Underwater nondestructive testing method and system for maritime work structure

Abstract

The invention discloses an underwater nondestructive testing method and system for a maritime work structure, and belongs to the field of underwater nondestructive testing, and the method comprises the following steps: obtaining environment sensing data of underwater detection equipment, and determining environment adaptation parameters according to the environment sensing data; according to the environment adaptation parameters, generating an optimized motion path, and collecting structure surface detection data along the motion path; obtaining a defect distribution diagram according to the structure surface detection data; according to the defect distribution diagram, high-priority defect areas are divided; embedding the data of the high-priority defect area into a carrier signal to obtain steganographic data; performing end-to-end encryption on the steganographic data to obtain encrypted transmission data; and sending the encrypted transmission data to a cloud server through an underwater communication protocol, and verifying the integrity of the encrypted transmission data to obtain final nondestructive testing storage data. According to the invention, collaborative optimization of autonomous navigation, accurate defect detection and secure data transmission of equipment in a deep sea environment is realized, and the detection efficiency and data security are improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of underwater non-destructive testing, and particularly relates to a method and system for underwater non-destructive testing of marine structures. BACKGROUND

[0002] Underwater non-destructive testing of marine engineering structures is a key technology for ensuring the safe operation of infrastructure such as offshore oil platforms, offshore wind power foundations, and cross-sea bridges, and is directly related to the long-term stability of marine resource development and major projects. With the expansion of marine engineering, the detection requirements are becoming increasingly complex, and traditional detection methods are inadequate in adapting to diverse environments and complex structures. Current solutions rely heavily on manual operation or single equipment, making it difficult to meet the diverse detection needs in deep-sea extreme conditions, and there are obvious shortcomings in data security and detection efficiency. These limitations result in detection processes being susceptible to environmental interference, high costs for data processing and equipment adaptation, and difficulty in meeting the high standards of modern marine engineering.

[0003] Therefore, the core challenge of the current prior art is to achieve the autonomous operation capability of the detection equipment and the security protection of the detection data. First, the detection equipment needs to adapt to the complex seabed environment to complete the structure scanning and defect identification without human intervention. For example, in deep-sea oil platform detection, the equipment needs to navigate autonomously and accurately locate defects in strong sea currents and low visibility conditions, but existing equipment relies on surface support and lacks autonomy, resulting in low operational efficiency. Second, detection data is vulnerable to leakage risks during transmission and storage, especially when it involves critical structural defect information, data protection becomes a pressing problem. For example, if the detection data of a cross-sea bridge pile foundation is maliciously obtained, it may expose structural safety hazards and affect engineering decisions. Therefore, how to develop underwater detection equipment with autonomous navigation and intelligent identification capabilities, while ensuring the security of detection data through steganographic protection technology, has become a key issue for improving the efficiency and reliability of underwater non-destructive testing of marine engineering structures. SUMMARY

[0004] To solve the above technical problems, the present application provides a method and system for underwater non-destructive testing of marine structures. One method for underwater non-destructive testing of marine structures includes:

[0005] Obtaining environmental perception data of an underwater detection device, and determining environmental adaptation parameters based on the environmental perception data;

[0006] Generating an optimized motion path based on the environmental adaptation parameters, and collecting structure surface detection data along the motion path;

[0007] Obtaining a defect distribution map based on the structure surface detection data;

[0008] Dividing a high-priority defect area based on the defect distribution map;

[0009] embedding the high-priority defect area data into a carrier signal to obtain stego data;

[0010] performing end-to-end encryption on the stego data to obtain encrypted transmission data;

[0011] sending the encrypted transmission data to a cloud server through an underwater communication protocol and verifying the integrity thereof to obtain final lossless detection storage data.

[0012] Preferably, the process of obtaining the environment adaptation parameter according to the environment perception data comprises:

[0013] obtaining sensor raw data of the underwater detection device, denoising and standardizing the sensor raw data to obtain preprocessed data;

[0014] inputting the preprocessed data into a deep learning model to extract an environment feature vector;

[0015] calculating a sea current intensity parameter and a visibility parameter according to the environment feature vector, respectively;

[0016] performing weighted fusion on the sea current intensity parameter and the visibility parameter to obtain a comprehensive environment parameter;

[0017] if the comprehensive environment parameter exceeds a preset threshold, correcting the comprehensive environment parameter to obtain the environment adaptation parameter.

[0018] Preferably, the process of generating an optimized motion path according to the environment adaptation parameter comprises:

[0019] obtaining real-time environment data, including sea current velocity direction data and obstacle distribution data;

[0020] if the sea current velocity is greater than a preset threshold, generating an initial trajectory point set using an A-star algorithm;

[0021] adjusting the initial trajectory point set according to a dynamic obstacle detection result to obtain an adjusted trajectory point set;

[0022] performing Kalman filter smoothing processing on the adjusted trajectory point set to obtain a smoothed trajectory curve;

[0023] if the smoothed trajectory curve meets the accuracy requirement, converting it into a control instruction to form the optimized motion path.

[0024] Preferably, the process of collecting structure surface detection data along the motion path comprises:

[0025] synchronously obtaining sonar data and optical data along the motion path;

[0026] de-noising and normalizing the sonar data and optical data to obtain a first sensor dataset;

[0027] inputting the first sensor dataset into a convolutional neural network to extract initial surface features to obtain a first feature set;

[0028] if a resolution of the first feature set is lower than a preset threshold, performing interpolation enhancement to obtain a second feature set;

[0029] analyzing structural geometric properties of the second feature set to determine a spatial distribution pattern of surface features to obtain the structural surface detection data.

[0030] Preferably, a process of obtaining a defect distribution map according to the structural surface detection data comprises:

[0031] performing high-resolution imaging on the structural surface detection data to obtain a first image;

[0032] de-noising and enhancing the first image to obtain a second image;

[0033] performing semantic segmentation on the second image using a U-Net algorithm to extract potential defect regions to obtain a defect region set;

[0034] extracting texture and geometric features of the defect region set to generate a feature vector set;

[0035] inputting the feature vector set into a support vector machine to classify and determine a defect type;

[0036] calculating defect spatial coordinates according to the defect region set and the defect type to generate a defect position set;

[0037] generating the defect distribution map according to the defect position set and the defect type.

[0038] Preferably, a process of dividing a high-priority defect region according to the defect distribution map comprises:

[0039] inputting the defect distribution map into a defect classification model to evaluate a severity of a defect to obtain a severity evaluation value;

[0040] if the severity evaluation value is higher than a preset threshold, marking a corresponding region as a high-priority region to obtain a high-priority defect region set;

[0041] performing clustering on the high-priority defect region set using a K-means clustering algorithm to obtain a clustering distribution;

[0042] selecting a sub-region with the highest defect density from each cluster to determine a set of key defect central regions;

[0043] If the defect density of the critical defect cluster area is higher than a preset density threshold, a decision tree algorithm is used to sort the defect features to obtain a defect feature priority ranking;

[0044] According to the defect feature priority ranking, spatial features are extracted, and a spatial pattern of defect distribution is determined, so as to determine the high-priority defect area.

[0045] Preferably, the process of embedding the high-priority defect area data into a carrier signal to obtain steganographic data includes:

[0046] Obtain the coordinate range and priority label of the high-priority defect area to obtain an initial defect data set;

[0047] Embed the initial defect data set into a preset carrier signal using a data steganography algorithm to generate steganographic encoding data;

[0048] If the embedding rate of the steganographic encoding data is lower than a preset threshold, adjust the frequency parameters of the carrier signal and re-embed to obtain optimized steganographic data;

[0049] Perform frequency domain analysis on the optimized steganographic data using discrete cosine transform to extract a steganographic protection feature vector and generate a frequency domain feature set;

[0050] If the frequency domain feature set passes the integrity check, the optimized steganographic data is used as the steganographic data.

[0051] Preferably, the process of end-to-end encryption of the steganographic data to obtain encrypted transmission data includes:

[0052] Decode the steganographic data to recover the original defect data;

[0053] Encrypt the original defect data using an AES algorithm to generate an encryption key and encrypted data;

[0054] Package the encrypted data according to an end-to-end encryption protocol to obtain packaged data;

[0055] Calculate the integrity check value of the packaged data using an SHA-256 algorithm to generate check data;

[0056] If the check data is consistent with a preset integrity threshold, the packaged data is used as the encrypted transmission data.

[0057] Preferably, the process of sending the encrypted transmission data to a cloud server through an underwater communication protocol and verifying its integrity to obtain final lossless detection storage data includes:

[0058] Send the encrypted transmission data to the cloud server through the underwater communication protocol, and obtain transmission data;

[0059] Integrity verification is performed on the transmission data by using a CRC check algorithm, and a check result is obtained.

[0060] If the check result indicates that the data is complete, the transmission data is stored in the cloud server to obtain storage data.

[0061] The hash value of the storage data is calculated using the SHA-256 algorithm to obtain a hash identifier.

[0062] If the hash identifier is consistent with the original data hash value, it is confirmed that the data has not been tampered with, and the final lossless detection storage data is obtained, and the storage time stamp and data index of the lossless detection storage data are recorded to form a storage record.

[0063] The application also provides a marine structure underwater non-destructive testing system, comprising:

[0064] An environmental perception module is used to obtain environmental perception data of the underwater detection device, and environmental adaptation parameters are determined according to the environmental perception data.

[0065] A path planning module is used to generate an optimized motion path according to the environmental adaptation parameters.

[0066] A structure scanning module is used to collect structure surface detection data along the motion path.

[0067] A defect extraction module is used to obtain a defect distribution map according to the structure surface detection data.

[0068] A defect analysis module is used to divide high-priority defect areas according to the defect distribution map.

[0069] A data steganography module is used to embed high-priority defect area data into a carrier signal to obtain steganographic data.

[0070] A data encryption module is used to perform end-to-end encryption on the steganographic data to obtain encrypted transmission data.

[0071] A data transmission module is used to send the encrypted transmission data to the cloud server through the underwater communication protocol, and verify its integrity to obtain the final non-destructive testing storage data.

[0072] Compared with the prior art, the application has the following advantages and technical effects:

[0073] The present application aims at the problems of high current intensity, low visibility, great difficulty in structural defect detection and insufficient data transmission security in deep sea environment, and proposes an integrated solution. Through real-time analysis of sensor data by a deep learning model, the current and visibility are judged, and environment adaptation parameters are generated, and then a path planning algorithm is used to optimize the device navigation trajectory, realizing autonomous navigation in strong current and low visibility. Based on the optimized motion path, the structure surface features are extracted by convolutional neural network using sonar and optical sensor data, and the defect area is accurately located and the defect type is judged by combining image segmentation algorithm. Further, the defect classification model is used to analyze the defect severity, determine the high priority area, and embed the defect information by using data steganography technology, and generate encrypted transmission data by combining end-to-end encryption algorithm, and transmit to the cloud through underwater communication protocol, and verify the data integrity. The present application realizes the cooperative optimization of device autonomous navigation, defect accurate detection and data safe transmission in deep sea environment, and improves the detection efficiency and data security. BRIEF DESCRIPTION OF DRAWINGS

[0074] The accompanying drawings, which form a part of the present application, are intended to provide further understanding of the present application, and the illustrative embodiments of the present application and their description serve the purpose of explaining the present application. The accompanying drawings should not be construed as an inappropriate limitation on the present application. In the drawings:

[0075] Figure 1 The method flowchart of the embodiment of the present application is shown in the figure;

[0076] Figure 2 The system structure diagram of the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0077] It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0078] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0079] As shown in the figure, Figure 1 The present embodiment provides an underwater nondestructive testing method for offshore structures, comprising:

[0080] Obtain the environment perception data of the underwater detection device, and determine the environment adaptation parameters according to the environment perception data;

[0081] Generate an optimized motion path according to the environment adaptation parameters, and collect structure surface detection data along the motion path;

[0082] According to the structure surface detection data, a defect distribution map is obtained;

[0083] According to the defect distribution map, a high-priority defect area is divided;

[0084] The high-priority defect area data is embedded into the carrier signal to obtain steganographic data;

[0085] The steganographic data is end-to-end encrypted to obtain encrypted transmission data;

[0086] The encrypted transmission data is sent to the cloud server through the underwater communication protocol, and its integrity is verified to obtain the final lossless detection storage data.

[0087] Further, the environmental perception data of the underwater detection device is obtained, and according to the environmental perception data, the process of determining the environmental adaptation parameter includes:

[0088] The sensor raw data of the underwater detection device is obtained, and the sensor raw data is denoised and standardized to obtain preprocessed data;

[0089] The preprocessed data is input into a deep learning model to extract an environmental feature vector;

[0090] According to the environmental feature vector, the current intensity parameter and the visibility parameter are calculated respectively;

[0091] The current intensity parameter and the visibility parameter are weighted and fused to obtain a comprehensive environmental parameter;

[0092] If the comprehensive environmental parameter exceeds a preset threshold, the comprehensive environmental parameter is corrected to obtain an environmental adaptation parameter.

[0093] Further, the embodiment obtains sensor data from the underwater detection device, uses preprocessing technology to denoise and standardize the data, and obtains processed sensor data. Through a preset deep learning model, the processed sensor data is feature-extracted to obtain a feature vector of the deep sea environment. A convolutional neural network is used to classify the feature vector to determine the current intensity level and obtain a current intensity parameter. A recurrent neural network is used to perform time series analysis on the feature vector to determine the change trend of the visibility condition and obtain a visibility parameter. According to the current intensity parameter and the visibility parameter, a weighted fusion algorithm is used to calculate an environmental adaptation parameter to obtain a comprehensive environmental parameter. If the comprehensive environmental parameter exceeds a preset threshold, the sensor data is analyzed again to determine an environmental abnormal point to obtain a corrected environmental parameter. Through the corrected environmental parameter, a mapping function is used to generate a real-time environmental adaptation strategy to obtain a final environmental adaptation output.

[0094] Specifically, obtaining sensor data from the underwater detection device is the basis of deep sea environment monitoring.

[0095] For example, underwater sonar and pressure sensors can collect data on sea current velocity and water depth, assuming that a device records a sea current velocity of 0.5 meters per second and a water pressure of 1000 Pa per second.

[0096] Preprocessing techniques are crucial for data denoising and normalization. In one possible implementation, the denoising of the present embodiment employs a moving average filter, with a window size of 5 seconds for sea current velocity data, smoothing out abnormal fluctuations, such as sudden 1.2-meter-per-second data to 0.6 meters per second. Normalization normalizes the data to the interval of 0 to 1, facilitating subsequent model processing. This can improve data quality and ensure the accuracy of feature extraction. Feature extraction is performed through a pre-set deep learning model to obtain deep-sea environmental feature vectors.

[0097] For example, an autoencoder can extract key features such as flow field gradients and turbulence intensity from data such as sea current velocity, water pressure, and temperature, generating a 32-dimensional feature vector. This method can capture complex environmental patterns and improve the accuracy of subsequent classification and analysis. A convolutional neural network is used to classify the feature vector to determine the sea current intensity level.

[0098] In one embodiment, the convolutional neural network involved in the present embodiment includes three convolutional layers and two fully connected layers, classifying the feature vector into weak, medium, and strong flow levels. Assuming that a certain feature vector is classified as strong flow, the sea current intensity parameter is quantified as 0.8. This classification can provide environmental basis for device operation and optimize decision-making efficiency. A recurrent neural network is used for time series analysis to determine the visibility trend.

[0099] For example, based on water turbidity and light intensity data, the recurrent neural network analyzes the past 10 minutes of data to predict a 5-minute future visibility decline trend, outputting a visibility parameter of 0.3. This prediction helps to adjust device strategies in advance and reduce misjudgments caused by changes in visibility. A weighted fusion algorithm combines sea current intensity and visibility parameters to calculate a comprehensive environmental parameter.

[0100] In one possible implementation, sea current intensity is given a weight of 0.6 and visibility a weight of 0.4, and the comprehensive parameter is calculated as 0.62. If the threshold is set to 0.7, no secondary analysis is needed because the threshold has not been exceeded. This fusion method balances multiple factors and improves the comprehensiveness of environmental assessment. If the comprehensive parameter exceeds 0.7, secondary analysis is performed to determine environmental abnormal points.

[0101] For example, if the sea current velocity is detected to have suddenly increased to 2 meters per second, secondary analysis can locate the abnormal source to a vortex near the device, and the comprehensive parameter is corrected to 0.65. Real-time environmental adaptation strategies are generated through a mapping function, such as adjusting the device's posture to reduce resistance, and the final output strategy is "reduce propulsion power by 20%". This strategy can reduce device wear and tear and prolong its service life.

[0102] It should be noted that the mapping function can be trained based on historical data to ensure that the strategy dynamically matches the environment, thereby improving the robustness and adaptability of the system.

[0103] Further, according to the environment adaptation parameter, the process of generating the optimized motion path comprises:

[0104] Obtain real-time environment data, including sea current velocity direction data and obstacle distribution data;

[0105] If the sea current velocity is greater than a preset threshold, an A-star algorithm is used to generate an initial trajectory point set;

[0106] According to the dynamic obstacle detection result, the initial trajectory point set is adjusted to obtain an adjusted trajectory point set;

[0107] The adjusted trajectory point set is subjected to Kalman filter smoothing processing to obtain a smoothed trajectory curve;

[0108] If the smoothed trajectory curve meets the accuracy requirement, it is converted into a control instruction to form an optimized motion path.

[0109] Further, the embodiment obtains real-time environment data through an environment perception module, including strong sea current velocity direction and obstacle distribution under low visibility conditions, to obtain an environment parameter set. If the strong sea current velocity in the environment parameter set exceeds a preset threshold, an A-star algorithm is used to generate an initial navigation trajectory, and path points are calculated in combination with real-time environment data to determine an initial trajectory point set. According to the initial trajectory point set, a dynamic obstacle detection technology is used to identify the position of dynamic obstacles under strong sea current and low visibility conditions to obtain an obstacle distribution map. If the obstacle distribution map shows that there is an intersection between the dynamic obstacles and the initial trajectory point set, a trajectory adjustment strategy is used to recalculate the path points based on the A algorithm to generate an adjusted trajectory point set. Through an environment adaptation parameter, in combination with the adjusted trajectory point set, a Kalman filter algorithm is used to smooth the device navigation trajectory to obtain a smoothed trajectory curve. If the navigation accuracy of the smoothed trajectory curve meets a preset threshold, the smoothed trajectory curve is converted into a device control instruction to determine an optimal motion trajectory. The device control instruction drives the navigation system to update the device position in real time, obtains new real-time environment data, and loops to the first step.

[0110] For example, the process of the embodiment environment perception module acquiring real-time environment data is realized based on multi-sensor fusion technology. In the deep sea environment, the strong current speed and direction are measured by a Doppler current meter, and the obstacle distribution relies on the cooperative work of a sonar and a laser radar. Assuming that in a certain deep sea exploration task, the current speed measured by the current meter is 2.5 m / s, the direction is northwest, the sonar scans a dense obstacle group 50 meters in front, and the visibility is only 0.5 meters. After the environment parameter set integrates these data, it is judged that the current speed exceeds the preset threshold of 1.8 m / s, triggering the subsequent navigation calculation.

[0111] In a possible implementation, when the A-star algorithm generates an initial navigation trajectory, the current direction and obstacle distribution are given priority. Based on the environment data, the algorithm takes the current position of the device as the starting point and a sampling point 100 meters away as the target point to generate a path that avoids obstacles. The initial trajectory point set may contain 10 key points, and the distance between each point is about 10 meters. The dynamic obstacle detection technology then intervenes, and through continuous scanning by the sonar, it identifies moving obstacles, such as a group of deep-sea organisms passing through the trajectory area at a speed of 0.3 m / s. The obstacle distribution map shows that these dynamic obstacles intersect with the 4th trajectory point.

[0112] Specifically, the trajectory adjustment strategy re-invokes the A-star algorithm, combines the predicted positions of the dynamic obstacles, adjusts the 4th to 6th trajectory points, and generates a new path to bypass the obstacle group. The new trajectory point set may offset the path by 5 meters to ensure the safe passage of the device. The Kalman filter algorithm then smoothes the adjusted trajectory point set.

[0113] For example, the device may deviate in position under the influence of strong currents, and the filter algorithm corrects the trajectory deviation by fusing inertial navigation and sonar data, so that the navigation accuracy of the trajectory curve is within 0.1 meters, meeting the preset threshold.

[0114] In a possible implementation, when the smoothed trajectory curve is converted into device control instructions, specific speed and direction instructions are generated.

[0115] For example, the device needs to move along the new trajectory at a speed of 1.2 m / s and adjust the heading every 5 seconds. The navigation system drives the device according to the instructions, and after updating the position in real time, the environment perception module acquires data again, such as the current speed decreasing to 1.5 m / s and the visibility increasing to 1 meter, and the cycle is processed to optimize subsequent navigation.

[0116] It should be noted that this cycle mechanism ensures that the device continuously adapts to dynamic changes in the complex deep sea environment, improving the stability and safety of navigation.

[0117] Further, the process of collecting structure surface detection data along the motion path includes:

[0118] synchronously acquiring sonar data and optical data along a motion path;

[0119] denoising and normalizing the sonar data and the optical data to obtain a first sensor data set;

[0120] inputting the first sensor data set into a convolutional neural network to extract initial surface features to obtain a first feature set;

[0121] if a resolution of the first feature set is lower than a preset threshold, performing interpolation enhancement to obtain a second feature set;

[0122] analyzing structural geometric characteristics of the second feature set to determine a spatial distribution pattern of the surface features to obtain structural surface detection data.

[0123] Further, the embodiment acquires sonar sensor data and optical sensor data, adopts a preprocessing technique to denoise and normalize the data to obtain a first sensor data set. Through the first sensor data set, a convolutional neural network is adopted to scan a complex structure, extract initial surface features, and obtain a first feature set. If the resolution of the first feature set is lower than a preset threshold, the first feature set is enhanced, an interpolation algorithm is adopted to improve the feature resolution, and a second feature set is obtained. According to the second feature set, geometric characteristics of the complex structure are analyzed, a spatial distribution pattern of the surface features is determined, and geometric distribution data is obtained. Through the geometric distribution data, a path planning algorithm is adopted to optimize a motion path, the connectivity between path points is determined, and an optimized path data set is obtained. If there is a conflict between the path points in the optimized path data set, the position of the path points is adjusted through a constraint condition, and a final path data set is obtained. According to the final path data set, control instructions are generated to drive the equipment to move along the optimized path, and a scanning task of the complex structure is completed.

[0124] In an embodiment, the sonar sensor and optical sensor data preprocessing technique mainly aims at noise interference problems in seabed complex structure scanning. Sonar data usually contains multipath reflection noise and equipment vibration noise, and the embodiment adopts a combination of median filtering and Gaussian filtering for denoising processing. The image data acquired by the optical sensor in a low-visibility marine environment has low contrast and color distortion problems, and data standardization is achieved through histogram equalization and gamma correction, which unifies the data of different sensors to the same numerical range, facilitating subsequent fusion processing.

[0125] For example, the present embodiment utilizes a convolutional neural network to play a key role in complex seabed structure feature extraction. The network adopts a multi-scale convolution kernel design, containing three different sizes of convolution kernels, 3x3, 5x5 and 7x7, to extract detailed texture, medium-scale shape and large-scale contour features respectively. For the scanning task of seabed rock structure, the network can identify key features such as crack distribution, concave-convex fluctuations and material changes on the rock surface, forming an initial feature set containing spatial position information and feature attributes.

[0126] In one possible implementation, the feature resolution enhancement process involved in the present embodiment adopts a bicubic interpolation algorithm and a super-resolution reconstruction technique. When the resolution of the initial feature set is 512x512 pixels, if the preset threshold requires 1024x1024 pixels, the system automatically starts the enhancement process. The bicubic interpolation algorithm generates new pixel points by calculating the weighted average of the surrounding 16 pixel points, maintaining the smoothness of the feature edges. The super-resolution reconstruction technique uses a deep learning model to predict high-frequency detail information to supplement the texture features lost in the interpolation process.

[0127] For example, in the geometric property analysis process, when the system scans the seabed pipeline structure, it identifies the bending degree and connection node position of the pipeline by calculating the surface normal vector and curvature change. The geometric distribution data contains the three-dimensional coordinates, normal vector direction and local curvature value of each feature point, forming a complete spatial geometric description. These data provide accurate spatial constraints for subsequent path planning.

[0128] In one embodiment, the path planning algorithm involved in the present embodiment adopts an improved RRT algorithm to handle navigation problems in complex seabed environments. The algorithm first randomly samples path points in three-dimensional space, and then checks whether there are obstacles between adjacent path points. For seabed canyon terrain, the algorithm avoids steep rock wall areas and selects relatively flat paths for scanning. Path point connectivity judgment is based on two dimensions of line-of-sight detection and collision detection to ensure that the device can safely pass through each path segment.

[0129] For example, the present embodiment solves the problem of path point overlap and deadlock through the path conflict processing mechanism through constraint condition optimization. When the distance between two path points is detected to be less than the minimum turning radius of the device, the system automatically adjusts the position of one of the path points to maintain a safety distance of more than 0.5 meters. The constraint conditions include kinematic constraints of the device, environmental obstacle constraints and task coverage requirements, and the multi-objective optimization algorithm is used to find the optimal path configuration that satisfies all the constraints.

[0130] It should be noted that the control instruction generation process of the embodiment converts the path data into motion commands executable by the device. The instructions contain parameters such as target position coordinates, motion speed, and attitude angle, and a piecewise linear interpolation method is used to generate a smooth motion trajectory, avoiding sharp changes in direction that can affect scanning accuracy.

[0131] Further, according to the structure surface detection data, the process of obtaining the defect distribution map includes:

[0132] High-resolution imaging is performed on the structure surface detection data to obtain a first image;

[0133] The first image is denoised and enhanced to obtain a second image;

[0134] The U-Net algorithm is used to perform semantic segmentation on the second image to extract potential defect regions and obtain a defect region set;

[0135] Texture and geometric features of the defect region set are extracted to generate a feature vector set;

[0136] The feature vector set is input into a support vector machine to classify and determine the defect type;

[0137] According to the defect region set and the defect type, the spatial coordinates of the defects are calculated to generate a defect position set;

[0138] According to the defect position set and the defect type, a defect distribution map is generated.

[0139] Further, the embodiment obtains original feature data from the structure surface and generates a first image using a high-resolution imaging device. The first image is denoised and enhanced using preprocessing techniques to obtain a second image. The U-Net algorithm is used to perform image segmentation on the second image to extract potential defect regions and obtain a defect region set. If the defect region set is not empty, texture and geometric features are extracted for each defect region to generate a feature vector set. According to the feature vector set, a support vector machine algorithm is used for classification to determine the defect type. The spatial coordinates of each defect are calculated using the defect region set and the defect type to generate a defect position set. According to the defect position set and the defect type, a visualization tool is used to generate a defect distribution map.

[0140] In one embodiment, the high-resolution imaging device uses a laser scanner or a high-precision camera that can obtain surface image data with a resolution of 0.1 millimeters. When the device scans the surface of a metal pipe, the original image often contains environmental light interference, device vibration noise, and other factors, which need to be preprocessed through Gaussian filtering and histogram equalization techniques to increase the image contrast to more than 1.5 times the original data.

[0141] Specifically, the U-Net algorithm adopts an encoder-decoder architecture in the image segmentation process, which can accurately identify defect areas such as cracks, corrosion, and dents. The algorithm first divides the input image into small blocks of 256x256 pixels, extracts multi-scale features through convolution layers, and then restores the original resolution through upsampling. When a suspected defect area is detected, the system automatically marks the pixel coordinate range to form a defect area set containing position information.

[0142] In one possible implementation, the feature extraction process calculates parameters such as texture roughness, edge sharpness, and shape regularity for each defect area.

[0143] For example, crack defects usually exhibit high edge sharpness and low shape regularity, while corrosion areas exhibit high texture roughness. The system combines these parameters into a 128-dimensional feature vector, providing a data basis for subsequent classification.

[0144] For example, the support vector machine classifier learns the feature boundaries of different defect types through training samples, and can accurately classify defects into three categories: surface scratches, deep cracks, and point corrosion. The classifier uses a radial basis function kernel, and the classification accuracy can reach more than 95%. When a new feature vector is input, the algorithm calculates its distance from each class support vector, and selects the class with the smallest distance as the determination result.

[0145] It should be noted that spatial coordinate calculation needs to be combined with the actual position information of the device and the image pixel coordinates. The system converts two-dimensional image coordinates to three-dimensional space coordinates through a coordinate transformation matrix, with an accuracy of millimeters. The defect position set contains the X, Y, Z coordinate values of each defect and the corresponding defect type identifier.

[0146] In one embodiment, the visualization tool displays the defect distribution using a combination of heat maps and three-dimensional models. Different defect types are identified by different colors, and the severity of the defect is represented by the color depth. This visualization method can intuitively reflect the overall health status of the structure and provide an important basis for maintenance decisions. Through this complete defect detection process, automatic identification and precise positioning of surface defects on complex structures can be achieved.

[0147] Further, according to the defect distribution map, the process of dividing high-priority defect areas includes:

[0148] Input the defect distribution map into the defect classification model to evaluate the severity of the defect and obtain a severity evaluation value;

[0149] If the severity evaluation value is higher than a preset threshold, the corresponding area is marked as a high-priority area, and a high-priority defect area set is obtained;

[0150] The K-means clustering algorithm is used to cluster the high-priority defect region set to obtain a clustering distribution.

[0151] A sub-region with the highest defect density is selected from each cluster to determine a key defect concentrated region.

[0152] If the defect density of the key defect concentrated region is higher than a preset density threshold, a decision tree algorithm is used to sort the defect features to obtain a defect feature priority ranking.

[0153] According to the defect feature priority ranking, spatial features are extracted to determine the spatial pattern of defect distribution, thereby determining the high-priority defect region.

[0154] Further, the embodiment obtains defect data input from the defect distribution map by a preset defect classification model, classifies by using a support vector machine algorithm to obtain a defect severity assessment. If the defect severity assessment value is higher than a preset threshold, the corresponding region is marked as a high-priority region to determine a high-priority defect region set. According to the high-priority defect region set, the K-means clustering algorithm is used to group the regions to obtain a clustering distribution of the defect regions. Through the clustering distribution, a sub-region with the highest defect density is obtained from each cluster to determine a key defect concentrated region. If the defect density of the key defect concentrated region is higher than a preset density threshold, a decision tree algorithm is used to analyze the defect features to obtain a defect feature priority ranking. According to the defect feature priority ranking, spatial features of the defect distribution are obtained from the key defect concentrated region to determine the spatial pattern of the defect distribution. Through the spatial pattern of the defect distribution, an interpolation algorithm is used to generate a continuity prediction map of the defect distribution to obtain an extended distribution of the high-priority defect region.

[0155] In an embodiment, the preset defect classification model establishes a defect severity discrimination standard through training samples.

[0156] Specifically, the model divides defects into three levels of slight, moderate, and severe according to parameters such as depth, area, and shape complexity, and the evaluation value corresponding to the severe level is usually set to 0.8 or above. When the evaluation value of the concrete surface crack detected reaches 0.85, the system automatically marks the region as a high-priority region to form a defect region set that needs to be focused on.

[0157] For example, in the bridge surface detection scenario, the K-means clustering algorithm groups the high-priority defect regions according to the spatial proximity. The algorithm sets the number of clusters to 5, and classifies the crack defects scattered in different positions of the bridge deck into the vicinity of the corresponding cluster center. Each cluster represents a relatively concentrated defect distribution region, which facilitates subsequent focused analysis and processing.

[0158] It should be noted that the defect density calculation of the embodiment adopts a comprehensive index of the number of defects in a unit area and the total area of defects. When the defect density of a certain clustering area exceeds the preset 15 defects per square meter, the system identifies the area as a key defect concentration area. The setting of this density threshold is based on the balance between structural safety requirements and maintenance costs.

[0159] In one embodiment, the embodiment performs multi-dimensional analysis on defect features based on a decision tree algorithm, including geometric features such as length, width, depth, and direction angle of defects, and apparent features such as surface roughness and color difference. The algorithm generates a priority ranking list according to the influence of each feature on structural safety. Generally, crack depth and length are given the highest priority weight because these features are directly related to the load-bearing capacity of the structure.

[0160] For example, in the spatial pattern recognition process, the system analyzes the distribution law of defects in the key defect concentration area. When it is found that the defects show linear distribution characteristics, it indicates that there may be a structural stress concentration problem. When the defects show radial distribution, it may cause local material aging or external impact damage. The identification of this spatial pattern provides an important basis for the subsequent development of maintenance strategies.

[0161] It should be noted that the interpolation algorithm involved in the embodiment adopts the Kriging interpolation method, which predicts the defect development trend in the surrounding area based on the distribution characteristics of the known defect points. The algorithm considers the directionality and continuity of defect expansion, and generates a continuous prediction map covering the entire key area. This prediction can identify the potential defect expansion path and provide guidance for preventive maintenance.

[0162] In one embodiment, the expansion distribution map displays the possible development range of the high-priority defect area. By comparing the current defect distribution with the predicted expansion distribution, maintenance personnel can develop targeted treatment plans in advance to avoid structural safety risks caused by further deterioration of defects.

[0163] Further, the process of embedding high-priority defect area data into a carrier signal to obtain steganographic data includes:

[0164] Obtain the coordinate range and priority label of the high-priority defect area to obtain an initial defect data set;

[0165] Embed the initial defect data set into the preset carrier signal using a data steganography algorithm to generate steganographic encoding data;

[0166] If the embedding rate of the steganographic encoding data is lower than the preset threshold, adjust the frequency parameters of the carrier signal and re-embed to obtain optimized steganographic data;

[0167] The optimized stego data is subjected to frequency domain analysis by using discrete cosine transform to extract a stego protection feature vector and generate a frequency domain feature set.

[0168] If the frequency domain feature set passes the integrity check, the optimized stego data is taken as the stego data.

[0169] Further, the embodiment obtains high-priority defect region data from the data source, determines the coordinate range and priority label of the defect region, and obtains an initial defect data set. The initial defect data set is subjected to encoding processing by using a data steganography algorithm, and the defect information is embedded in a preset carrier signal to generate stego encoded data. If the embedding rate of the stego encoded data is lower than a preset threshold, the frequency parameter of the carrier signal is adjusted, and the embedding operation is re-executed to obtain optimized stego data. The optimized stego data is subjected to frequency domain analysis by using a discrete cosine transform algorithm to extract a stego protection feature vector and generate a frequency domain feature set. According to the frequency domain feature set, the integrity of the stego protection is determined, and if the integrity check passes, detection data of the stego protection is generated. The coordinate and priority information of the defect region are extracted from the detection data of the stego protection to generate a structured detection data set. The structured detection data set is subjected to classification processing by using a support vector machine algorithm to determine a final defect classification result.

[0170] In an embodiment, the core principle of the data steganography algorithm is to embed the coordinate information and priority label of the defect region into the redundant space of the carrier signal.

[0171] Specifically, when the high-priority defect region data containing the X coordinate range 120-450 and the Y coordinate range 80-320 is obtained, the system first converts these numerical values into a binary encoding sequence, and then selects the low-frequency component of the audio carrier signal as the embedding carrier.

[0172] For example, the frequency parameter of the carrier signal is initially set to 8 kHz sampling rate, and when the embedding rate detection result is 0.15 which is lower than the preset threshold 0.2, the system automatically adjusts the carrier signal frequency to 16 kHz to increase the available embedding space. This dynamic adjustment mechanism ensures that the defect information can be completely embedded in the carrier, avoiding the risk of data loss.

[0173] In a possible implementation, when the discrete cosine transform algorithm performs frequency domain analysis on the optimized stego data, the time domain signal is converted into a frequency domain coefficient matrix. The system extracts the first 64 coefficients in the low-frequency region as the feature vector, and these coefficients contain the main features of the stego information. By analyzing the distribution pattern and amplitude change of the coefficients, the integrity of the stego protection can be effectively detected.

[0174] For example, when the energy concentration index in the frequency domain feature set reaches 0.85 or above, it indicates that the steganographic protection integrity is good. The system further verifies the data integrity through the hash check algorithm, and generates detection data containing the original defect coordinates and priority labels. This double verification mechanism significantly improves the reliability of defect information transmission.

[0175] In one embodiment, the generation process of the structured detection data set of the present embodiment involves reorganizing the steganographically protected detection data according to a predetermined format. The system sorts the extracted defect region coordinates according to the priority level, forming a structured data table containing position information, severity score and time stamp.

[0176] Specifically, the communication support vector machine algorithm uses radial basis function as the kernel function when processing structured detection data, and performs nonlinear classification of defect data in a multi-dimensional feature space. The algorithm learns the boundary features of different defect types through training samples and establishes a classification decision surface. When new defect data is input, the system determines the final classification result of the defect according to the distance relationship between the data point and the decision surface.

[0177] It should be noted that this defect detection method based on steganographic protection has significant security advantages. By hiding sensitive defect information in ordinary carrier signals, it effectively prevents defect data from being maliciously intercepted or tampered with during transmission, while ensuring the accuracy and traceability of defect classification results.

[0178] Further, the process of end-to-end encryption of steganographic data to obtain encrypted transmission data includes:

[0179] Decoding the steganographic data to recover the original defect data;

[0180] Encrypting the original defect data using the AES algorithm to generate an encryption key and encrypted data;

[0181] According to the end-to-end encryption protocol, the encrypted data is encapsulated to obtain encapsulated data;

[0182] The SHA-256 algorithm is used to calculate the integrity check value of the encapsulated data to generate check data;

[0183] If the check data is consistent with the preset integrity threshold, the encapsulated data is taken as the encrypted transmission data.

[0184] Furthermore, this embodiment extracts the original data stream from the steganography detection data and restores it to the initial data using a preset decoding method. Using the initial data, an encryption key and encrypted data are generated using the AES encryption algorithm. If the encryption key is successfully generated, the encrypted data is encapsulated according to an end-to-end encryption protocol to obtain encapsulated data. For the encapsulated data, an integrity check value is calculated using the SHA-256 algorithm to generate check data. If the check data matches a preset integrity threshold, the check data is sent via a secure transmission protocol to obtain transmitted data. Based on the transmitted data, a decryption algorithm and encryption key are used for decryption processing to generate decrypted data. Using the decrypted data, the data integrity and steganography protection status are verified, and the final output data is determined.

[0185] In one possible implementation, when extracting the original data stream from the steganalytic detection data, the system first identifies the frequency domain marker bits in the carrier signal and locates the embedded defect information by inverse discrete cosine transform.

[0186] Specifically, when a marker signal with a frequency of 2.4 kHz is detected, the decoder automatically switches to the corresponding extraction mode, recombines the defect coordinate data scattered in 8 frequency bands, and restores the defect area information containing 127 pixels on the X-axis and 89 pixels on the Y-axis.

[0187] For example, during the AES encryption algorithm generation process, the system dynamically selects the key length based on the length of the extracted defect data. When the defect data exceeds 1024 bytes, a 256-bit key length is automatically adopted to ensure the security of high-priority defect information. The encryption key is generated by combining a timestamp and a device identifier to form a unique identifier. This method effectively prevents defect detection data from being maliciously tampered with or stolen during transmission.

[0188] In one embodiment, the end-to-end encryption protocol employs a layered protection mechanism during encapsulation. The system categorizes encrypted defective data into three priority levels: high-priority data uses double encryption encapsulation, medium-priority data uses standard encapsulation, and low-priority data uses lightweight processing. Timestamps and sequence numbers are added during encapsulation to ensure the integrity and timing of data packets.

[0189] Specifically, when calculating the integrity check value using the SHA-256 algorithm, the system generates a 64-bit hexadecimal checksum for each data block. When the defect detection data contains information from multiple regions, the checksum for each region is calculated separately, and then the overall checksum is generated.

[0190] For example, a data packet containing 5 defective areas will generate 6 check values. The first 5 correspond to each area, and the 6th is the overall check code.

[0191] For example, during the transmission of the secure transmission protocol, the system adopts an adaptive transmission strategy. When the network delay is less than 50 milliseconds, the real-time transmission mode is adopted; when the delay is more than 100 milliseconds, the cache transmission mode is switched to, and the check data is temporarily stored and sent in batches. The data integrity is continuously monitored during the transmission, and once the check value is found to be inconsistent, the retransmission mechanism is triggered immediately.

[0192] In one possible implementation, after the decryption processing stage system verifies the validity of the key, the data is restored according to the reverse encryption process. The decryption process adopts a segmented processing method, and each time 256 bytes of data block is processed to ensure stable decryption of large-capacity defect data. After decryption, the system automatically compares the original data characteristic value to verify the decryption accuracy.

[0193] For example, during the data integrity verification, the system checks the steganographic protection state identifier to confirm whether the defect information is completely preserved. By comparing the data structure before and after decryption, the consistency of the coordinate information, priority label and other key parameters is verified. The final output data contains complete defect detection results, providing a reliable basis for subsequent quality control. This multi-level security protection mechanism can significantly improve the security and reliability of defect detection data.

[0194] Further, the process of obtaining the final lossless detection storage data by transmitting the encrypted transmission data to the cloud server through the underwater communication protocol and verifying the integrity thereof includes:

[0195] The encrypted transmission data is transmitted to the cloud server through the underwater communication protocol to obtain the transmission data.

[0196] The transmission data is verified for integrity by using the CRC check algorithm to obtain a check result.

[0197] If the check result indicates that the data is complete, the transmission data is stored in the cloud server to obtain storage data.

[0198] The hash value of the storage data is calculated by using the SHA-256 algorithm to obtain a hash identifier.

[0199] If the hash identifier is consistent with the original data hash value, it is confirmed that the data has not been tampered with, the final lossless detection storage data is obtained, and the storage time stamp and data index of the lossless detection storage data are recorded to form a storage record.

[0200] Further, the embodiment obtains original data through an underwater communication protocol, encrypts the data by using an AES encryption algorithm to obtain encrypted data. If the encrypted data is successfully generated, the encrypted data is transmitted to a cloud server through the underwater communication protocol to obtain transmission data. According to the transmission data, a CRC check algorithm is used to verify the data integrity to obtain a check result. If the check result shows that the data is complete, the data is stored on the cloud server to obtain stored data. According to the stored data, a SHA-256 algorithm is used to generate a data hash value to obtain a hash identifier. By comparing the hash identifier with the original data hash value, if they are consistent, it is confirmed that the data has not been tampered with, and the final stored data is obtained. According to the final stored data, a storage timestamp and a data index are recorded to obtain a storage record.

[0201] In a possible implementation, the underwater communication protocol uses acoustic modulation technology to realize data transmission, and carries the original data through a 12-kilohertz acoustic signal as a carrier frequency. When the underwater sensor collects the marine environment monitoring data, the system first preprocesses the original data packet of 128 bytes to ensure that the data format meets the subsequent encryption requirements.

[0202] Specifically, the AES encryption algorithm uses a 256-bit key length to perform symmetric encryption on the data. The system generates a random initialization vector with a length of 16 bytes, which is combined with the original data and then input into the AES encryption engine. During the encryption process, the algorithm divides the data into 128-bit data blocks, and each data block is subjected to 10 rounds of encryption operation to generate the corresponding ciphertext block. After the encryption is completed, the system verifies whether the length of the ciphertext is consistent with the expected value to confirm that the encrypted data is successfully generated.

[0203] In an embodiment, the underwater communication protocol establishes a secure connection channel with the cloud server and uses frequency division multiple access technology to avoid signal interference. During the transmission process, the system encapsulates the encrypted data into a standard data frame, each frame containing 64 bytes of payload and 8 bytes of control information. The communication protocol automatically processes the signal attenuation and multipath propagation problems in the underwater environment to ensure that the data is complete and reaches the cloud server.

[0204] Specifically, the CRC check algorithm uses a 32-bit cyclic redundancy check code to verify the integrity of the transmission data. The system calculates the CRC value of the received data frame and compares the calculation result with the check field in the data frame. When the two CRC values are completely matched, the check result shows that no error or loss has occurred in the data transmission process. If the check fails, the system automatically triggers the retransmission mechanism to require the sending end to retransmit the data frame.

[0205] In one possible implementation, the cloud server adopts a distributed storage architecture to save the data that passes the verification. The system assigns a unique storage identifier to each data packet and writes the data to multiple storage nodes to achieve redundant backup. After the storage operation is completed, the system returns storage confirmation information, including the physical storage location and access path of the data.

[0206] Specifically, the SHA-256 algorithm performs a hash operation on the stored data to generate a 64-bit hexadecimal hash value. The algorithm processes the stored data in groups of 512 bits, and each group of data undergoes 64 rounds of compression function operation to generate an intermediate hash value. The final hash identifier is unique and irreversible, and can accurately reflect the complete state of the data content.

[0207] In one embodiment, the system compares the newly generated hash identifier with the original hash value saved before data transmission bit by bit. The comparison process uses the XOR operation to detect the difference between the two hash values, and when the operation result is all zeros, it confirms that the data has not been tampered with. This verification mechanism can detect any small changes in the data, providing reliable protection for data integrity.

[0208] Specifically, the storage record contains a time stamp accurate to milliseconds and a hierarchical data index structure. The time stamp uses the UTC standard time format to ensure time consistency worldwide. The data index uses a B-tree structure to organize, supporting fast data retrieval and range query operations, significantly improving data access efficiency.

[0209] As shown in Figure 2 Based on the same inventive concept, the embodiment also provides a marine structure underwater non-destructive testing system,

[0210] An environment perception module is configured to acquire environment perception data of the underwater detection device and determine environment adaptation parameters based on the environment perception data.

[0211] A path planning module is configured to generate an optimized motion path based on the environment adaptation parameters.

[0212] A structure scanning module is configured to collect structure surface detection data along the motion path.

[0213] A defect extraction module is configured to obtain a defect distribution map based on the structure surface detection data.

[0214] A defect analysis module is configured to divide a high-priority defect area based on the defect distribution map.

[0215] A data steganography module is configured to embed data of the high-priority defect area into a carrier signal to obtain steganographic data.

[0216] A data encryption module is configured to perform end-to-end encryption on the steganographic data to obtain encrypted transmission data.

[0217] The data transmission module is configured to send the encrypted transmission data to a cloud server via an underwater communication protocol and verify the integrity of the transmission data to obtain final nondestructive testing storage data.

[0218] The underwater nondestructive testing system for offshore structures provided in the embodiment has all the advantages of the method for underwater nondestructive testing of offshore structures provided in the first embodiment.

[0219] The above merely provides the preferred embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of non-destructive testing of a marine structure underwater, characterised in that, The method comprises the following steps: obtaining environmental perception data of the underwater detection equipment, and determining environmental adaptation parameters according to the environmental perception data; generating an optimized motion path according to the environmental adaptation parameters, and collecting structure surface detection data along the motion path; obtaining a defect distribution map according to the structure surface detection data; dividing a high-priority defect area according to the defect distribution map; embedding the high-priority defect area data into a carrier signal to obtain steganographic data; performing end-to-end encryption on the steganographic data to obtain encrypted transmission data; sending the encrypted transmission data to a cloud server through an underwater communication protocol, and verifying the integrity to obtain final nondestructive testing storage data.

2. The method of claim 1, wherein the process of obtaining environmental perception data of the underwater detection equipment and determining environmental adaptation parameters according to the environmental perception data comprises: obtaining sensor raw data of the underwater detection equipment, denoising and standardizing the sensor raw data to obtain preprocessed data; inputting the preprocessed data into a deep learning model to extract environmental feature vectors; calculating current intensity parameters and visibility parameters according to the environmental feature vectors, respectively; weighting and fusing the current intensity parameters and the visibility parameters to obtain comprehensive environmental parameters; if the comprehensive environmental parameters exceed a preset threshold, correcting the comprehensive environmental parameters to obtain the environmental adaptation parameters.

3. The method of claim 1, wherein the process of generating an optimized motion path according to the environmental adaptation parameters comprises: obtaining real-time environmental data, including current velocity direction data and obstacle distribution data; if the current velocity is greater than a preset threshold, generating an initial trajectory point set using A-star algorithm; adjusting the initial trajectory point set according to dynamic obstacle detection results to obtain an adjusted trajectory point set; performing Kalman filter smoothing processing on the adjusted trajectory point set to obtain a smoothed trajectory curve; if the smoothed trajectory curve meets the accuracy requirement, converting it into control instructions to form the optimized motion path.

4. The method of claim 1, wherein the process of collecting structure surface detection data along the motion path comprises: synchronously obtaining sonar data and optical data along the motion path; denoising and standardizing the sonar data and optical data to obtain a first sensor data set; inputting the first sensor data set into a convolutional neural network to extract initial surface features and obtain a first feature set; if the resolution of the first feature set is lower than a preset threshold, performing interpolation enhancement to obtain a second feature set; analyzing the structure geometric properties of the second feature set to determine the spatial distribution pattern of the surface features, and obtaining the structure surface detection data.

5. The method of claim 1, wherein the process of obtaining a defect distribution map according to the structure surface detection data comprises: performing high-resolution imaging on the structure surface detection data to obtain a first image; denoising and enhancing the first image to obtain a second image; ​ ​ ​ ​ adopting a U-Net algorithm to perform semantic segmentation on the second image, extracting a potential defect region, and obtaining a defect region set; extracting texture and geometric features of the defect region set to generate a feature vector set; inputting the feature vector set into a support vector machine to classify and determine a defect type; calculating defect spatial coordinates according to the defect region set and the defect type, and generating a defect position set; generating the defect distribution map according to the defect position set and the defect type.

6. The method of claim 1, wherein the process of dividing a high-priority defect region according to the defect distribution map comprises: inputting the defect distribution map into a defect classification model to evaluate defect severity and obtain a severity evaluation value; if the severity evaluation value is higher than a preset threshold, marking the corresponding region as a high-priority region to obtain a high-priority defect region set; adopting a K-means clustering algorithm to cluster the high-priority defect region set to obtain a clustering distribution; selecting a sub-region with the highest defect density from each cluster to determine a high-priority defect region in a key defect set; if the defect density of the high-priority defect region in the key defect set is higher than a preset density threshold, adopting a decision tree algorithm to sort defect features to obtain a defect feature priority order; extracting spatial features according to the defect feature priority order to determine a spatial pattern of defect distribution, thereby determining the high-priority defect region.

7. The method of claim 1, wherein the process of embedding the high-priority defect region data into a carrier signal to obtain steganographic data comprises: obtaining a coordinate range and a priority label of the high-priority defect region to obtain an initial defect data set; adopting a data steganography algorithm to embed the initial defect data set into a preset carrier signal to generate steganographic encoding data; if the embedding rate of the steganographic encoding data is lower than a preset threshold, adjusting the frequency parameters of the carrier signal and re-embedding to obtain optimized steganographic data; adopting a discrete cosine transform to perform frequency domain analysis on the optimized steganographic data, extracting steganographic protection feature vectors to generate a frequency domain feature set; if the frequency domain feature set passes an integrity check, the optimized steganographic data is taken as the steganographic data.

8. The method of claim 1, wherein the process of performing end-to-end encryption on the steganographic data to obtain encrypted transmission data comprises: decoding the steganographic data to restore original defect data; adopting an AES algorithm to encrypt the original defect data to generate an encryption key and encrypted data; packaging the encrypted data according to an end-to-end encryption protocol to obtain packaged data; adopting a SHA-256 algorithm to calculate an integrity check value of the packaged data to generate check data; if the check data is consistent with a preset integrity threshold, the packaged data is taken as the encrypted transmission data.

9. The method of claim 1, wherein the process of sending the encrypted transmission data to a cloud server through an underwater communication protocol and verifying its integrity to obtain final non-destructive testing storage data comprises: sending the encrypted transmission data to the cloud server through the underwater communication protocol to obtain transmission data; The CRC check algorithm is used to verify the integrity of the transmission data, and a check result is obtained. If the check result indicates that the data is complete, the transmission data is stored in a cloud server to obtain stored data. The SHA-256 algorithm is used to calculate the hash value of the stored data to obtain a hash identifier. If the hash identifier is consistent with the original data hash value, it is confirmed that the data has not been tampered with, and the final lossless detection storage data is obtained, and the storage time stamp and data index of the lossless detection storage data are recorded to form a storage record.

10. An underwater non-destructive testing system for offshore structures, characterised in that, It comprises: An environmental perception module for obtaining environmental perception data of an underwater detection device and determining environmental adaptation parameters based on the environmental perception data; A path planning module for generating an optimized motion path based on the environmental adaptation parameters; A structure scanning module for collecting structure surface detection data along the motion path; A defect extraction module for obtaining a defect distribution map based on the structure surface detection data; A defect analysis module for dividing high-priority defect areas based on the defect distribution map; A data steganography module for embedding high-priority defect area data into a carrier signal to obtain steganographic data; A data encryption module for end-to-end encryption of the steganographic data to obtain encrypted transmission data; A data transmission module for sending the encrypted transmission data to a cloud server via an underwater communication protocol and verifying its integrity to obtain the final lossless detection storage data.

Images