitute the model's feature vector, reflecting the geometric shape and texture characteristics of the defect.

[0068] For example, after detecting an abnormal region, the system extracts the following five-dimensional feature values: area of ​​1024 pixels, aspect ratio of 4.2, standard deviation of edge curvature of 0.12, boundary closure of 0.65, and grayscale contrast of 2.8. The system inputs this feature vector into a pre-defined SVM classifier, combines it with the training-obtained discrimination boundary, outputs a "crack" category label, and assigns a corresponding confidence score. The model is trained using 1500 manually annotated underwater defect image samples, covering five categories: cracks, corrosion, pitting, erosion, and non-defects. 80% of the images are used for training, and 20% are used for validation. During model training, a radial basis function (RBF) is used as the kernel function, the penalty factor C is set to 10, and the γ parameter is set to 0.05. Grid search is used to optimize the classification effect.

[0069] After deployment, the model is loaded into the system with fixed parameters, supporting rapid classification of each abnormal area after defect identification. The system uses the maximum class probability as the classification criterion and applies the identification results to subsequent spatial positioning and joint modeling of the defect-affected area, providing a clear defect type input for the final structural calculation of the clamp diameter.

[0070] In step S16, based on the defect distribution information, combined with the preset circular pile curvature and preset defect density, the adjustment range of the clamp diameter needs to be calculated to obtain the adjustment range, including: The adjustment range for the clamp diameter is calculated based on the defect distribution information, combined with the preset circular pile curvature and preset defect density, to obtain the adjustment range, including: Based on the defect distribution information, the surface of the circular pile is analyzed using a curvature measurement method to identify areas of abnormal curvature and obtain a circular pile curvature dataset. Based on the defect distribution information, defect distribution density estimation is performed to identify high-density defect regions and obtain a defect density dataset. Based on the circular pile curvature dataset and the defect density dataset, a spatial mapping relationship between the defect location and the clamp installation stress point is established, and the correlation parameters between the location and the stress point are extracted to obtain the parameter mapping result. Based on the parameter mapping results, the adjustment range of the clamp diameter is calculated using regression analysis, and the range is divided into intervals to obtain the adjustment interval.

[0071] In one specific embodiment, to achieve precise adjustment of the clamp diameter, the system first performs structural curvature analysis on the surface of the circular pile based on the acquired defect distribution information. Specifically, the system uses the discrete Gaussian curvature measurement method. This method uses the normal vector of each surface sampling point as a reference to construct a local triangular mesh around it, and estimates the Gaussian curvature value of that point by calculating the included angle and area between each mesh. Gaussian curvature can be used to identify local unevenness and curvature abrupt changes on the surface of the circular pile, and its physical meaning is to reflect the degree of geometric curvature per unit area. For example, in actual operation, the system is set to sample once every 500 mm². If the Gaussian curvature value in a certain mesh is greater than 0.12 (in m⁻²), it is marked as a curvature abnormality area and recorded in the circular pile curvature dataset.

[0072] In one specific embodiment, the system performs density estimation on the spatial distribution of defects to determine areas of concentrated defects. This process employs a fixed-side-length grid method, unfolding the circular pile axially into a two-dimensional rectangular surface, dividing it into several grid regions with sides of 0.2m × 0.2m. The proportion of defect pixels within each grid is calculated, and the density level is determined accordingly. For example, when the defect coverage area within a grid exceeds 40%, the system defines it as a high-density defect region and adds it to the defect density dataset. This method effectively quantifies the degree of defect aggregation, aiding in subsequent assessments of whether adjustments to the clamp's deformation and fit structure are necessary.

[0073] In one specific embodiment, to link structural geometry with defect characteristics, the system establishes a spatial mapping model between defect points and the stress points of the clamps to be installed. This model uses the three-dimensional coordinate system of the circular pile as a reference, and for each high-density defect point, calculates its Euclidean distance to the target stress point, its normal angle, and its projected area ratio. These parameters reflect the influence of the defect point on the force transmission path. For example, if a defect point has a distance of 45mm from the stress point, a normal angle of 37°, and a projected area ratio of 0.27, the system evaluates this point as a "medium-strong interference point" and calculates an influence coefficient based on a weighted average of the three factors, forming a parameter mapping result table.

[0074] In one specific embodiment, the system utilizes ridge regression to establish a mathematical model relating the hoop diameter to the aforementioned influencing parameters. The core advantage of ridge regression lies in its ability to stably output parameter estimates even in the presence of multicollinearity and to suppress overfitting. During training, the system uses historical engineering project data and simulated assembly data as sample sets, setting a regularization coefficient α=0.3 to complete model coefficient fitting. For example, if the predicted hoop diameter adjustment value output by the model is +3.4mm within a specific circular pile area, the system defines the diameter adjustment interval as [+2.9mm, +3.9mm] within a fluctuation range of ±0.5mm, serving as the input for the next stage of iterative optimization.

[0075] In step S17, based on the adjustment range and combined with the simulation results of preset geometric parameters and preset fit, the diameter configuration needs to be iteratively optimized to obtain the optimal clamp diameter value, including: Based on the adjustment range, and combined with the simulation results of preset geometric parameters and preset fit, the diameter configuration is iteratively optimized to obtain the optimal clamp diameter value; Based on the adjustment range, the actual geometric parameters of the circular pile are collected and compared with the preset geometric benchmark to identify structural areas with differences and obtain the geometric calibration requirement distribution. Based on the geometric calibration requirement distribution, dynamic calibration processing of geometric parameters is performed. Combined with the preset fit simulation results, an optimized distribution scheme of the force points for clamp installation is constructed to obtain the optimized configuration result of the force points. Based on the optimized configuration of the force points, the fit error is calculated, the error exceeding the limit area is identified, the error adjustment requirement distribution is formed, and the error evaluation result set is obtained. Based on the error evaluation result set, the diameter parameter is adjusted in multiple rounds using an iterative optimization algorithm, and the optimal clamp diameter value is obtained by comparison and verification according to the clamp diameter configuration rules.

[0076] In one specific embodiment, to obtain the optimal clamp diameter value within the adjustment range, the system first performs actual geometric parameter acquisition on the target circular pile area. This is achieved by simultaneously using laser scanning and underwater visual measurement instruments to record the diameter, height, ellipticity, and local concavity / convexity variations at various locations on the circular pile. Taking a 2.5-meter diameter circular pile as an example, the preset geometric reference values ​​include a standard diameter of 2.50 ± 0.01 meters, an axial tilt angle not exceeding 1.5°, and a maximum roundness deviation not exceeding 5 mm. When the measured diameter at a certain detection point is 2.536 meters and the roundness deviation reaches 7 mm, the system automatically determines that there is a structural deviation in that area and marks it as a geometric calibration requirement area. The entire geometric deviation distribution forms a "geometric calibration requirement distribution," which serves as the target object for subsequent calibration processing.

[0077] Specifically, for the aforementioned calibration requirement area, the system invokes the dynamic adjustment module for geometric parameters to execute a fitting-correction operation process. This process first uses a cubic B-spline interpolation algorithm to fit curves at each measurement point, establishing a fitting curve for the circular pile cross-section. Then, based on the set maximum deformation compensation limit (e.g., ±5mm), a geometric correction matrix is ​​constructed to correct the fitting profile of the area, and the corrected geometric model is output. The system then inputs the corrected model into a preset fit simulation module. This module establishes a contact model between the circular pile and the clamp based on the finite element method, setting the material elastic modulus to 210GPa, Poisson's ratio to 0.3, and the element size to 0.01m. It simulates the installation process through node constraints, obtaining optimized results for the distribution of stress points under different diameter parameters. For example, in the current configuration, the number of stress concentration points increases from 4 to 6, and the balance of the spacing between points improves by 12%, demonstrating good optimization results.

[0078] Specifically, the system calculates the fit error of the force distribution scheme. Fit is defined as the maximum gap error between the actual contact surface and the inner wall of the designed clamp, and the system uses the least squares deviation evaluation method for quantification. If the maximum deviation in a certain area reaches 0.024 meters, exceeding the preset tolerance (the tolerance threshold is set at 0.02 meters), the system determines that the fit in that area is substandard, forms an "error adjustment requirement distribution," and records the relevant error values ​​as an "error evaluation result set."

[0079] Specifically, the system initiates an iterative diameter optimization process based on the error assessment result set. The optimization method used is the gradient descent algorithm, with an initial diameter value of 2.51 meters, a step size of 0.005 meters, and a maximum number of iterations of 50. In each iteration, the system re-invokes the fit simulation module and the error assessment process to verify the diameter adjustment results. If the error value is below 0.02 meters for three consecutive iterations and the force distribution is stable, the current diameter configuration is deemed to meet the accuracy requirements. For example, in the 9th iteration, the system outputs a diameter value of 2.505 meters, a maximum fit error of 0.015 meters, and a force balance index of 0.91, satisfying the clamp diameter configuration rules, and is ultimately determined to be the optimal clamp diameter value. This value will serve as the basis for the generation and execution of subsequent control commands.

[0080] In step S18, it is necessary to generate underwater installation control commands based on the optimal clamp diameter value, verify parameter stability, and output the final installation parameter scheme, including: The underwater installation control command is generated based on the optimal clamp diameter value, the parameter stability is verified, and the final installation parameter scheme is output.

[0081] Based on the optimal clamp diameter value, an initial installation instruction set for the underwater equipment is generated, and consistency verification is performed against the preset installation parameter standard to obtain the initial control instruction dataset. Based on the initial control command dataset, a simulation analysis of fit error and installation force point distribution is performed, and the results are judged against a preset stability standard to obtain the matching verification results. When the matching verification result does not meet the preset stability standard, the fitting error value and the force point distribution scheme are adjusted and optimized to generate an adjustment control dataset; Based on the adjusted control dataset, combined with the control requirements of the underwater equipment, parameter integration and path mapping are performed to output the final installation parameter scheme.

[0082] In one specific embodiment, after obtaining the optimal clamp diameter value (e.g., 2.505 meters), the system first calls the equipment control module to generate a corresponding initial installation instruction set. This set includes key parameters such as the clamp manipulator's opening angle, positioning and movement path, and installation rhythm control. Taking the current example, the system sets the initial opening angle to 120°, the movement path step size to 0.05 meters, and the control frequency to 10Hz. It then performs consistency verification against a set of preset installation parameter standards (e.g., trajectory deviation <0.02 meters, action response delay <0.1 seconds). The verification process calls the parameter consistency verification module and outputs a complete set of initial control instruction datasets.

[0083] Specifically, the system inputs the aforementioned instruction dataset into the simulation analysis module to perform a joint simulation of fit error and force point distribution. The fit simulation relies on a finite element contact model, setting the material stiffness to 210 GPa and the friction coefficient to 0.15, to simulate gaps under different postures during installation and determine if any fit deviation zones exist. The concurrently performed force point simulation model treats the clamp as a multi-point force-bearing body, setting the number of force points to 6, with an average spacing of 0.42 meters, and a symmetry error not exceeding ±0.03 meters. The system calculated a fit error of 0.022 meters, with the maximum local error exceeding the preset tolerance (<0.02 meters), and the force distribution being uneven, resulting in a matching verification result that does not meet the stability standard.

[0084] In response to the above results, the system automatically enters the instruction adjustment process to optimize the fit error value and the position of the force points. The adjustment strategy is based on the gradient descent method combined with a heuristic position fine-tuning algorithm, gradually adjusting the opening angle (increase or decrease by 0.5°), the number of path subdivision points, and the positioning posture sequence, while updating the installation action instructions in real time. After optimization, a new adjustment control dataset is generated, and error simulation and force balance evaluation are re-executed in the simulation verification module until all key indicators fall within the set stability range.

[0085] Specifically, the system integrates the adjustment control dataset with the structural control requirements of the underwater equipment (such as hydraulic response time, feedback encoder resolution, cable winding constraints, etc.) to establish a path mapping relationship between the equipment's motion path and the target installation parameters. The final installation parameter scheme, consisting of the robotic arm trajectory, attitude control sequence, angle stepping plan, and hydraulic flow setting, is output through the parameter mapping interface. This scheme is encapsulated in a standard JSON control command format and synchronously pushed to the command cache of the underwater execution module for automatic deployment in subsequent installation tasks.

[0086] It should be noted that the system's preset stability standards include two core indicators: the fit error must not exceed 0.02 meters, and the stress distribution balance index must not be lower than 0.9. The fit error is assessed through simulation calculations of the average and maximum gaps between the circular pile and the clamp in different contact areas, ensuring no significant loosening or jamming after installation. The stress balance index is calculated based on the consistency of load distribution at each stress point; if the maximum load deviation exceeds 10%, it is considered unbalanced, thus ensuring the structural stability and stress reliability of the equipment during long-term underwater operation.

[0087] Reference Figure 2 The second embodiment of the present invention provides a clamp diameter adjustment and operating system for underwater robots to detect apparent defects in circular piles, including: The data acquisition module is used to acquire image data and sonar data of the underwater circular pile surface to obtain an initial surface information dataset; The data cleaning module is used to perform image enhancement and sonar filtering based on the initial surface information dataset, remove interference noise, and obtain a clear surface information set. The correction feature module is used to extract surface texture and structural features based on the set of clear surface information, correct imaging errors, and obtain a corrected feature information library. The defect mapping module is used to identify defect regions using a convolutional neural network based on the feature information database, and obtain a defect distribution mapping map. The defect distribution module is used to analyze the location and shape characteristics of defects based on the defect distribution map, and combine the preset sonar positioning results to obtain the defect distribution information of the defect-affected area. The adjustment range module is used to calculate the adjustment range of the clamp diameter based on the defect distribution information, combined with the preset circular pile curvature and preset defect density. The optimal diameter module is used to iteratively optimize the diameter configuration based on the adjustment range, combined with preset geometric parameters and preset fit simulation results, to obtain the optimal clamp diameter value. The installation control module is used to generate underwater installation control commands based on the optimal clamp diameter value, verify parameter stability, and output the final installation parameter scheme.

[0088] It should be noted that the underwater robot clamp diameter adjustment and operating system for detecting apparent defects in round piles provided in this embodiment of the invention is used to execute all the process steps of the underwater robot clamp diameter adjustment and operation method for detecting apparent defects in round piles in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0089] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a data cleaning program. When the processor executes the computer program, it implements the steps in the embodiments of the above-described underwater robot methods for adjusting the diameter of the clamp for detecting apparent defects in circular piles, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the defect distribution module.

[0090] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0091] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0092] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0093] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0094] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0095] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0096] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for adjusting the diameter of a clamp used by an underwater robot to detect apparent defects in a round pile, characterized in that... include: Acquire image and sonar data of the underwater circular pile surface to obtain an initial surface information dataset; Based on the initial surface information dataset, image enhancement and sonar filtering are performed to remove interference noise and obtain a clear surface information set. Based on the set of clear surface information, surface texture and structural features are extracted, imaging errors are corrected, and a corrected feature information library is obtained. Based on the feature information database, a convolutional neural network is used to identify defect regions and obtain a defect distribution mapping map; Based on the defect distribution map, the location and shape characteristics of the defects are analyzed, and combined with the preset sonar positioning results, the defect distribution information of the defect-affected area is obtained. Based on the defect distribution information, combined with the preset circular pile curvature and preset defect density, the adjustment range of the clamp diameter is calculated to obtain the adjustment range; Based on the adjustment range, and combined with the simulation results of preset geometric parameters and preset fit, the diameter configuration is iteratively optimized to obtain the optimal clamp diameter value; Based on the optimal clamp diameter value, underwater installation control commands are generated, parameter stability is verified, and the final installation parameter scheme is output.

2. The method for adjusting and operating the clamp diameter for underwater robot detection of apparent defects in round piles according to claim 1, characterized in that, The step of performing image enhancement and sonar filtering based on the initial surface information dataset to remove interference noise and obtain a clear surface information set includes: Based on the initial surface information dataset, the image data is denoised using a wavelet transform algorithm to remove high-frequency noise generated by water flow disturbance, resulting in denoised image data. Based on the denoised image data, histogram equalization is used to enhance the image, thereby improving image contrast and texture clarity, resulting in enhanced image data. Based on the initial surface information dataset, the sonar data is time-domain filtered using an adaptive Kalman filter method to suppress background noise interference, resulting in filtered sonar data. Based on the sonar filtering data, the main frequency signal is extracted using the short-time Fourier transform method, and invalid frequency bands are removed to obtain enhanced sonar data; Based on the enhanced image data and enhanced sonar data, alignment is performed according to timestamps, and multi-source data fusion is completed to obtain a clear surface information set.

3. The method for adjusting and operating the clamp diameter for underwater robot detection of apparent defects in round piles according to claim 1, characterized in that, The step of extracting surface texture and structural features based on the set of clear surface information, correcting imaging errors, and obtaining a corrected feature information library includes: The set of clear surface information is decomposed into multiple scales using the curvature scale analysis method to extract local curvature abrupt change regions and obtain the texture feature dataset. Based on the texture feature dataset, a spatial registration algorithm is used to align the multi-scale texture features, and the structural deformation deviation value is extracted by combining the preset sonar point cloud reconstruction results to obtain the aligned texture and deformation feature dataset. Based on the alignment texture and the deformation feature dataset, an eigenvalue correction algorithm based on the light field refraction model is applied to correct the pixel offset error caused by imaging distortion, and the correction feature data is obtained. Based on the corrected feature data, texture anomalies and deformation deviation values ​​are fused, and time alignment and consistency checks are performed to obtain the corrected feature information library.

4. The method for adjusting and operating the clamp diameter for underwater robot detection of apparent defects in round piles according to claim 1, characterized in that, The step of identifying defect regions using a convolutional neural network based on the feature information database to obtain a defect distribution map includes: The feature information database is divided into regions, and a set of candidate abnormal regions is extracted to obtain the initial abnormal distribution range; Based on the initial anomaly distribution range, perform texture feature edge detection to obtain the defect region; The defect region is depth-quantized by combining the gray-level co-occurrence matrix analysis method to obtain the numerical distribution data of the defect depth; Based on the numerical distribution data, a preset convolutional neural network model is used to extract and identify features of abnormal regions to obtain defect region determination results. Based on the defect region determination results, the defect boundary contour and depth values ​​are fused to generate a two-dimensional defect location matrix. A visualization mapping method is used to render a heat map of the defect region in the defect region determination results to obtain a defect distribution mapping map.

5. The method for adjusting and operating the clamp diameter for underwater robot detection of apparent defects in round piles according to claim 1, characterized in that, The step of analyzing the defect location and shape characteristics based on the defect distribution map, and combining this with preset sonar positioning results, to obtain defect distribution information in the defect-affected area includes: Based on the defect distribution map, image region segmentation is performed, and the coordinate records of the defect regions are extracted to obtain preliminary defect location distribution results; Based on the defect location distribution results, geometric feature extraction is performed to analyze the defect location contour structure and generate a defect shape feature set. Based on the set of defect shape features, and by comparing it with a preset defect classification model, defect type labeling information is obtained. Based on the defect type labeling information, the preset sonar signal anomaly location results are matched, and the suspected anomaly points are jointly analyzed to obtain the spatial distribution range of the defect anomaly points. Based on the spatial distribution range, defect type and abnormal point information are fused, boundary division and contour extraction are performed to obtain defect distribution information of the defect-affected area.

6. The method for adjusting and operating the clamp diameter for underwater robot detection of apparent defects in round piles according to claim 1, characterized in that, The adjustment range for the clamp diameter is calculated based on the defect distribution information, combined with the preset circular pile curvature and preset defect density, to obtain the adjustment range, including: Based on the defect distribution information, the surface of the circular pile is analyzed using a curvature measurement method to identify areas of abnormal curvature and obtain a circular pile curvature dataset. Based on the defect distribution information, defect distribution density estimation is performed to identify high-density defect regions and obtain a defect density dataset. Based on the circular pile curvature dataset and the defect density dataset, a spatial mapping relationship between the defect location and the clamp installation stress point is established, and the correlation parameters between the location and the stress point are extracted to obtain the parameter mapping result. Based on the parameter mapping results, the adjustment range of the clamp diameter is calculated using regression analysis, and the range is divided into intervals to obtain the adjustment interval.

7. The method for adjusting and operating the clamp diameter for underwater robot detection of apparent defects in round piles according to claim 1, characterized in that, Based on the adjustment range, and combined with the simulation results of preset geometric parameters and preset fit, the diameter configuration is iteratively optimized to obtain the optimal clamp diameter value; Based on the adjustment range, the actual geometric parameters of the circular pile are collected and compared with the preset geometric benchmark to identify structural areas with differences and obtain the geometric calibration requirement distribution. Based on the geometric calibration requirement distribution, dynamic calibration processing of geometric parameters is performed. Combined with the preset fit simulation results, an optimized distribution scheme of the force points for clamp installation is constructed to obtain the optimized configuration result of the force points. Based on the optimized configuration of the force points, the fit error is calculated, the error exceeding the limit area is identified, the error adjustment requirement distribution is formed, and the error evaluation result set is obtained. Based on the error evaluation result set, the diameter parameter is adjusted in multiple rounds using an iterative optimization algorithm, and the optimal clamp diameter value is obtained by comparison and verification according to the clamp diameter configuration rules.

8. The method for adjusting and operating the clamp diameter for underwater robot detection of apparent defects in round piles according to claim 1, characterized in that, The underwater installation control command is generated based on the optimal clamp diameter value, the parameter stability is verified, and the final installation parameter scheme is output. Based on the optimal clamp diameter value, an initial installation instruction set for the underwater equipment is generated, and consistency verification is performed against the preset installation parameter standard to obtain the initial control instruction dataset. Based on the initial control command dataset, a simulation analysis of fit error and installation force point distribution is performed, and the results are judged against a preset stability standard to obtain the matching verification results. When the matching verification result does not meet the preset stability standard, the fitting error value and the force point distribution scheme are adjusted and optimized to generate an adjustment control dataset; Based on the adjusted control dataset, combined with the control requirements of the underwater equipment, parameter integration and path mapping are performed to output the final installation parameter scheme.

9. A clamp diameter adjustment and operating system for underwater robots to detect apparent defects in round piles, characterized in that, include: The data acquisition module is used to acquire image data and sonar data of the underwater circular pile surface to obtain an initial surface information dataset; The data cleaning module is used to perform image enhancement and sonar filtering based on the initial surface information dataset, remove interference noise, and obtain a clear surface information set. The correction feature module is used to extract surface texture and structural features based on the set of clear surface information, correct imaging errors, and obtain a corrected feature information library. The defect mapping module is used to identify defect regions using a convolutional neural network based on the feature information database, and obtain a defect distribution mapping map. The defect distribution module is used to analyze the location and shape characteristics of defects based on the defect distribution map, and combine the preset sonar positioning results to obtain the defect distribution information of the defect-affected area. The adjustment range module is used to calculate the adjustment range of the clamp diameter based on the defect distribution information, combined with the preset circular pile curvature and preset defect density. The optimal diameter module is used to iteratively optimize the diameter configuration based on the adjustment range, combined with preset geometric parameters and preset fit simulation results, to obtain the optimal clamp diameter value. The installation control module is used to generate underwater installation control commands based on the optimal clamp diameter value, verify parameter stability, and output the final installation parameter scheme.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the clamp diameter adjustment and operation method for underwater robot detection of apparent defects in round piles as described in any one of claims 1 to 7.

Images