A method and device for underwater pier detection and repair based on flexible tentacles

Abstract

The application provides a kind of underwater pier detection and repair method and device based on flexible tentacle, wherein the method comprises: robot carries out autonomous navigation and environmental perception, travels to the preset detection area;When approaching the target pier of the detection area, perform bionic winding action through flexible tentacle, adaptively wrap the target pier;The attachment area of the flexible tentacle is optically scanned, the original underwater image is collected, and the original underwater image is processed, to determine the crack area;According to the identified crack area, carry out hierarchical evaluation, generate decision instruction and carry out crack repair.Through the application, energy consumption, noise emission are significantly reduced, and physical damage to the ecological community on the pier surface is minimized, solving the contradiction between the existing ROV attachment stability and low disturbance demand.

Description

Technical Field

[0001] This invention relates to the field of underwater robot technology, and in particular to a method and apparatus for underwater bridge pier inspection and repair based on flexible tentacles. Background Technology

[0002] Underwater bridge structures, especially piers, abutments, and pile foundations, are subjected to long-term water erosion, alternating wet and dry conditions, and chemical corrosion, making them prone to cracks, spalling, exposed reinforcement, and corrosion. Regular and precise inspection and maintenance of these underwater structures are crucial for ensuring bridge operational safety and extending their service life. However, traditional underwater structure inspection methods and existing mainstream technologies still have significant shortcomings in terms of efficiency, accuracy, safety, and functionality, making it difficult to meet the needs of modern, intelligent infrastructure operation and maintenance. The following analysis delves into the bottlenecks of existing Remotely Operated Vehicle (ROV) technology to clarify the technical necessity of this invention.

[0003] 1. Insufficient near-wall stability and positioning: Existing ROVs generally rely on multi-vector thrusters to resist water flow and maintain hovering or close proximity to structures. When near bridge pier surfaces, the thruster wake violently agitates the water, causing sediment to suspend and the water to become turbid, severely degrading the data quality of optical and acoustic sensors. Simultaneously, in complex flow fields, ROVs are prone to continuous attitude drift and jitter, making true "static positioning" impossible. This results in blurred images and misaligned sequences, posing significant challenges to subsequent automatic identification. Some technologies attempt to use suction cups, magnetic adsorption, or robotic arms for fixation, but rigid suction cups require extremely high surface flatness and have poor sealing and insufficient adsorption force on rough concrete surfaces with deposits; magnetic adsorption is only suitable for steel structures; and rigid robotic arms suffer from collision damage to structural surfaces, poor flexibility, and complex control.

[0004] 2. Weak Intelligent Sensing and Automatic Defect Recognition Capabilities: Currently, most ROVs used for inspection are essentially "mobile cameras," whose core function is to transmit underwater video streams back to the surface control console for real-time visual interpretation by the operator. This mode not only places extremely high demands on the operator and is prone to fatigue, but also cannot eliminate human subjectivity. Although a few cutting-edge studies have attempted to integrate target detection algorithms based on convolutional neural networks, these algorithms are usually trained on images under surface or good water quality conditions. When directly applied to real-world images severely degraded by complex underwater optical effects (such as color distortion, fogging, and non-uniform lighting), their performance drops sharply, resulting in high false positive and false negative rates, and making it particularly difficult to reliably identify minute cracks and defects with blurred boundaries.

[0005] 3. Limited System Functionality and Lack of Task Expansion Capability: Commercially available inspection ROVs are typically customized closed systems with fixed sensors (such as cameras and sonar), and their task is limited to data acquisition. If different operations such as cleaning, repair, or sampling are required, a separate dedicated operational ROV must be used, or the task must be performed by a diver. This "one machine, one function" model leads to problems such as high equipment investment, fragmented workflows, and difficulties in coordinating different tasks, failing to achieve an efficient closed loop of "inspection-evaluation-treatment."

[0006] 4. Significant Continuous Power Consumption and Ecological Disturbance: The operation mode, which relies on thrusters to maintain attitude, consumes a great deal of energy, limiting the continuous operating time of ROVs. More importantly, the noise, eddies, and suspended sediment generated by the continuous operation of the thrusters will cause continuous disturbance to the aquatic ecological environment around the bridge piers, potentially affecting fish habitats and destroying benthic communities, which contradicts increasingly stringent environmental protection requirements.

[0007] There is currently no effective solution to the contradiction between the existing ROV's attachment stability and the requirement for low disturbance. Summary of the Invention

[0008] This invention provides a method and apparatus for underwater bridge pier inspection and repair based on flexible tentacles, in order to solve the defect of the contradiction between the attachment stability and low disturbance requirements of existing ROVs.

[0009] In a first aspect, the present invention provides a method for underwater bridge pier detection and repair based on flexible tentacles, comprising:

[0010] The robot navigates autonomously and perceives its environment, then travels to a pre-defined detection area.

[0011] When approaching the target pier in the detection area, the device performs a biomimetic wrapping motion using flexible tentacles to adaptively wrap around the target pier.

[0012] Optical scanning is performed on the attachment area of ​​the flexible tentacles to acquire raw underwater images, and image processing is performed on the raw underwater images to determine the crack area;

[0013] Based on the identified crack areas, a graded assessment is conducted, decision instructions are generated, and crack repair is carried out.

[0014] According to the present invention, a method for underwater bridge pier inspection and repair based on flexible tentacles is provided. The robot performs autonomous navigation and environmental perception, and travels to a pre-set inspection area, including:

[0015] The robot's current position information is calculated in real time using an ultra-short baseline positioning system and a pose reference system that integrates an inertial navigation unit.

[0016] As the robot approaches the target bridge pier, it scans the fan-shaped area in front of it to generate a preliminary point cloud map.

[0017] Based on the simultaneous localization and mapping algorithm, combined with the primary point cloud map, an underwater 3D environment model of the detection area is constructed and updated, and the optimal path to the detection area is planned.

[0018] According to the present invention, a method for underwater bridge pier detection and repair based on flexible tentacles is provided. When approaching a target bridge pier in the detection area, the method utilizes flexible tentacles to perform a biomimetic wrapping action to adaptively wrap the target bridge pier, comprising:

[0019] When the distance between the robot and the surface of the target bridge pier reaches a preset range, the flexible tentacle actively extends, and the tactile sensing fibers at the end begin to contact and sense the surface contour of the target bridge pier.

[0020] The flexible tentacles perform a simulated wrapping action, adaptively wrapping around the bridge pier, and tightening and locking through an internal drive mechanism to form a stable mechanical connection with the target bridge pier;

[0021] During the process of wrapping the bridge pier, the pressure of each segment of the flexible tentacle is adjusted in real time;

[0022] After the bonding is complete, turn off the robot's main thrusters.

[0023] According to the present invention, a method for underwater bridge pier detection and repair based on flexible tentacles is provided, wherein optical scanning is performed on the attachment area of ​​the flexible tentacles to acquire original underwater images, and image processing is performed on the original underwater images to determine the crack area, including:

[0024] Based on the stable attachment of the flexible tentacles, the robot's detection soft robotic arm moves along the planned path to perform optical scanning of the attachment area from different angles and distances, acquiring raw underwater images;

[0025] The original underwater image is enhanced to obtain a high-quality image;

[0026] Defect identification is performed on the high-quality image to determine the crack area of ​​the target bridge pier.

[0027] According to the present invention, a method for underwater bridge pier detection and repair based on flexible tentacles is provided, wherein the original underwater image is subjected to image enhancement processing to obtain a high-quality image, including:

[0028] A transformer-based denoising diffusion network is established. Gaussian noise is gradually added to a clear water surface image to obtain a noisy image.

[0029] A conditional guidance mechanism is adopted, using the original underwater image as a condition, until the denoising process is completed and the high-quality image is generated.

[0030] According to the present invention, a method for underwater bridge pier detection and repair based on flexible tentacles is provided, which involves defect identification of the high-quality image and determining the crack area of ​​the target bridge pier, including:

[0031] The high-quality images are input into a crack detection and segmentation model built on the YOLO framework to capture micro-crack features;

[0032] Multi-scale feature optimization is performed using the attention scale sequence fusion module, and the bounding box, confidence score, and pixel-level segmentation mask of the crack in the target region are output.

[0033] According to the present invention, a method for underwater bridge pier detection and repair based on flexible tentacles is provided, which performs graded assessment based on the identified crack areas, generates decision instructions, and performs crack repair, including:

[0034] A preliminary classification and risk assessment are conducted based on the size, shape, and density of the cracked areas.

[0035] If the decision is to repair the cracked area, a decision instruction is generated, the robot's toolbox is activated, the specified working tool is replaced, and the robot is guided to the cracked area to perform the repair operation.

[0036] According to the present invention, a method for underwater bridge pier detection and repair based on flexible tentacles, after repairing the cracked area, includes:

[0037] Release the flexible tentacles from their entanglement, activate the robot's main thrusters to detach the robot from the target pier surface, and move it to the next adjacent attachment area to repair the crack.

[0038] During the detection and repair process of the detection area, the original underwater image, the high-quality image, the identification results of the crack area, the repair operation log, and the sensor data are transmitted to the surface control station to generate a detection report;

[0039] After the repair work in the detection area is completed, the robot is controlled to return to the water surface recovery point along a safe path.

[0040] Secondly, the present invention also provides an apparatus for underwater bridge pier detection and repair based on flexible tentacles, used to implement the above-mentioned method for underwater bridge pier detection and repair based on flexible tentacles, comprising:

[0041] The main frame integrates an operating compartment, a control compartment, and a power compartment; the control compartment is equipped with an artificial intelligence processing unit for real-time operation of underwater image enhancement, crack recognition algorithms, and multi-sensor fusion positioning algorithms.

[0042] A buoyancy adjustment device used to adjust the buoyancy of its own weight relative to the water surface;

[0043] A biomimetic flexible tentacle mechanism is used to wrap around and encase the target bridge pier in the detection area.

[0044] According to the device for underwater bridge pier detection and repair based on flexible tentacles provided by the present invention, the artificial intelligence processing unit integrates an intelligent control and sensing system, including:

[0045] The perception fusion layer is used to scan the underwater environment, generate an underwater three-dimensional environment model of the detection area, acquire raw underwater images, and perform image enhancement processing on the raw underwater images.

[0046] The intelligent decision-making layer is used to detect and identify cracks in the target bridge pier, determine the crack area, and generate decision instructions.

[0047] The collaborative execution layer is used to perform crack repair operations according to the decision instructions.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] This invention provides a method for underwater bridge pier inspection and repair based on flexible tentacles. The biomimetic flexible tentacles achieve "zero-disturbance adaptive bonding," ensuring data acquisition quality and platform stability. The flexible tentacles can adaptively wrap around and envelop the surface of concrete bridge piers with uneven surfaces, attached organisms, and irregular geometry, overcoming inherent defects such as the high flatness requirements of rigid suction cups, the sensitivity of eddy current adsorption to materials, and the potential for surface damage by robotic arms. After the tentacles complete wrapping and locking, the robot's main thrusters can be completely shut off, and the system enters a "static wall-attached" state. This eliminates water turbidity, robot vibration, and sediment suspension caused by continuous thruster operation, providing a stable and clear data acquisition environment for optical and acoustic sensors. The flexible wrapping combined with a "micro-negative buoyancy" design constitutes a passively stable system resistant to water flow impact. The operation process does not require continuous thrust output, significantly reducing energy consumption and noise emissions, and minimizing physical damage to the ecological community on the bridge pier surface. This solves the contradiction between the adhesion stability and low-disturbance requirements of existing ROVs, achieving an organic unity between engineering operations and environmental protection. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0051] Figure 1 This is a flowchart of the underwater bridge pier detection and repair method based on flexible tentacles provided by the present invention;

[0052] Figure 2 This is a schematic diagram of the workflow of the robot performing underwater inspection and repair in an embodiment of the present invention;

[0053] Figure 3 This is a schematic diagram of the underwater bridge pier detection and repair device based on flexible tentacles provided by the present invention;

[0054] Figure 4 This is a front view of the overall structure of the underwater bridge pier detection and repair device based on flexible tentacles provided by the present invention;

[0055] Figure 5 A top view of the overall structure of the underwater bridge pier detection and repair device based on flexible tentacles provided by this invention;

[0056] Figure 6 This is a side view of the overall structure of the underwater bridge pier detection and repair device based on flexible tentacles provided by the present invention.

[0057] Figure label:

[0058] 1: Main frame; 2: Buoyancy adjustment device; 3: Toolbox; 4: Detection soft robotic arm; 5: Grasping soft robotic arm; 6: Main control cabin; 7: Power supply cabin; 8: Propulsion system; 9: Vision module. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0060] This invention provides a method for underwater bridge pier inspection and repair based on flexible tentacles. Figure 1 This is a flowchart of the underwater bridge pier detection and repair method based on flexible tentacles provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps:

[0061] In step S101, the robot performs autonomous navigation and environmental perception, and travels to the pre-set detection area;

[0062] Step S102: When approaching the target pier in the detection area, the flexible tentacles perform a biomimetic wrapping action to adaptively wrap the target pier.

[0063] Step S103: Optical scanning is performed on the attachment area of ​​the flexible tentacles to acquire the original underwater image, and the original underwater image is processed to determine the crack area.

[0064] Step S104: Based on the identified crack areas, a graded assessment is performed, a decision instruction is generated, and crack repair is carried out.

[0065] In this method, firstly, the ROV (Remotely Operated Vehicle) autonomously navigates and perceives its environment, navigating to a pre-defined or operator-designated detection area. When the robot approaches the target pier in the detection area, it switches to a wall-adhering operation mode. A gripping soft robotic arm (i.e., a flexible tentacle) actively extends, performing a biomimetic wrapping motion to envelop and adhere to the surface of the target pier. Based on stable adhesion, an optical scan of the adhesion area is performed to acquire raw underwater images of the target pier. These raw underwater images are then processed to enhance image quality and identify crack areas on the target pier. Finally, based on the crack area identification results, decision-making instructions are generated, and repair work is carried out on the crack areas.

[0066] In the aforementioned process, the biomimetic flexible tentacles achieved "zero-disturbance adaptive adhesion," ensuring data acquisition quality and the stability of the operating platform. Like an octopus, the flexible tentacles adaptively wrap around and envelop the surface of concrete bridge piers, which have uneven surfaces, attached organisms, and irregular geometric features. This overcomes the inherent defects of rigid suction cups, such as high flatness requirements, eddy current adsorption's sensitivity to materials, and the potential for surface damage from robotic arm gripping. After the tentacles complete the wrapping and locking, the robot's main thrusters can be completely shut off, and the system enters a "static wall-attachment" state. This eliminates water turbidity, robot vibration, and sediment suspension caused by continuous thruster operation, providing a stable and clear data acquisition environment for optical and acoustic sensors. The flexible wrapping combined with a "micro-negative buoyancy" design constitutes a passively stable system resistant to water flow impact. The operation process does not require continuous thrust output, significantly reducing energy consumption and noise emissions, and minimizing physical damage to the ecological community on the bridge pier surface. This solves the contradiction between the adhesion stability and low-disturbance requirements of existing ROVs, achieving an organic unity between engineering operations and environmental protection.

[0067] Figure 2 This is a schematic diagram illustrating the workflow of the robot performing underwater inspection and repair in an embodiment of the present invention, as shown below. Figure 2As shown, in some embodiments, step S101 involves the robot performing autonomous navigation and environmental perception, and traveling to a pre-defined detection area. This includes: using an ultra-short baseline positioning system and a pose reference system that integrates an inertial navigation unit to calculate the robot's current position information in real time; scanning the fan-shaped area in front of the robot to generate a primary point cloud map during the approach to the target bridge pier; and constructing and updating an underwater three-dimensional environment model of the detection area based on a simultaneous localization and mapping algorithm and the primary point cloud map, and planning the optimal path to reach the detection area.

[0068] For example, the robot connects to the surface control station via a fiber optic composite cable with power and high-speed data communication capabilities. After entering the water, it first uses an ultra-short baseline positioning system and an attitude reference system that integrates an inertial navigation unit to calculate its own three-dimensional position, heading, and roll angles in real time, achieving initial positioning with centimeter-level accuracy.

[0069] As the system approaches the target bridge pier area, the forward-looking sonar is activated to scan the fan-shaped area ahead, generating a preliminary point cloud map that includes the pier outline, distance, and potential obstacles. Simultaneously, the array camera and flow velocity and turbidity sensors collect data synchronously.

[0070] The robot's central control unit runs a simultaneous localization and mapping (SLAM) algorithm, fusing sonar point clouds, visual features, and inertial data to construct and update an underwater 3D environment model of the work area in real time. Based on this model, the robot's intelligent path planning module calculates a safe and efficient approach path, guiding the robot to a pre-set or operator-specified starting detection area.

[0071] In some embodiments, step S102, when approaching the target pier in the detection area, involves using flexible tentacles to perform a biomimetic wrapping motion to adaptively wrap the target pier. This includes: when the distance between the robot and the surface of the target pier reaches a preset range (0.5-1 meter), the flexible tentacles actively extend, and the tactile sensing fibers at their ends begin to contact and sense the surface contour of the target pier; the flexible tentacles perform a simulated wrapping motion to adaptively wrap the pier, and tighten and lock it through an internal drive mechanism to form a stable mechanical connection with the target pier; during the wrapping process, the pressure of each segment of the flexible tentacles is adjusted in real time to ensure a firm fit without damaging the surface; after the fit is completed, the robot's main thrusters are turned off, and the system enters a "static wall-hugging" state entirely maintained by the flexible tentacles.

[0072] As the physical interaction and operational basis of this method, this embodiment aims to solve the problem of stable and non-destructive adhesion of ROVs to complex bridge pier surfaces. It abandons traditional rigid adsorption or dynamic pushing methods and innovatively adopts multi-degree-of-freedom flexible tentacles based on the biomimetic principle of octopus tentacles. The flexible tentacles are made of flexible materials such as silicone and possess active extension, bending, and wrapping functions, enabling them to adaptively wrap and adhere to irregular concrete surfaces with unevenness and attached organisms. Their core function is to provide a zero-active-disturbance, highly stable, and highly adaptable observation and operational platform. After the flexible tentacles have completed wrapping and adhesion, the main thruster can be completely shut off, fundamentally eliminating water turbidity, machine vibration, and physical disturbance to the ecological environment caused by thruster operation.

[0073] It is worth noting that during long-distance transfers and precise positioning approaches, the propulsion system takes precedence while the flexible tentacles retract; during wall-hugging detection and operations, the flexible tentacles take precedence while the propulsion system is deactivated. These distinct operating modes ensure both mobility and operational stability.

[0074] Based on this, in step S103, the attachment area of ​​the flexible tentacles is optically scanned to acquire original underwater images, and the original underwater images are processed to determine the crack area. This includes: with the flexible tentacles stably attached, the robot's detection soft robotic arm moves along the planned path to perform optical scanning of the attachment area from different angles and distances to acquire original underwater images; the original underwater images are enhanced to obtain high-quality images; and the high-quality images are used for defect identification to determine the crack area of ​​the target bridge pier.

[0075] In this embodiment, based on stable attachment, the detection soft robotic arm begins operation, carrying a vision module and moving along a planned sub-path to perform high-resolution optical scanning of the attachment area from multiple angles and distances. The LED / laser supplementary lighting system adaptively adjusts to obtain the best illumination effect.

[0076] The crack recognition algorithm integrates an end-to-end intelligent processing pipeline consisting of an underwater image enhancement module and a crack detection and segmentation module (SDI-ASF-YOLO11). This algorithm is specifically optimized for complex underwater optical environments, ensuring high-precision crack recognition and segmentation even under low visibility conditions.

[0077] Specifically, image enhancement processing is performed on the original underwater image to obtain a high-quality image, including: establishing a transformer-based denoising diffusion network, using a clear surface image as a condition, and progressively adding Gaussian noise to obtain a noisy image; and employing a conditional guidance mechanism, using the original underwater image as a condition, until the denoising process is completed and a high-quality image is generated.

[0078] For example, the specific details of the image enhancement process are as follows:

[0079] 1. Forward diffusion process (image degradation modeling)

[0080] This process is used during the training phase to build a degradation model from a sharp image to a noisy image. Given a sharp image of water... Gaussian noise is gradually added over T steps, eventually transforming the noise into pure Gaussian noise. Each step of the noise addition process follows the formula below:

[0081]

[0082] in, In the forward process, the output image is the noisy image obtained after diffusion at step t, representing the result at step t after diffusion. The resulting, blurrier image after adding noise; During the forward process, the input image is at the t-th time step. The noisy image after one-step diffusion is represented by the image at time t. In step 1, the relatively clear image state is the starting point for calculating the noisier image in the next step; It is the transition probability of the forward diffusion process, representing the probability given the previous state. At that time, the current step Conditional distribution; The noise scheduling parameter represents the diffusion process and controls the variance of the noise added at step t. Its value is between (0,1) and usually increases with t. It is the identity matrix, representing the covariance matrix of the noise; It represents a normal distribution (Gaussian distribution) and is used to model the changes in the image at each diffusion step.

[0083] 2. Condition-guided inverse denoising process (image enhancement and reconstruction)

[0084] This is crucial for model training and inference. The model needs to learn from noise. Recover a clear image This embodiment employs a conditional guidance mechanism, using the acquired, degraded underwater image c as a condition to guide the denoising process in generating an image with the same content as the original but enhanced quality. This process is defined as follows:

[0085]

[0086] in, It is a parameterized, conditionally guided backdiffusion process that learns from noise. Restore to clear image Mapping; In the reverse process, the input image represents the relatively blurry image state to be denoised at step t, which is the starting point for predicting a clearer image. In the reverse process, the output image represents the clearer image state predicted by the model at step t-1, which is the result after denoising; c is the conditional information, which refers to the original degraded underwater image. This conditional input provides contextual guidance for the denoising process, ensuring that the generated image is consistent with the content of c but with enhanced quality, thereby achieving targeted image restoration rather than random generation. This represents the mean vector of the Gaussian distribution predicted by the denoising neural network, given condition c and time step t. This represents the covariance matrix predicted by the network.

[0087] 3. Network Architecture and Training Optimization

[0088] Network Architecture: This embodiment employs a Transformer-based denoising diffusion network. Specifically, it denoises the image... The conditional image c is concatenated along the channel dimension and input into a network consisting of convolutional layers and multiple Transformer modules. Through its self-attention mechanism, the Transformer module can better model the long-range dependencies of the image, thereby more effectively recovering the coherent texture and edge features of the crack.

[0089] Training objective: During training, the network learns to predict noise added to an image. The loss function uses mean squared error loss:

[0090]

[0091] in, This represents the training objective (loss function) of the diffusion model. is the actual noise at step t, used to represent the noise added during the actual diffusion process; It is a denoising neural network based on the current noisy state. The noise is predicted based on the conditional image c and the time step t.

[0092] Inference acceleration: To improve efficiency, a segmented skip sampling strategy is adopted. The reverse inference process can be summarized as a series of interval step length sequences. Where S (sampling step size S=10) is less than the total time step T. The step size is used in the early critical steps of the reverse process (e.g., t from a to c). Perform dense sampling, using a step size in the later stages (t from c to b). Sparse sampling is employed to achieve high-quality image reconstruction in just S=10 steps, significantly faster than the traditional T=1000 steps. A piecewise sampling method uses different sampling step sizes within the time step sequence, as detailed below:

[0093]

[0094] in and These are the different step sizes ([a, c] and [c, b]) for sampling within different intervals; c is the segmentation point (here, c is independent of the conditional image variable c), which divides the sampling interval into early stages. and late Two stages; S represents the total number of steps in the accelerated sampling process; This represents a non-uniform subsequence selected from the original T steps during accelerated reasoning (sampling).

[0095] Defect identification of high-quality images and determination of crack regions in target bridge piers include: inputting high-quality images into a crack detection and segmentation model based on the YOLO framework to capture micro-crack features; using an attention-scale sequence fusion module for multi-scale feature optimization, and outputting the bounding box, confidence score, and pixel-level segmentation mask of the cracks in the target region.

[0096] This embodiment accepts enhanced, high-quality images and is responsible for the precise location (detection) and contour delineation (segmentation) of cracks. Based on the original YOLO11, it integrates two core improved modules, the details of which are as follows:

[0097] 1. Semantic detail injection module

[0098] Function: It solves the problem of semantic information loss when feature maps of different scales are fused, especially enhancing the ability to perceive small-scale crack features.

[0099] Structure: Multiple input feature maps from different layers of the backbone network are convolved and their channel counts are unified. For each feature map, its importance weights are calculated through channel attention modules and spatial attention modules. All feature maps are unified to the same spatial resolution through adaptive average pooling or bilinear interpolation. Finally, the weighted multi-scale feature maps are adaptively fused to output a feature map rich in multi-scale semantic information.

[0100] 2. Attention Scale Sequence Fusion Module

[0101] Function: Enables more efficient multi-scale feature fusion, allowing the model to simultaneously focus on the overall morphology of large cracks and the local details of micro cracks.

[0102] Structure: The Scale Sequence Feature Fusion (SSFF) module unifies the size of feature maps from different scales, upsamples and stacks them, and then fuses them using 3D convolution to capture global semantic information across scales. The Triple Feature Encoder (TFE) module concatenates feature maps from large, medium, and small scales to retain richer local details. The outputs of SSFF and TFE are fed into the Channel and Position Attention Module (CPAM). This module calculates attention weights from both the channel and spatial dimensions, making the network focus more on feature channels and spatial locations related to cracks. Finally, a Cross Stage Partial (CSP) structure is used for multi-branch convolution and concatenation to achieve deep fusion of crack information.

[0103] 3. Output and Loss Function

[0104] The model ultimately outputs the crack's bounding box, class confidence score, and pixel-level segmentation mask. The entire model training process is a multi-task learning process, and its total loss function is a weighted sum of bounding box regression loss, classification loss, segmentation loss, etc. Among these, the diffusion model loss based on noise prediction... It is a key component in ensuring model performance.

[0105] In some embodiments, step S104, which involves classifying and evaluating the identified crack areas, generating decision instructions, and repairing the cracks, includes: performing preliminary classification and risk assessment based on the size, shape, and density of the crack areas; if the decision is to repair the crack areas, generating decision instructions, mobilizing the robot's toolbox, replacing the specified work tools, and guiding it to the crack areas for repair operations.

[0106] In this embodiment, the control system analyzes the crack detection results in real time. For identified cracks, a preliminary classification and risk assessment are performed based on their size, shape, and density. Simultaneously, the operator can review and assign tasks through the surface control station's interactive interface.

[0107] If a decision is made to repair a crack (such as through grouting), the control system will generate a sequence of work instructions. First, the gripping soft robotic arm is fine-tuned to align the toolbox interface area with the target work point. Then, the quick-change mechanism inside the toolbox is activated, automatically moving the currently mounted vision sensor module back into the chamber and removing and locking the designated crack grouting repair tool onto the interface.

[0108] The robotic arm, equipped with an auxiliary positioning camera, may perform secondary precise positioning. Guided by the control system, the nozzle of the grouting tool is precisely aligned with the crack location by the robotic arm, initiating the grouting process. During the operation, pressure sensors provide real-time feedback to ensure effective filling of the repair material.

[0109] In this process, the vision system (detection software robotic arm) is responsible for "seeing" and "diagnosing," while the tool system (toolbox) is responsible for "execution." The switching and coordination between the two are uniformly scheduled by the central control system based on the "diagnostic results," thus realizing a closed loop of perception and execution.

[0110] In summary, this method integrates a modular, switchable tool system. Once the identification system detects and locates a crack, the control unit can autonomously decide or respond to commands, dispatching the tentacles or specialized mechanisms to switch the current tool to a grouting repair head and guiding it precisely to the crack location for operation. This forms an integrated intelligent operation loop of "detection-decision-repair," significantly improving overall operation and maintenance efficiency. This method overcomes the limitations of traditional ROVs with their single function, integrating multiple functions such as detection, cleaning, repair, and sampling into a single unit. Through a quick-change interface, tools can be automatically switched as needed during a single dive mission, seamlessly connecting the "detection and evaluation" and "repair processing" processes that traditionally require multiple deployments and multiple devices, greatly shortening the operation cycle and reducing overall costs. Tool switching and operation execution are not preset or remotely controlled, but directly driven by the identification results of the aforementioned intelligent vision system. This method can "see the crack - analyze the crack - decide to call the repair tool - guide the tool to align for operation," forming a complete autonomous closed loop of "perception-cognition-decision-execution," reducing human intervention and improving the accuracy and consistency of operations.

[0111] Furthermore, after repairing the current crack area, the process includes: untangling the flexible tentacles, activating the robot's main thrusters to detach the robot from the target pier surface, and moving to the next adjacent attachment area for crack repair; during the detection and repair process of the detection area, the original underwater images, high-quality images, crack area identification results, repair operation logs, and sensor data are transmitted to the surface control station to generate a detection report; after the repair work in the detection area is completed, the robot is controlled to return to the surface recovery point along a safe path.

[0112] Specifically, after completing the inspection and repair work at the current attachment point, the flexible tentacles unwrap, the main thrusters activate, and the robot detaches from the bridge pier surface. Based on global planning, the system drives the ROV to the next adjacent attachment point, repeating the inspection and repair process until comprehensive inspection and selective repair of the entire target area are completed.

[0113] Throughout the entire operation, all raw data, enhanced images, identification results (including crack masks with spatial coordinates), repair operation logs, and sensor data are transmitted to the surface control station in real time or near real time via fiber optic composite cables for instant monitoring, offline analysis, and generation of detection reports.

[0114] After completing all scheduled tasks, the robot returns to the surface recovery point along a safe path according to instructions. Personnel on the water then retrieve the robot using a launching system, completing the operation.

[0115] This invention also provides a device for underwater bridge pier detection and repair based on flexible tentacles. The device for underwater bridge pier detection and repair based on flexible tentacles provided by this invention will be described below. The device for underwater bridge pier detection and repair based on flexible tentacles described below can be referred to in correspondence with the method for underwater bridge pier detection and repair based on flexible tentacles described above. Figure 3-6 As shown, Figure 3 This is a schematic diagram of the underwater bridge pier detection and repair device based on flexible tentacles provided by the present invention. Figure 4 This is a front view of the overall structure of the underwater bridge pier detection and repair device based on flexible tentacles provided by the present invention. Figure 5 A top view of the overall structure of the underwater bridge pier detection and repair device based on flexible tentacles provided by this invention. Figure 6 This is a side view of the overall structure of the underwater bridge pier detection and repair device based on flexible tentacles provided by the present invention. The device includes:

[0116] The main frame 1 integrates the work compartment, control compartment and power compartment; the control compartment is equipped with an artificial intelligence processing unit for real-time operation of underwater image enhancement, crack recognition algorithm and multi-sensor fusion positioning algorithm;

[0117] Buoyancy adjustment device 2 is used to adjust its own weight and buoyancy relative to the water surface;

[0118] The biomimetic flexible tentacle mechanism includes a detection soft robotic arm 4 and a grasping soft robotic arm (flexible tentacle) 5, used to wrap around and encase the target bridge pier in the detection area.

[0119] Specifically, the main frame 1 is made of high-strength, corrosion-resistant ultra-high molecular weight polyethylene (UHMWPE) material, which has low density (0.92–0.96 g / cm³) and high toughness (elongation at break >350%), effectively resisting underwater impact and withstanding working pressure at a depth of at least 200 meters (no plastic deformation under hydrostatic pressure of 2 MPa). The overall design adopts a compartmentalized layout, divided into three modules:

[0120] The work cabin integrates a core bionic working mechanism. The front section is a toolbox 3, featuring a modular design and a built-in quick-change interface for storing and switching various work terminals. Symmetrically arranged on both sides of the front are gripping soft robotic arms 5, while a detection soft robotic arm 4 is positioned in the center. The detection soft robotic arm 4 integrates a vision module 9 (including a miniature camera and supplementary lighting unit) at its end, responsible for close-range, multi-angle image acquisition. The gripping soft robotic arm 5, acting as a flexible tentacle, is constructed with a silicone substrate and built-in drive fibers, responsible for wrapping, holding, and fixing the bridge piers, providing a stable base for inspection and repair operations.

[0121] Control Cabin: Contains the main control cabin 6 and the power supply cabin 7. The main control cabin 6 houses an industrial-grade main controller based on a multi-core ARM (Acorn RISC Machine) architecture and a dedicated artificial intelligence processing unit, used for real-time execution of underwater image enhancement, crack detection algorithms, and multi-sensor fusion positioning algorithms; it is the "brain" of the entire system. The power supply cabin 7 integrates a high-energy-density lithium battery pack and an intelligent power distribution management system, providing stable power to the entire system.

[0122] The power compartment primarily houses the propulsion system, consisting of four horizontal vector thrusters and two vertical thrusters, with each horizontal thruster generating at least 100N of thrust. Its main function is to enable the ROV's overall maneuverability, attitude adjustment, and precise positioning and approach before the grab arms become entangled and secured, rather than providing the main force during the core wall-hugging detection phase.

[0123] The buoyancy adjustment device, through a combination of buoyancy materials and counterweights, ensures that the robot's net weight in water is slightly greater than its net buoyancy. This design, combined with a low center of gravity layout, enables low-power "bottom-sitting" standby, reducing unnecessary thruster power consumption. When the gripping arm wraps around the bridge pier, it can utilize its own weight to assist in initial contact, reducing the energy consumption of the tentacle drive and enhancing the overall system's resistance to current.

[0124] The biomimetic flexible tentacle mechanism is primarily implemented by the grasping soft robotic arm 5. The grasping soft robotic arm 5 adopts an octopus-tentacle-like structure, possessing multi-degree-of-freedom bending, twisting, and elongation capabilities. Its surface is covered with a flexible friction layer and integrates distributed pressure-sensing fibers, enabling it to sense contact force and achieve adaptive wrapping. Once the tentacle wraps around and locks onto the bridge pier, it generates strong static friction and mechanical interlocking, rigidly connecting the robot to the pier. At this point, the main thruster can be completely shut off, entering a static operation mode with "zero active disturbance."

[0125] The artificial intelligence processing unit integrates an intelligent control and perception system, including: a perception fusion layer, used to scan the underwater environment, generate an underwater three-dimensional environment model of the detection area, and acquire raw underwater images and perform image enhancement processing on the raw underwater images; an intelligent decision-making layer, used to detect and identify cracks in the target bridge pier, determine the crack area, and generate decision instructions; and a collaborative execution layer, used to execute crack repair operations according to the decision instructions.

[0126] Specifically, the perception fusion layer integrates optical, acoustic, inertial, and environmental sensor networks to provide high-quality, multimodal input data for intelligent algorithms.

[0127] Visual perception pathway: Raw underwater images are acquired by a vision module (multi-view array camera and adaptive LED / laser illumination system). Image data is transmitted to the control cabin in real time via a high-speed bus. At the algorithm processing front end, the underwater image enhancement unit first preprocesses the raw images. This unit is based on an improved conditional denoising diffusion model, guided by the aforementioned acquired degraded images, and removes noise and color cast caused by underwater scattering and absorption through an efficient reverse iterative process to reconstruct an enhanced image with clear details and faithful colors.

[0128] Multimodal perception fusion: Forward-looking imaging sonar provides obstacle information and coarse structural outlines at mid-to-long range; deep fusion of ultra-short baseline localization system and inertial measurement unit provides real-time robot position and attitude information with centimeter-level accuracy; miniature water quality sensors (turbidity, flow velocity) monitor the working environment in real time. All data are registered and fused in the spatiotemporal domain to generate an instant environmental map for navigation and obstacle avoidance, and assign precise absolute three-dimensional coordinates to each defect subsequently identified.

[0129] The intelligent decision-making layer is the command center of the system, and its decisions are based on the information from the perception fusion layer and the output of the dedicated recognition algorithm.

[0130] Core cognitive algorithm: The enhanced high-quality image is input into the crack intelligent detection and segmentation model in real time. Based on the efficient YOLO11 framework, this model introduces a Semantic and Detail Infusion (SDI) module to enhance the extraction of micro-crack features, and uses an attention-scale sequence fusion module to optimize the fusion of multi-scale crack features. Finally, it outputs the crack bounding box, confidence score, and pixel-level segmentation mask.

[0131] Adaptive task planning: The decision-making system not only plans a global coverage path based on the environmental map, but more importantly, it can receive and analyze crack identification results in real time. For example, when a dense crack area or a serious defect of a specific shape is identified, the decision layer will automatically generate instructions to schedule the gripping arm to adjust its contact position, or plan a local fine rescanning path for the area, and decide to call the corresponding repair tools in the toolbox (such as grouting heads), realizing a cognitive closed loop from "seeing" to "understanding" to "planning".

[0132] The collaborative execution layer is responsible for translating intelligent decisions into precise, coordinated physical actions. When the decision-making layer issues a work instruction, the collaborative control system dispatches a soft robotic arm carrying a vision module to perform a fixed-point re-inspection, while simultaneously directing the gripping soft robotic arm to make fine adjustments to optimize its working posture. Subsequently, the control system drives the toolbox's quick-change mechanism to switch the current tool (such as a camera) to the designated work tool (such as a cleaning brush head or grouting head), which is then guided by the robotic arm to the target location to perform precise operations.

[0133] The system employs a dynamic power consumption management strategy. During long-distance maneuvers, priority is given to ensuring the power of the propulsion system; during wall-hugging detection and operations, priority is given to ensuring the power supply of the artificial intelligence processing unit and sensor array in the control cabin, while significantly reducing or shutting down the thruster power to achieve optimal energy efficiency.

[0134] The system supports online updates and incremental learning of the algorithm model. New data and model parameters can be injected through the waterborne command and control station, enabling the system's identification and decision-making capabilities to continuously evolve and adapt to a wider range of defect types and more complex environments.

[0135] This device employs a highly collaborative, task-oriented system architecture, achieving optimized overall performance and enhanced reliability. All hardware designs, including the lightweight frame, compartmentalized layout, bionic tentacles, and modular toolboxes, are closely aligned with and serve the core tasks of "stable fit, intelligent recognition, and precise operation," eliminating redundancy and improving overall reliability and environmental adaptability. By fusing visual, acoustic, inertial navigation, and environmental sensor data in the spatiotemporal domain, the system not only provides itself with centimeter-level navigation and positioning accuracy but also assigns precise absolute coordinates to each identified defect and senses real-time changes in the operating environment, enabling adaptive adjustments. The system balances high performance with high reliability. Its modular design facilitates maintenance and transportation; its low-disturbance operation mode complies with environmental regulations; and its integrated intelligent operation significantly reduces reliance on human labor and operational risks.

[0136] Based on the aforementioned method and device for underwater bridge pier inspection and repair using flexible tentacles, an underwater remotely operated vehicle (ROV) was designed. The specific structure of the ROV body is as follows:

[0137] 1. ROV Carrier: As the main platform of the entire unit, it is constructed with a high-strength ultra-high molecular weight polyethylene frame, possessing excellent corrosion resistance and impact resistance. This platform integrates the buoyancy adjustment system, energy system, propulsion system, and installation interfaces for all functional modules, providing a stable and reliable underwater transport and execution base for the core intelligent algorithm system and multi-functional operating mechanisms.

[0138] 2. Bionic Flexible Tentacle System: This system is the core actuator of the ROV. It mainly consists of a gripping soft robotic arm (flexible tentacle) and a detection soft robotic arm. The gripping soft robotic arm is driven by pneumatic / hydraulic muscles or shape memory alloys, enabling multi-degree-of-freedom bending and wrapping. Its surface integrates tactile sensing fibers to sense contact force and achieve adaptive wrapping, ultimately forming a stable mechanical interlock on the bridge pier surface. The detection soft robotic arm is a lighter soft structure with a vision module integrated at its end, responsible for carrying optical sensors for fine scanning. The two sets of tentacles work together to achieve "adaptive attachment" and "precise observation / operation," which is the physical basis for the "zero-disturbance" operation mode.

[0139] 3. Control Unit: The control cabin is the nerve center and intelligent core of the ROV. It integrates an industrial-grade main controller based on a multi-core ARM architecture, a dedicated artificial intelligence processing unit (such as a computing module equipped with a GPU or NPU), an optical transceiver, a network switch, and a power management module. Its core feature is the built-in high-performance computing core dedicated to real-time operation of underwater image enhancement algorithms and intelligent crack detection and segmentation models. This unit is responsible not only for multi-sensor data fusion, motion control, and global path planning, but more importantly, its embedded "intelligent decision-making module" can receive and analyze visual recognition results in real time, dynamically generating operational instructions (such as tool switching commands and local scanning paths), truly realizing a closed loop from "perception" to "cognition" to "decision."

[0140] 4. Modular Toolbox System: Located above the front work area of ​​the ROV, this is a modular storage and exchange platform integrating standard interfaces and quick-change mechanisms. The toolbox can accommodate and lock various operational terminal modules, such as high-definition inspection cameras, crack grouting repair heads, and rotary cleaning brush heads. Its drive mechanism, under control unit commands, can automatically retrieve, retract, and lock different tools within tens of seconds, providing crucial hardware support for achieving integrated "inspection-repair" functionality.

[0141] 5. Multi-source sensing and observation system: This system adopts a distributed, multi-modal sensor layout.

[0142] Optical sensing path: The core is handled by the vision module, which integrates a high-sensitivity underwater camera and an LED / laser illumination unit adapted to the underwater spectral characteristics. The acquired raw images are directly transmitted to the control cabin via a dedicated high-speed channel for real-time enhancement and recognition processing.

[0143] Acoustic sensing system: This includes a forward-looking imaging sonar and a multibeam echo sounder, mounted on both sides of the front of the frame. The sonar is used for mid- to long-range obstacle avoidance and large-area structural contour scanning. Its data is spatiotemporally fused with visual data to assist in constructing a 3D environmental model.

[0144] Environmental sensing group: including miniature flow meters, turbidity sensors, depth gauges, etc., are distributed and installed on the frame to monitor the hydrological conditions at the operation site in real time, providing environmental parameters for tentacles to adhere to control and path planning.

[0145] The assembly and connection methods for ROV equipment are as follows:

[0146] Connection between the control unit and the main frame: The control compartment, as a sealed, independent unit, is fixed to the mounting plane of the central compartment of the main frame using stainless steel bolts at the four corners. All power and signal cables are led out through waterproof through-cabin connectors at the bottom of the compartment. A dedicated computing submodule for artificial intelligence algorithm processing connects to the main controller backplane via a high-speed serial bus, ensuring ultra-low latency for image processing data streams.

[0147] Connection between the bionic tentacle system and the frame: The drive base of the grasping soft robotic arm is rigidly connected to the main load-bearing structure on both sides of the front of the frame via a reinforced flange interface and multiple high-strength bolts. The drive pipelines (pneumatic / hydraulic) and signal cables are connected to the control unit inside the cabin through flexible protective sleeves. The base of the detection soft robotic arm is installed in a similar manner at the center of the front of the frame, and its cables are also waterproofed and tensile-resistant. The vision module, as the end effector of the detection arm, is connected to the end of the arm via a quick-release mechanical interface for easy maintenance or replacement.

[0148] Modular toolbox and frame connection: The toolbox is a single module, inserted into a dedicated slot at the top front of the frame via a bottom slider guide mechanism, and secured by hydraulic or electric locking pins on both sides. The toolbox's control and power interfaces automatically align after being pushed into place, enabling plug-and-play functionality.

[0149] Connection of Multi-Source Sensing Modules to the Frame: Sonar System: The metal mounting bracket of the forward-looking sonar is bolted to the crossbeam at the front of the frame via shock-absorbing rubber pads to isolate thruster vibration. Data and power cables are laid along the frame's cable channels and connected to the control compartment. Environmental Sensor Group: Each sensor is installed in an appropriate location on the frame according to its function (e.g., the flow meter is installed on the upstream side), all fixed with small stainless steel brackets, and the data is aggregated to the control unit via a unified waterproof bus.

[0150] Connection between the propulsion system and the frame: The six thrusters of the propulsion system (4 horizontal + 2 vertical) are directly fixed to the reinforced nodes at the front, rear, and sides of the frame via their own aluminum alloy brackets and stainless steel bolts. The power and communication cables for each thruster run separately through the cabin to the thruster drive plate in the control compartment, ensuring independent controllability.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for underwater bridge pier inspection and repair based on flexible tentacles, characterized in that, include: The robot navigates autonomously and perceives its environment, then travels to a pre-defined detection area. When approaching the target bridge pier in the detection area, a biomimetic wrapping motion is performed using flexible tentacles to adaptively wrap around the target bridge pier, including: When the distance between the robot and the surface of the target bridge pier reaches a preset range, the flexible tentacle actively extends, and the tactile sensing fibers at the end begin to contact and sense the surface contour of the target bridge pier. The flexible tentacles perform a simulated wrapping action, adaptively wrapping around the bridge pier, and tightening and locking through an internal drive mechanism to form a stable mechanical connection with the target bridge pier; During the process of wrapping the bridge pier, the pressure of each segment of the flexible tentacle is adjusted in real time; After the bonding is complete, turn off the robot's main thrusters; Optical scanning is performed on the attachment area of ​​the flexible tentacle to acquire raw underwater images, and image processing is performed on the raw underwater images to determine the crack area, including: Based on the stable attachment of the flexible tentacles, the robot's detection soft robotic arm moves along the planned path to perform optical scanning of the attachment area from different angles and distances, acquiring raw underwater images; The original underwater image is subjected to image enhancement processing to obtain a high-quality image, including: A transformer-based denoising diffusion network is established. Gaussian noise is gradually added to a clear water surface image to obtain a noisy image. A conditional guidance mechanism is adopted, using the original underwater image as a condition, until the denoising process is completed and the high-quality image is generated; Defect identification is performed on the high-quality image to determine the crack area of ​​the target bridge pier, including: The high-quality images are input into a crack detection and segmentation model built on the YOLO framework to capture micro-crack features; Multi-scale feature optimization is performed using the attention scale sequence fusion module, and the bounding box, confidence score, and pixel-level segmentation mask of the crack in the target region are output. Based on the identified crack areas, a graded assessment is conducted, decision instructions are generated, and crack repair is carried out.

2. The method for underwater bridge pier detection and repair based on flexible tentacles according to claim 1, characterized in that, The robot performs autonomous navigation and environmental perception, navigating to a pre-defined detection area, including: The robot's current position information is calculated in real time using an ultra-short baseline positioning system and a pose reference system that integrates an inertial navigation unit. As the robot approaches the target bridge pier, it scans the fan-shaped area in front of it to generate a preliminary point cloud map. Based on the simultaneous localization and mapping algorithm, combined with the primary point cloud map, an underwater 3D environment model of the detection area is constructed and updated, and the optimal path to the detection area is planned.

3. The method for underwater bridge pier detection and repair based on flexible tentacles according to claim 1, characterized in that, Based on the identified crack areas, a graded assessment is performed, decision instructions are generated, and crack repair is carried out, including: A preliminary classification and risk assessment are conducted based on the size, shape, and density of the cracked areas. If the decision is to repair the cracked area, a decision instruction is generated, the robot's toolbox is activated, the specified working tool is replaced, and the robot is guided to the cracked area to perform the repair operation.

4. The method for underwater bridge pier detection and repair based on flexible tentacles according to claim 1, characterized in that, After repairing the crack in the current crack area, the following steps are included: Release the flexible tentacles from their entanglement, activate the robot's main thrusters to detach the robot from the target pier surface, and move it to the next adjacent attachment area to repair the crack. During the detection and repair process of the detection area, the original underwater image, the high-quality image, the identification results of the crack area, the repair operation log, and the sensor data are transmitted to the surface control station to generate a detection report; After the repair work in the detection area is completed, the robot is controlled to return to the water surface recovery point along a safe path.

5. A device for underwater bridge pier detection and repair based on flexible tentacles, used to implement the method for underwater bridge pier detection and repair based on flexible tentacles as described in any one of claims 1-4, characterized in that, include: The main frame integrates the work compartment, control compartment, and power compartment; The control cabin is equipped with an artificial intelligence processing unit, which is used to run underwater image enhancement, crack recognition algorithms and multi-sensor fusion positioning algorithms in real time. A buoyancy adjustment device used to adjust the buoyancy of its own weight relative to the water surface; A biomimetic flexible tentacle mechanism is used to wrap around and encase the target bridge pier in the detection area.

6. The device for underwater bridge pier detection and repair based on flexible tentacles according to claim 5, characterized in that, The artificial intelligence processing unit integrates an intelligent control and sensing system, including: The perception fusion layer is used to scan the underwater environment, generate an underwater three-dimensional environment model of the detection area, acquire raw underwater images, and perform image enhancement processing on the raw underwater images. The intelligent decision-making layer is used to detect and identify cracks in the target bridge pier, determine the crack area, and generate decision instructions. The collaborative execution layer is used to perform crack repair operations according to the decision instructions.

Images