Submarine cable near-shore detection operation method

By combining geographic information systems, acoustic detection, laser imaging and fiber sensing data, and using deep learning models, the problem of insufficient efficiency and accuracy of submarine cable detection and positioning is solved, and efficient and accurate submarine cable positioning and status evaluation is achieved. , improves the safety maintenance and monitoring capabilities of submarine cables.

CN118940044APending Publication Date: 2024-11-12STATE GRID ZHEJIANG ELECTRIC POWER CO LTD ZHOUSHAN POWER SUPPLY CO +2

Patent Information

Application Number
CN202411123401.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-15
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

In the prior art, the intelligence level of submarine cable detection and positioning measures is low, resulting in the need to improve detection efficiency and accuracy and high cost.

Method used

The combination of a variety of technical means is adopted, including the fusion of geographic information systems, acoustic detection, laser imaging, and fiber sensing data, and data analysis is carried out through deep learning models to build a three-dimensional visual model to achieve accurate positioning and state evaluation of submarine cables.

Benefits of technology

It improves the efficiency and accuracy of submarine cable detection, reduces measurement noise, enhances data quality and reliability, reduces human intervention, improves work efficiency and the accuracy of results, and ensures the safe maintenance and monitoring of submarine cables.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940044A_ABST
    Figure CN118940044A_ABST
Patent Text Reader

Abstract

The invention discloses a near-shore detection operation method for a submarine cable, relates to the technical field of cable detection and positioning, and aims to solve the problems of complex operation and low efficiency of submarine cable detection and positioning at present. The method comprises the steps of collecting historical submarine cable distribution information, geological data and hydrological data of a near-shore area, and performing preliminary analysis through a geographic information system; arranging a plurality of detection devices, and obtaining acoustic detection data, laser imaging data and optical fiber sensing data of the submarine cable; fusing the data to obtain a fused data set, and inputting the data set into a trained deep learning model to obtain cable layout, depth and state information; a three-dimensional visual model is constructed, and the seabed environment and the cable state are evaluated; through the sea surface display system, the mobile mark is placed on the cable and a nearby area to realize cable position calibration. And generating a detection report including the position, depth and state of the cable according to detection and analysis results. According to the invention, the cable information can be efficiently and accurately obtained.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of cable detection and positioning, and in particular to a method for nearshore detection of submarine cables. Background Art

[0002] Submarine cables are cables laid at the bottom of the ocean, mainly used to transmit electricity, communications and data. Submarine cables are usually composed of multiple layers, including conductors, insulation layers, reinforcement layers and protective layers to ensure the durability and reliability of the cables in the marine environment. According to different uses, submarine cables can be divided into the following types: communication cables, power cables and scientific research cables. At the same time, in order to further improve the stability and reliability of submarine cables, it is necessary to detect and locate submarine cables to facilitate maintenance and repair and optimize wiring. However, the current intelligence level of submarine cable detection and positioning measures needs to be improved, and the operation is relatively complicated, which reduces the efficiency of detection and positioning to a certain extent.

[0003] Prior art 1, Chinese patent, application number: 202210282733.3 discloses a submarine cable path detection device and method and a cable fault location method, including a detection ship, a master control module, a rudder control module, a positioning and information transmission system, a cable path detection component, a fault detection component and an onshore centralized control center. By applying a low-frequency voltage to the fault phase of the submarine cable, an alternating magnetic field is generated in the sea area around the cable, and then the direction of the submarine cable is detected and judged, which overcomes the technical defect of the inconvenience of its maintenance caused by the actual line of the submarine cable deviating from the original laid line, and saves time and improves efficiency. The input is used to locate the fault point, which overcomes the problem of time-consuming and labor-intensive determination of the fault point of the submarine cable and the low accuracy of fault point location, and improves the accuracy and efficiency of fault point location. However, its detection means are relatively single, and its accuracy and detection efficiency need to be further improved.

[0004] Prior art 2, Chinese patent, application number: 202311627168.0 discloses a method for offshore submarine cable positioning based on USBL and SINS / GPS system. In the installation error setting stage, the installation error of the ultra-short baseline (USBL) and strapdown inertial navigation (SINS) / global positioning system (GPS) combined positioning system is optimized based on the whale optimization algorithm (WOA), wherein the SINS / GPS system is directly installed on the ship, and the USBL is installed on a ship-based six-degree-of-freedom stabilization device; at the same time, the position of the beacon relative to the USBL transducer is uniformly calculated to the earth coordinate system to achieve accurate positioning of the submarine cable. The input improves the positioning accuracy and stability by minimizing the cost function and eliminating the additional requirements for the separation-based calibration method, and realizes the unified calibration of the system installation error, which provides an important reference for the installation error correction of the underwater combined positioning system and the positioning of the submarine cable. However, its system is relatively complex and requires professional personnel to operate, resulting in a low degree of intelligence in the system.

[0005] Prior art three, Chinese patent, application number: 202110925087.3 discloses a method and device for detecting the buried depth of submarine cables in shallow waters, the method comprising: obtaining the average sound speed and draft depth during the navigation process of the survey ship according to the specified ship speed and the navigation pre-laid survey line; inputting the average sound speed and draft depth into the depth sounder to measure the water depth and obtain instantaneous water depth data and positioning data; using the side-scan sonar to conduct a seabed topography survey and obtain side-scan sonar data; using the shallow layer profiler to lay out the specified encrypted cable detection survey line to obtain profile data; drawing a track map; generating a water depth map; generating a landform map; obtaining a cable status profile map containing the buried depth of the submarine cable; according to the track map, water depth map, landform map and cable status profile map, the submarine cable route is divided into sections to obtain the submarine cable buried result map, which greatly reduces the cost of the traditional submarine cable buried depth survey operation, and also improves the safety of the operation, so as to realize the investigation of the buried depth of the submarine cable in shallow waters. However, it requires the cooperation of multiple devices, resulting in a high cost of detection.

[0006] At present, the existing technologies 1, 2 and 3 have a single technical means, resulting in the problem that the detection efficiency and accuracy of submarine cables need to be further improved. Therefore, the present invention provides a method for nearshore detection of submarine cables, which improves the detection efficiency and accuracy of submarine cables by combining multiple technical means, and provides a solid foundation for the safe maintenance of submarine cables. Summary of the invention

[0007] The main purpose of the present invention is to provide a method for nearshore detection of submarine cables to solve the problem that the conventional technology has a single technical means, resulting in the need to further improve the detection efficiency and accuracy of submarine cables.

[0008] To achieve the above object, the present invention provides the following technical solutions: A method for nearshore detection of submarine cables, comprising: Collect historical submarine cable laying information, geological data and hydrological data in the nearshore area, and conduct preliminary analysis through the geographic information system; deploy a variety of detection equipment to obtain acoustic detection data, laser imaging data and fiber optic sensor data of submarine cables; The acoustic detection data, laser imaging data and fiber optic sensor data are integrated to obtain a fused data set, which is then input into a trained deep learning model to obtain cable layout, depth and status information; a three-dimensional visualization model is constructed based on the obtained cable layout, depth and status information to evaluate the seabed environment and cable status; Through the sea surface display system, mobile markers are placed on the cable and nearby areas to calibrate the cable position; based on the detection and analysis results, a detection report is generated, including the cable position, depth and status.

[0009] As a further improvement of the present invention, preliminary analysis is performed through a geographic information system, comprising the following steps: Identify various available data sources, including government databases, industry reports, scientific research papers, and historical survey databases; obtain relevant historical submarine cable deployment information from each identified data source, as well as geological and hydrological data in the area, including cable type, deployment time, deployment depth, and surrounding soil, rock type, water velocity, and wave height; Standardize all data from the acquired data sources, import the collected data into the geographic information system, and store the data in the spatial database; generate heat maps, perform buffer analysis, and relative position analysis to identify the relationship between cable layout and geographic features; analyze the distribution pattern of historical cable layout and predict the geographic areas and risk factors that may be associated with future cable layout; Through the relative position and distribution pattern of cable laying, various detection equipment are deployed, and acoustic sensors, laser imagers and fiber optic sensor equipment are deployed to the predetermined seabed locations.

[0010] As a further improvement of the present invention, the evaluation of the seabed environment and the cable status includes the following steps: De-noise the acoustic detection data, laser imaging data, and fiber optic sensor data, convert all types of data into a unified format, synchronize the timestamps of data from different sources, and spatially match the data from different sensors based on geographic coordinates; select a data fusion network, synthesize the data, and obtain a fused data set that contains information on the location, shape, and status of the cable; Select a convolutional neural network, use historical data to train the deep learning model, and predict the layout, depth, and status of the cable based on the input data; input the fused data set into the trained deep learning model, and after reasoning, output the location, depth, and status information of the cable in the form of probability distribution or confidence, indicating the most likely cable characteristics; The cable layout, depth and status features are extracted from the deep learning model output. Using visualization tools, the extracted cable features are converted into a three-dimensional visualization model. The cable's location, shape, depth and corresponding environmental characteristics are all represented. Combined with the three-dimensional visualization model, an in-depth analysis is conducted to evaluate the seabed environment, potential risks and the operating status of the cable.

[0011] As a further improvement of the present invention, the data fusion network synthesizes the data, including the following steps: Establish a multimodal input layer of acoustic data, laser imaging data and optical fiber data to retain the time series characteristics of acoustic signals, the spatial characteristics of laser imaging, and the characteristic types of data types of continuous monitoring data of optical fiber sensing; Construct multiple parallel feature extraction networks to automatically extract meaningful features from each type of data, and use the attention mechanism to weight the features so that the attention mechanism can adaptively focus on important information; use the message passing network to perform multi-level feature fusion after feature extraction, combining low-level features with high-level features to maintain contextual relationships; Through the dynamic modeling mechanism, the dynamic model parameters are adjusted through real-time feedback so that it can be re-learned when new data is collected; the fused output will be processed by a graph neural network to capture the relationship between the cable and environmental characteristics, connect each feature with its spatial position, and output the cable position, depth and status information.

[0012] As a further improvement of the present invention, multi-level feature fusion is performed, including the following steps: The low-level and high-level features obtained by the feature extraction network are embedded in the space mapping, the features are mapped to a low-dimensional space through a linear layer, and then a nonlinear transformation is applied for enhancement; Build a graph model that treats features at different levels as nodes in a graph. The connections between features represent their relationships. Each node interacts with its neighbor nodes through a message passing mechanism to learn deeper feature relationships and contextual information. During the fusion process, an adaptive weighting mechanism is applied to dynamically adjust the weights according to the importance of the features and the context. More important features occupy a larger proportion in the final fusion result. The weight calculation is based on automatic learning based on the attention mechanism.

[0013] As a further improvement of the present invention, constructing a graph model includes the following steps: The acoustic features, laser imaging features and optical fiber sensing features of multi-level feature fusion correspond to one node; the nodes are layered according to the level of the features; Calculate the similarity between different features through distance measurement, and use similarity to build edges; assign attributes to nodes and edges. Node attributes include: feature vector, corresponding to the feature expression of the node; hierarchical information, identifying the hierarchy of features; edge attributes include: edge weight, connection strength or similarity; edge type, indicating what type of relationship; Use the graph processing library to build and manage the graph structure, create an empty graph, add nodes, and add edges; apply the message passing mechanism to interact with information in the graph model, and each node will update its own status based on the information of its neighbor nodes; the node receives information from the neighbor nodes through the edges it connects, and calculates the new feature representation of each node according to the message update rules.

[0014] As a further improvement of the present invention, the dynamic modeling mechanism comprises the following steps: Establish a real-time feedback system to monitor the performance of the model when new data is input. By evaluating the deviation between the output results and the actual situation in real time, feedback signals are automatically generated to guide the adjustment of model parameters. Through regularization techniques, constraints are imposed on model parameters to keep the parameters balanced when updating; Based on real-time feedback information, the fusion weights of each modal feature are adaptively adjusted; node features are dynamically updated according to the input new data, and during the updating process, the weights and relationships of the edges are dynamically adjusted.

[0015] As a further improvement of the present invention, the training of the deep learning model includes the following steps: De-noise historical data from acoustic detection, laser imaging, and fiber optic sensing, synchronize data from different sensors based on timestamps, and spatially match data using geographic coordinates; Introduce regularization terms in the loss function, use historical data to train the deep learning model, monitor the learning curve during training and perform early stopping calculations, use adaptive learning rate strategies to optimize parameter update speed, use validation sets to evaluate model performance, analyze the performance of deep learning models in different situations, and adjust network parameters and architectures, number of layers, and convolution kernel size in a timely manner; The performance of the deep learning model was evaluated on an independent test set, and standard indicators including accuracy, recall, and F1 score were used for quantitative evaluation. The ROC curve was drawn to analyze the performance of the deep learning model on different cable states.

[0016] As a further improvement of the present invention, monitoring the learning curve during the training process and performing early stopping calculations include the following steps: At the end of each training round, record the current training loss, validation loss, and other related indicators, and use data visualization tools to draw learning curves in real time to show the changing trends of the training set and validation set losses; Set an early stopping threshold to decide when to stop training; at the end of each round, check the change in validation loss. If the validation loss does not improve within the set rounds, enter the early stopping mechanism; Verify that the loss reaches a minimum value and save the current deep learning model parameters; if the early stopping condition is met, terminate the training.

[0017] As a further improvement of the present invention, the cable position marking is realized, comprising the following steps: Set moving marks through the sea surface display system to calibrate the analyzed cable positions; By interpolation, the moving marker is adjusted according to the real-time estimated cable position; Record all data during the position calibration process, including the real-time position and depth of the mobile marker and its relationship with the cable position, and feed it back to the central monitoring system in real time; Based on the results of all detection and position calibration, a detection report is automatically generated, including the specific location, depth and current status of the cable, with 3D visualization results and risk assessment information.

[0018] The data collection and detection equipment of the present invention is deployed to collect and integrate the historical submarine cable laying information, geological data and hydrological data in the nearshore area, and use the geographic information system (GIS) for preliminary analysis to provide background information on the possible cable laying location and environmental characteristics; a variety of detection equipment (such as acoustic sensors, laser equipment and fiber optic sensors) are deployed to obtain different types of data to ensure multi-dimensional coverage of the entire detection area. The significance achieved: Through historical data and environmental analysis, the target area can be better understood, the information blind spots in cable detection can be reduced, and the pertinence and effectiveness of subsequent detection can be improved; the comprehensive application of a variety of detection equipment provides a rich information basis for subsequent data fusion and analysis, ensuring the acquisition of more comprehensive and accurate submarine cable information. Data fusion and deep learning analysis, by fusing acoustic detection data, laser imaging data and fiber optic sensor data, form an integrated data set. This fusion can reduce measurement noise and improve the overall quality and reliability of data; the fused data set is analyzed using a trained deep learning model to automatically extract the layout, depth and status information of the cable, greatly improving the analysis efficiency and accuracy; a three-dimensional visualization model is constructed based on the acquired information, so that the status of the cable and its surrounding environment can be seen at a glance. Significance achieved: Through deep learning technology, complex data sets can be automatically processed, human intervention can be reduced, and work efficiency and accuracy of results can be improved; the three-dimensional visualization model provides a more intuitive interface, which is convenient for multiple parties and relevant personnel to understand the submarine environment and the status of the cable, and provide support for subsequent decision-making. Cable calibration and detection report generation, through the sea surface display system to mark the movement of the cable and its surrounding areas, so as to accurately calibrate the cable position. This process ensures the traceability of the cable in subsequent maintenance and monitoring work; based on the detection and analysis results, a detailed detection report is generated, including the location, depth and status information of the cable, providing reference materials for relevant management and decision makers. Significance achieved: During the calibration process, construction interference and accidental injuries can be reduced, submarine cables can be protected, and the safety of offshore operations can be improved; the generation of the detection report ensures that all relevant parties can clearly understand the current status of the cable, providing a valuable basis for future monitoring, maintenance and management decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic diagram of a process flow of an embodiment of the present invention; Figure 2 A schematic flow chart of the steps of performing preliminary analysis through a geographic information system according to an embodiment of the present invention; Figure 3 A schematic diagram of a flow chart of steps for evaluating the submarine environment and cable status according to an embodiment of the present invention; Figure 4 A schematic diagram of a process flow for implementing cable location marking according to an embodiment of the present invention; Figure 5 It is a structural schematic diagram of an embodiment of an electronic device of the present invention; Figure 6 It is a schematic structural diagram of an embodiment of the storage medium of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] like Figure 1 As shown, this embodiment provides an embodiment of a method for nearshore detection of a submarine cable, and the method for nearshore detection of a submarine cable specifically includes the following steps: S1: Collect historical submarine cable laying information, geological data and hydrological data in the nearshore area, and conduct preliminary analysis through the geographic information system; deploy a variety of detection equipment to obtain acoustic detection data, laser imaging data and fiber optic sensing data of submarine cables; S2: Fuse the acoustic detection data, laser imaging data and fiber optic sensor data to obtain a fused data set, and input the data set into the trained deep learning model to obtain the cable layout, depth and status information; construct a three-dimensional visualization model based on the obtained cable layout, depth and status information to evaluate the seabed environment and cable status; S3: Through the sea surface display system, mobile markers are placed on the cable and nearby areas to calibrate the cable position; based on the detection and analysis results, a detection report is generated, including the cable position, depth and status.

[0022] In this embodiment, a combination of multiple detection technologies is used, such as the integrated use of low-frequency acoustic detection, laser scanning, magnetic detection and fiber optic sensing technology to improve accuracy and reliability; an unmanned submersible (ROV) or an autonomous underwater vehicle (AUV) is equipped with a variety of sensors and imaging equipment; a sonar system is used to perform a wide range of preliminary scans to identify potential cable paths and draw a rough topographic map; in a specific area, laser scanning or high-resolution sonar imaging is used for detailed detection to obtain the precise location of the cable and the characteristics of the seabed; fiber optic sensors are deployed along the cable laying path to monitor temperature, strain and other data in real time to provide important information for subsequent positioning.

[0023] In step S1 of this embodiment, data collection and detection equipment deployment are used to collect and integrate historical submarine cable deployment information, geological data and hydrological data in the nearshore area, and a geographic information system (GIS) is used for preliminary analysis to provide background information on the possible cable deployment location and environmental characteristics; a variety of detection equipment (such as acoustic sensors, laser equipment and optical fiber sensors) are deployed to obtain different types of data to ensure multi-dimensional coverage of the entire detection area. The significance achieved: Through historical data and environmental analysis, the target area can be better understood, the information blind spots in cable detection can be reduced, and the pertinence and effectiveness of subsequent detection can be improved; the comprehensive application of multiple detection equipment provides a rich information basis for subsequent data fusion and analysis, ensuring that more comprehensive and accurate submarine cable information is obtained. Step S2 data fusion and deep learning analysis, by fusing acoustic detection data, laser imaging data and optical fiber sensor data to form an integrated data set. This fusion can reduce measurement noise and improve the overall quality and reliability of data; the fused data set is analyzed using a trained deep learning model to automatically extract the layout, depth and status information of the cable, greatly improving the analysis efficiency and accuracy; a three-dimensional visualization model is constructed based on the acquired information, so that the status of the cable and its surrounding environment can be seen at a glance. Significance achieved: Through deep learning technology, complex data sets can be automatically processed, human intervention can be reduced, and work efficiency and accuracy of results can be improved; the three-dimensional visualization model provides a more intuitive interface, which is convenient for multiple parties and relevant personnel to understand the submarine environment and the status of the cable, and provide support for subsequent decision-making. Step S3 Cable calibration and detection report generation, through the sea surface display system to mark the movement of the cable and its surrounding areas, so as to accurately calibrate the cable position. This process ensures the traceability of the cable in subsequent maintenance and monitoring work; based on the detection and analysis results, a detailed detection report is generated, including the location, depth and status information of the cable, providing reference materials for relevant management and decision makers. Significance achieved: During the calibration process, construction interference and accidental injuries can be reduced, submarine cables can be protected, and the safety of offshore operations can be improved; the generation of the detection report ensures that all relevant parties can clearly understand the current status of the cable, providing a valuable basis for future monitoring, maintenance and management decisions.

[0024] In summary, through the above steps, the detection process of submarine cables in this embodiment can not only achieve efficient and accurate acquisition of cable information, but also help improve the safety and efficiency of the entire marine engineering. Combined with advanced technologies such as data fusion, deep learning and three-dimensional visualization, the entire process ensures the integrity, reliability and intuitiveness of information, meeting the high standards of modern submarine engineering construction and maintenance. The implementation of this operation process can effectively reduce potential risks and extend the service life of cables, thereby improving the overall performance and service quality of the submarine cable system.

[0025] Furthermore, if Figure 2As shown, the process of performing preliminary analysis by using a geographic information system in step S1 specifically includes the following steps: S11: Identify various available data sources, including government databases, industry reports, scientific research papers, and historical survey databases; obtain relevant historical submarine cable deployment information from each identified data source, as well as geological and hydrological data in the area, including cable type, deployment time, deployment depth, and surrounding soil, rock type, water velocity, and wave height; S12: Standardize all data from the acquired data sources, import the collected data into the geographic information system, and store the data in the spatial database; generate heat maps, perform buffer analysis, and relative position analysis to identify the relationship between cable layout and geographic features; analyze the distribution pattern of historical cable layouts, and predict geographic areas and risk factors that may be associated with future cable layouts; S13: Deploy various detection equipment based on the relative position and distribution pattern of the cable laying, and deploy acoustic sensors, laser imagers, fiber optic sensors and other equipment to the predetermined seabed locations.

[0026] In the S11 data source determination and acquisition of this embodiment, by determining multiple data sources including government databases, industry reports, scientific research papers and historical survey databases, the comprehensiveness and accuracy of the information are ensured, which can provide a broad background for subsequent analysis; the information obtained covers the historical submarine cable laying data and related geological and hydrological data, such as cable type, laying time, laying depth, and the soil, rock type, water flow speed, wave height, etc. of the surrounding environment. The significance achieved: Obtaining comprehensive historical and environmental data can provide a reliable basis for subsequent detection work and improve the scientific nature of decision-making; multi-dimensional data collection can help identify potential environmental risk factors and help formulate effective response plans. S12 data standardization processing and analysis, standardize the acquired data and import it into GIS to ensure the consistency of data formats from different sources, which is convenient for subsequent analysis; through heat map generation, buffer analysis and relative position analysis, identify the relationship between cable laying and important geographical features (such as water bodies, terrain, soil types, etc.), and provide support for the formulation of detection strategies; through spatial analysis of historical laying data, extract the distribution pattern of cable laying, and predict possible laying areas and potential risks in the future. Significance achieved: Converting raw data into meaningful information makes geographic data have higher application value in practice and improves the efficiency of subsequent detection; identifying the relationship between cables and geographic features helps optimize the layout of detection equipment and improve the pertinence and effectiveness of detection. S13 detection equipment layout, through understanding the relative position and distribution pattern of cable layout, scientifically guides the layout of detection equipment, and places acoustic sensors, laser imagers, fiber optic sensors and other equipment at the seabed location most likely to obtain effective information; ensures that sensors with different technical backgrounds are reasonably matched to achieve multi-sensor collaborative work. Significance achieved: Accurate layout can significantly improve the accuracy of detection results and provide high-quality basic information for subsequent data acquisition; reasonable selection of layout location can better adapt to environmental changes and potential risks, ensuring that subsequent cable monitoring and maintenance work is more efficient and safe.

[0027] In summary, this embodiment forms a closed-loop process of data collection, analysis and application through the steps of preliminary analysis by geographic information system. The implementation of each step can significantly improve the scientificity, effectiveness and safety of submarine cable detection. It not only provides strong support for the laying of cables, engineering construction and subsequent maintenance, but also promotes the scientific and Internet-based development of submarine environmental protection and management. Through the preliminary analysis process, scientific understanding of historical data and environmental characteristics can help design specific plans for subsequent acoustic detection, laser imaging and fiber optic sensing. After clarifying the potential area, the detection location, the type of detection equipment and the basic parameters of the option can be selected in a targeted manner, thereby improving the efficiency and effectiveness of subsequent data collection.

[0028] Furthermore, if Figure 3 As shown, the process of evaluating the seabed environment and cable status in step S2 specifically includes the following steps: S21: De-noise the acoustic detection data, laser imaging data, and fiber optic sensor data, convert all types of data into a unified format, synchronize the timestamps of data from different sources, and spatially match the data from different sensors based on geographic coordinates; select a data fusion network, synthesize the data, and obtain a fused data set that contains information on the location, shape, and status of the cable; S22: Select a convolutional neural network, use historical data to train the deep learning model, and predict the layout, depth, and status of the cable based on the input data; input the fused data set into the trained deep learning model, and after reasoning, output the location, depth, and status information of the cable in the form of probability distribution or confidence, indicating the most likely cable characteristics; S23: Extract cable layout, depth and status features from the deep learning model output, and use visualization tools to convert the extracted cable features into a three-dimensional visualization model. The cable's location, shape, depth and corresponding environmental features (such as surrounding terrain, other obstacles, etc.) are all represented; combined with the three-dimensional visualization model, conduct in-depth analysis and assessment of the seabed environment, potential risks and the operating status of the cable.

[0029] Step S21 of this embodiment is data preprocessing and fusion. By denoising the acoustic detection data, laser imaging data and fiber optic sensor data, external interference is reduced, and the signal-to-noise ratio of the data is improved, which helps to improve the quality of subsequent analysis; all data are converted into a unified format, so that different types of sensor data can be easily operated and analyzed, which improves the compatibility and operability of the data; the data of different sensors are synchronized with timestamps and spatial coordinates are matched to ensure the relevance of data at the same time and space position, and enhance the reliability of the data; by selecting a data fusion network, data from different sources are synthesized to form a comprehensive data set, which contains more comprehensive cable information (such as location, shape and status), providing effective support for subsequent analysis. The significance achieved: Enhance the reliability of data and ensure that the results of all subsequent analyses are based on high-quality basic data; lay a good foundation for subsequent analysis and modeling, thereby improving the accuracy of analysis and the reliability of decision-making. Step S22: Application of deep learning model, select convolutional neural network (CNN) and use historical data to train the model so that it can effectively identify and predict the layout, depth and status information of the cable; input the fused data set into the trained deep learning model, and output the status information of the cable after calculation, which is usually presented in the form of probability distribution or confidence, indicating the accuracy and credibility of the recognition. Significance achieved: The deep learning model extracts deep features in the data, making the prediction of the cable position, depth and status more accurate; based on the output results of the model, the operating status and potential risks of the cable are efficiently evaluated, providing a scientific basis for equipment maintenance and submarine environmental protection. Step S23: Three-dimensional visualization and analysis evaluation, extract cable layout, depth and status features from the output of the deep learning model, and use visualization tools to build a three-dimensional visualization model, so that the specific characteristics of the cable (location, shape, depth, etc.) and its adjacent environment are intuitively presented; combined with the three-dimensional visualization model, the submarine environment is deeply analyzed to identify potential risk factors and the operating status of the cable, and a comprehensive evaluation is formed. Significance achieved: Through 3D visualization, relevant personnel can more intuitively understand complex data relationships and improve the communication efficiency between decision makers and relevant stakeholders; it can effectively identify environmental risks and cable status, support early warning of potential problems and the formulation of subsequent maintenance strategies, and improve the safety and reliability of submarine cables.

[0030] In summary, the three steps of this embodiment form a complete closed-loop system of data preprocessing, intelligent analysis and visualization in terms of technical effect. It not only improves the scientificity and accuracy of submarine cable monitoring, but also plays an important role in assessing the potential risks of the submarine environment, and ultimately provides strong support for the safe operation and maintenance of the cable.

[0031] Furthermore, the process of synthesizing data by the data fusion network in step S21 specifically includes the following steps: S211: Establish a multimodal input layer of acoustic data, laser imaging data and optical fiber data, and retain the time series characteristics of acoustic signals, the spatial characteristics of laser imaging, and the characteristic types of the data types of the continuous monitoring data of optical fiber sensing; S212: Build multiple parallel feature extraction networks to automatically extract meaningful features from each type of data, and use the attention mechanism to weight the features so that the attention mechanism can adaptively focus on important information; use the message passing network to perform multi-level feature fusion after feature extraction, combining low-level features with high-level features to maintain contextual relationships; Among them, the feature extraction network specifically includes: Adaptive convolution module: It is used to dynamically adjust the size and shape of the convolution kernel according to the actual characteristics of the input data. When processing acoustic data, laser imaging and fiber optic sensor data, it can automatically optimize the convolution parameters based on the characteristics of the input signal. It significantly improves the efficiency of feature extraction under different data modes, allowing the model to perform better in high-noise and variable environments. Multi-granularity feature extraction module: used to introduce a multi-granularity feature extraction module into the feature extraction network to extract features at different scales and resolutions; for example, by constructing sub-networks of different depths, multi-level extraction of short-term features of acoustic signals, global features of laser images, and time series features of optical fiber monitoring data can be achieved; multi-granularity feature extraction can ensure that the network captures features at multiple scales, thereby providing rich information representation and supporting subsequent processing; Integrated nonlinear transformation layer module: used to integrate multiple nonlinear activation functions, such as parameterized ReLU (PReLU) and Swish, in the feature extraction process. These activation functions are applied in feature extraction at different levels to improve the ability to capture nonlinear features. By selecting appropriate activation functions, the model can capture complex feature interactions during the extraction process and improve the richness of feature expression. S213: Through the dynamic modeling mechanism, the dynamic model parameters are adjusted through real-time feedback so that it can be re-learned when new data is collected; the fused output will be processed by a graph neural network to capture the relationship between the cable and environmental features, connect each feature with its spatial position, and output the cable position, depth and status information.

[0032] In step S211 of this embodiment, a multimodal input layer is established. By setting up a multimodal input layer, different types of data (acoustic signals, laser imaging, fiber optic sensing) can be classified and processed to ensure that their important features are not ignored during the processing; by retaining the features of multiple data types, the input data dimension of the model is improved, so that the model can process richer information. Significance achieved: By directly supporting multiple data types at the input layer, the integrity and accuracy of the fused data are enhanced, ensuring that subsequent processing and analysis are more reliable; this design can effectively cope with the complexity of data from different sensors in the real environment, improve the adaptability of the model, and provide a more sufficient information basis for the training and reasoning of subsequent deep learning models. Step S212 feature extraction and fusion, using multiple parallel feature extraction networks, can simultaneously extract meaningful features from different data sources, improve the speed and accuracy of information extraction; this mechanism allows the model to adaptively identify and focus on important features, thereby improving the quality of feature extraction; the message passing network is used to achieve the fusion of low-level and high-level features, so that features at different levels and from different sources maintain context information and obtain deeper knowledge representation. Significance achieved: Through parallel processing and attention mechanism, the model can accurately identify and utilize key features and improve the ability to express information; multi-level feature fusion ensures that the relationship between each feature is better captured, enhances the understanding of cable status and environmental characteristics, and supports subsequent more complex reasoning tasks. Step S213 dynamic modeling and graph neural network processing allow the model to continuously adjust and optimize parameters through real-time feedback, so that it can self-learn and adapt during the data collection process, ensuring the flexibility and accuracy of the model when facing new data; through graph neural network processing, the cable features are connected with their environmental features, which can better capture complex relationships, and the output results include cable location, depth and status information. Significance achieved: The dynamic modeling mechanism ensures that the model can be continuously optimized as the data features change, with a wider range of applicability, which helps to maintain model performance in complex and changing environments; the application of graph neural networks makes it easier for the model to capture the complex relationship between the cable and the surrounding environment, providing a more accurate cable status assessment, and providing rich information for potential risk identification and decision support.

[0033] In summary, each step of this embodiment provides effective technical support for data fusion and ensures the effective integration and utilization of different data sources. These steps effectively improve the responsiveness, adaptability and accuracy of the model at the system level, enabling more accurate evaluation of the submarine environment and cable monitoring. The implementation of each step not only enhances the effect of data fusion, but also provides a scientific and reasonable technical path and method for cable status inspection in complex environments.

[0034] Furthermore, the process of performing multi-level feature fusion in step S212 specifically includes the following steps: S2121: embedding space mapping of low-level and high-level features obtained through feature extraction network, mapping features to low-dimensional space through linear layers, and then applying nonlinear transformation for enhancement; S2122: Build a graph model that treats features at different levels as nodes in a graph. The connections between features represent their relationships. Each node interacts with its neighbor nodes through a message passing mechanism to learn deeper feature relationships and context information. S2123: During the fusion process, an adaptive weighting mechanism is applied to dynamically adjust the weights according to the importance of the features and the context. More important features have a larger proportion in the final fusion result. The weight calculation is based on automatic learning based on the attention mechanism.

[0035] Step S2121 of this embodiment embeds spatial mapping, and maps high-dimensional features to low-dimensional space through linear layers to achieve feature compression, making the feature representation more compact and reducing computational complexity; applying nonlinear transformations can make full use of the nonlinear relationship between features, enhance the richness of feature representation, and enable the model to capture more complex data patterns. Significance achieved: Through low-dimensional mapping of features, redundant information is reduced, the computational efficiency of the model is improved, and thus the training and reasoning speeds are improved; nonlinear transformations enable the final feature representation to more comprehensively reflect the complex relationship between metadata, improving the efficiency of information utilization and the predictive ability of the model. Step S2122 graph model construction and message passing, constructing a graph model, treating features at different levels as nodes, can effectively represent the significant relationship between features, and provide a natural way to capture the interaction between features; through the message passing mechanism, nodes can interact with neighboring nodes, so that each node can dynamically update its state according to neighborhood features, thereby learning deeper feature relationships. Significance achieved: This graph-based feature representation and interaction mechanism enhances the model's understanding of the complex relationship between features, enabling the model to identify potential contextual dependencies; the graph model accesses the interactive relationship between features, allowing information exchange in a more flexible way, and enhancing the effect of multi-level feature fusion. Step S2123 adaptive weighted fusion, applying the adaptive weighting mechanism, the model can dynamically adjust the weight according to the importance of the feature and contextual information, and through the attention mechanism, the model can automatically learn which features are more important; through the attention mechanism, the model can assign different weights to different features, and can focus more on those features with decision-making significance, thereby improving the representativeness of the final fusion feature. Significance achieved: The adaptive weighting mechanism can ensure that the final fusion feature pays more attention to key features, thereby improving the accuracy and reliability of the fused features and achieving more accurate decision support; through the attention mechanism, it can clearly show which features have a greater impact on the model's judgment, improving the transparency and interpretability of the model, and providing a better basis for practical applications.

[0036] In summary, the design and functions of the steps in this embodiment complement each other and work together on the overall process of multi-level feature fusion. The computational efficiency and feature expression ability are improved through low-dimensional embedding space mapping, the deep relationship between features is learned through the graph model, and effective feature selection and fusion are performed in combination with adaptive weighting, thereby significantly improving the overall model's processing ability and accuracy for multimodal data in complex scenarios. This not only makes the model perform well in specific tasks, but also provides strong support for subsequent applications and optimization.

[0037] Furthermore, the process of constructing a graph model in step S2122 specifically includes the following steps: S21221: The acoustic feature, laser imaging feature and optical fiber sensing feature of multi-level feature fusion correspond to one node; the nodes are layered according to the level of the features; S21222: Calculate the similarity between different features through distance measurement, and use similarity to build edges; assign attributes to nodes and edges. Node attributes include: feature vector, which is the feature expression of the corresponding node; hierarchical information, which identifies the hierarchy of the feature (low or high level); edge attributes include: edge weight, connection strength or similarity; edge type, which indicates the type of relationship; S21223: Use the graph processing library to build and manage the graph structure, create an empty graph, add nodes, and add edges; apply the message passing mechanism to interact with information in the graph model. Each node will update its own status based on the information of its neighbor nodes; the node receives information from the neighbor nodes through the edges connected to it, and calculates the new feature representation of each node according to the message update rules.

[0038] Preferably, in step S21221 of this embodiment, nodes are created and layered, and each node corresponds to a feature (acoustic feature, laser imaging feature, fiber optic sensing feature), which helps to clarify the key features that need to be analyzed and fused in subsequent processing; layering according to the hierarchical relationship of features (low-level, middle-level, high-level) can provide a logical basis for subsequent feature fusion, so that features at different levels can be correctly combined and utilized. Significance achieved: By layering the features, the graph model has a clearer hierarchical structure, which facilitates understanding and processing of the relationship between features; the hierarchical node configuration provides a basis for message transmission in the graph, and low-level features can pass information to high-level features to form an effective feature hierarchy structure, further enriching feature expression. Step S21222 edge construction and attribute assignment, calculate the similarity between different features through distance measurement, provide a quantitative basis for the connection between features, and ensure that the connection is based on the similarity of real data features; assign attributes to each node and edge, and use feature vectors, hierarchical information, edge weights and types to describe the properties of nodes and edges, and enhance the expressive power of the graph. Significance achieved: Through the construction of edges and attribute assignment, the graph model can effectively describe the relationship between features, can intuitively reflect the interaction between features, and improve the quality and depth of feature fusion; the edges and nodes with attributes can enable the model to dynamically reflect the importance and connection relationship of different features, making the model more flexible in different data environments. Step S21223 Management and message passing of graph structure, using the graph processing library to create and manage graph models, making node and edge operations (addition, deletion, modification, and query) efficient and convenient, enhancing the controllability and scalability of graph operations; applying the message passing mechanism, nodes can exchange information in the graph, and update according to the influence of neighbor nodes on their own features, realizing dynamic feature learning. Significance achieved: The message passing mechanism ensures that feature information is not limited to a single node, but effectively propagates information to the entire graph through connected edges, enhancing the contextual information between features; through the update of node features in the graph, deep learning frameworks such as graph neural networks (GNNs) can be used to further improve the accuracy and intelligence of feature extraction and fusion, and adapt to complex tasks.

[0039] In summary, each step of constructing the graph model in this embodiment enhances the ability of multi-level feature fusion by adding structure and attributes to the graph and realizing the interaction between potential features. The layering and similarity of nodes construct a reasonable and dynamic feature network, so that multimodal feature information can be fully utilized when performing feature fusion. The overall effect is that through more detailed feature modeling and information flow, the model can not only process complex data, but also adapt more flexibly to the needs of different scenarios and conditions.

[0040] Furthermore, the process of the dynamic modeling mechanism in step S213 specifically includes the following steps: S2131: Establish a real-time feedback system to monitor the performance of the model when new data is input, and automatically generate feedback signals to guide the adjustment of model parameters by real-time evaluation of the deviation between the output results and the actual situation; Among them, error calculation: compare the error between the model output and the actual value, and provide feedback according to the set optimization goal (such as minimizing the mean square error); Adaptive learning rate adjustment: Based on the current error, the learning rate is dynamically adjusted to achieve a more efficient training process, making the model more adaptable and flexible when dealing with new data; S2132: Regularization techniques are used to impose constraints on model parameters so that the parameters remain balanced when they are updated; S2133: Based on real-time feedback information, adaptively adjust the fusion weights of each modal feature; dynamically update node features according to the input new data. During the update process, the weights and relationships of the edges are dynamically adjusted.

[0041] In step S2131 of this embodiment, the establishment of a real-time feedback system can timely identify the deficiencies of the model when processing new data by calculating the deviation between the model output and the actual value; through the set optimization goal (such as minimizing the mean square error), the feedback system can systematically guide the model to adjust in the desired direction. Dynamic adjustment of the learning rate is achieved, and the learning rate is automatically adjusted according to the size of the current error, allowing the model to take a larger step for optimization when the error is large, and gradually refine the adjustment process; improve training efficiency, prevent the model from falling into a local minimum when facing new data, and more effectively respond to data changes. Significance achieved: Through real-time monitoring, the model can respond to new data changes, quickly adapt to the environment, and enhance the flexibility and robustness of the model; introduce a feedback mechanism so that the model can be continuously optimized during the training process, enhance the overall learning quality, and help achieve more accurate results. Step S2132 Application of regularization technology, introduce regularization technology (such as L1 regularization or L2 regularization), by adding constraints to the parameters, so that the extreme situation of parameters being too large or too small is avoided during the update. This balancing effect is crucial to controlling the complexity and stability of the model. Significance achieved: By maintaining the balance and reasonable range of parameters, the overfitting phenomenon of the model on the training set is reduced, thereby improving the generalization ability of the model on unseen data sets; optimizing the stability of the model, so that when facing different input data, the output of the model can still maintain a reasonable range of variation, which is conducive to reliability in practical applications. Step S2133 adaptive feature fusion, according to the current input new data, real-time adjustment of the fusion weights of different modal features (acoustic features, laser imaging features, fiber optic sensing features), so that the model can timely optimize the feature combination according to different situations; in the graph neural network, the node features are adaptively updated according to the latest data, so that the representation of each node can continuously reflect new information, and the weights and relationships of the edges are adjusted at the same time to ensure that the optimal information passage path is maintained in the feature fusion. Significance achieved: Adaptive adjustment of the fusion weight of the feature can ensure that more important features dominate the fusion process, thereby improving the representation ability after feature fusion, which is particularly important in complex environments or when the data noise is large; real-time update of node features and edge relationships ensures that the model can sensitively capture data changes in different situations, thereby improving the recognition and response capabilities of important patterns.

[0042] In summary, the entire dynamic modeling mechanism of this embodiment greatly improves the flexibility and robustness of the model in processing multimodal data through three steps: real-time feedback, regularization technology, and adaptive feature fusion. Each step not only promotes the learning and optimization process of the model, but also ensures the effectiveness and stability of the model when facing unknown or complex environments. Overall, this mechanism provides a technical framework for multimodal data analysis that can self-learn, dynamically adjust, and operate stably, achieving more efficient data fusion.

[0043] Furthermore, the training process of the deep learning model in step S22 specifically includes the following steps: S221: De-noise historical data from acoustic detection, laser imaging, and fiber optic sensing, synchronize data from different sensors based on timestamps, and spatially match data using geographic coordinates; Among them, the convolutional neural network is selected as the basic architecture, and multi-layer convolution and pooling modules are designed to extract the spatial features of the data, reduce its dimension and extract features. The feature extraction and fusion module introduces a feature fusion layer after the convolution layer to effectively aggregate data features from different modalities to enhance the model's adaptability to complex environments; it combines the graph neural network (GNN) to capture the spatial relationship and dependency between features to ensure that the relationship between the cable and the environment is fully represented; S222: Introduce regularization terms in the loss function, use historical data to train the deep learning model, monitor the learning curve during training and perform early stopping calculations, use an adaptive learning rate strategy to optimize the parameter update speed, use a validation set to evaluate model performance, analyze the performance of the deep learning model in different situations, and adjust network parameters and architectures in a timely manner, such as the number of layers, convolution kernel size, etc.; Among them, an adaptive learning rate strategy is used to optimize the parameter update speed. An initial learning rate is set. In each training round, the algorithm automatically calculates and updates the learning rate based on the current loss and gradient changes: if the loss improves, the learning rate is increased to speed up the convergence; if the loss worsens, the learning rate is reduced to avoid model oscillation; according to the progress of training, a learning rate decay strategy is set (for example, the learning rate is reduced every several rounds) to adjust the weights more finely in the later stage of training; S223: Evaluate the performance of the deep learning model on an independent test set, using standard metrics including accuracy, recall, and F1 score for quantitative evaluation; draw ROC curves to analyze the performance of the deep learning model on different cable states.

[0044] In step S221 of this embodiment, data preprocessing and feature extraction are performed to clean up historical data from acoustic, laser and fiber optic sensors through denoising, reduce random noise in the signal, improve data quality, and ensure the accuracy of subsequent analysis; through timestamp-based synchronization, ensure that the data collected by different sensors reflect the real situation at the same time, and avoid misjudgment caused by time differences; use geographic coordinates for spatial matching to ensure that different data are accurately located in physical space, thereby providing reliable basic data for subsequent analysis. The design of multi-layer convolution and pooling modules can effectively extract local features of data, and reduce feature dimensions and computational complexity through pooling layers, while enhancing the abstraction ability of features; feature extraction and fusion modules enhance the applicability of the model to complex seabed environments by aggregating data features from different modalities. This fusion enables the model to take advantage of the advantages of different sensors and extract more comprehensive information; by introducing GNN, the spatial relationship between features is captured to ensure that the mutual influence between the cable and the surrounding environment is fully considered. This design is particularly suitable for modeling graph structure data or spatial features to enhance the expression ability of the model. Significance achieved: Effective data preprocessing and feature extraction help improve the performance of the model and ensure that it can produce accurate results in complex tasks (such as cable status prediction); feature fusion and spatial relationship modeling enable the model to adapt to different situations in a changing seabed environment, improving its adaptability and robustness. Step S222 Model training and parameter optimization, introduce regularization terms (such as L1 or L2) in the loss function to prevent overfitting, ensure that the model not only performs well on the training set, but also maintains efficient prediction capabilities on unknown data; monitor the learning curve during the training process to judge the training effect of the model, use early stopping calculation to avoid overfitting, and ensure that the model stops training in the best state; use adaptive learning rate to optimize the parameter update speed, which can automatically adjust the learning rate so that it can advance quickly when converging during the model training process, while preventing invalid updates in the oscillation area; use the validation set to evaluate the model performance and help adjust the hyperparameters (such as the number of layers, convolution kernel size, etc.) so that the model has better adaptability and performance in different situations. Significance achieved: Through regularization and early stopping calculation, the training effect of the model is ensured to have better generalization ability, reduce dependence on specific samples, and thus improve the reliability of the prediction results; the adaptive learning rate and verification mechanism make the training process more efficient, reduce unnecessary computing burden, and ensure the best model effect under limited resources.Step S223: Evaluation and analysis of the model. The model performance is evaluated on an independent test set, which provides an objective indicator to judge the actual application effect of the model and ensures that the model can operate reliably in a real environment. The model is quantitatively evaluated using standard indicators such as accuracy, recall, and F1 score. These indicators provide the performance of the model in different categories (such as cable status), which helps to understand the advantages and disadvantages of the model. By drawing the ROC curve, the performance of the model under different classification thresholds can be analyzed to measure the trade-off between sensitivity and specificity, which helps to select the best classification threshold to improve the quality of decision-making. Significance achieved: Through performance evaluation on an independent test set, solid data support can be provided for the feasibility of the actual application of the model, and the user's trust can be enhanced. The analysis of standard indicators and ROC curves provides a basis for the optimization and adjustment of the model, so that subsequent work can be targeted and adjusted based on actual performance to improve decision-making efficiency.

[0045] In summary, the training process of the entire deep learning model in this embodiment achieves efficient and accurate cable status assessment through three steps: data preprocessing, model training and evaluation. Each step has its specific technical effect, which not only improves the accuracy and robustness of the model, but also provides a reliable guarantee for subsequent practical applications. Through these steps, the deep learning model can effectively cope with the challenges brought by the complex submarine environment and provide practical support for cable maintenance and monitoring.

[0046] Furthermore, the process of monitoring the learning curve during the training process and performing early stopping calculation in step S222 specifically includes the following steps: S2221: At the end of each training round, record the current training loss, validation loss and other related indicators, use data visualization tools to draw learning curves in real time, and show the changing trends of training set and validation set losses; Judging the state of the model based on the learning curve: Normal convergence: both training loss and validation loss decrease and are relatively close; Underfitting: Both training loss and validation loss are high; Overfitting: The training loss decreases while the validation loss increases, indicating that the model begins to "memorize" the training data instead of learning; S2222: Set an early stopping threshold to decide when to stop training; at the end of each round, check the change in validation loss. If the validation loss does not improve within the set rounds, enter the early stopping mechanism; S2223: Verify that the loss reaches a minimum value and save the current deep learning model parameters; meet the early stopping condition and terminate the training.

[0047] Step S2221 of this embodiment records and visualizes the learning curve, and realizes real-time monitoring of model performance by recording the training loss, validation loss and other relevant indicators (such as accuracy, precision, recall, etc.) of each training round; uses data visualization tools (such as Matplotlib, TensorBoard, etc.) to vividly display the changing trend of the training set and validation set losses to form a learning curve; according to the learning curve, the training status of the model can be quickly judged to determine whether the model is normally converged, underfitting or overfitting. Significance achieved: Visualizing the learning curve makes the training process more transparent, allowing researchers and engineers to better understand the model performance and training dynamics; by quickly identifying the training status, it allows timely adjustment of the training strategy, such as modifying the model structure or hyperparameters to optimize performance. Step S2222 sets the early stopping threshold, and by setting a specific early stopping threshold (for example, the validation loss has not improved within a certain round), provides a clear standard to guide when the model training is terminated; after each training round, the change of the validation loss should be checked to determine whether the early stopping mechanism should be entered. Significance achieved: By setting up an early stopping mechanism, the model is prevented from overfitting on the training set, thereby maintaining its generalization ability and ensuring that the performance on unseen data is also reliable; clear stopping conditions avoid unnecessary calculations and save training time and resources, especially when the training process of deep learning models usually consumes a lot of computing resources. Step S2223 saves the best model and terminates training. Once the verification loss reaches the minimum value, the current model parameters will be saved to ensure the best results for subsequent use; once the early stopping condition is met, the training will be terminated to avoid overfitting that may be caused by continued training of the model on the training set. Significance achieved: By saving the best model parameters, it is ensured that the final model has good performance, which can bring higher accuracy and reliability in the inference stage; terminating unnecessary training steps not only reduces the computational overhead, but also helps the training iteration process of the model to be more efficient, thereby accelerating the entire R&D cycle.

[0048] In summary, the three steps of this embodiment are indispensable in monitoring the learning process and performing early stopping calculations. They complement each other to ensure that the model can stop training in the best state, reduce the risk of overfitting, improve the generalization ability of the model, and effectively utilize computing resources. Through real-time monitoring and feedback mechanisms, researchers can adjust training strategies more flexibly to cope with complex deep learning tasks. This not only improves the performance of the model, but also provides a reliable basic support for subsequent practical applications.

[0049] Furthermore, the process of converting the extracted cable features into a three-dimensional visualization model in step S23 specifically includes the following steps: S231: Extract key features of the cable from the output of the deep learning model, such as the location (coordinates), shape (e.g., line segment, curve, etc.), depth (height relative to the seabed), and status information (e.g., whether there is damage, wear, etc.) of the cable; integrate the extracted features with relevant environmental data (e.g., terrain, obstacles, etc.) to create a data structure that associates all features and environmental factors in the system; S232: Create a three-dimensional coordinate system based on the acquired cable position data, construct the geometric shape of the cable using a grid model, generate curves based on the extracted cable shape features, and generate a volume model of the cable; integrate environmental features into the three-dimensional model; S233: Use visualization software to add appropriate textures and material properties to the cable and environmental models; based on the visualization model, combined with the extracted cable status information and environmental characteristics, conduct risk assessment and status analysis to identify potential risk factors.

[0050] Step S231 of this embodiment extracts and integrates cable features and environmental data, extracts key features of the cable from the deep learning model, including the location, shape, depth and status information of the cable; the accurate extraction of these features is the basis for subsequent visualization; the cable features are integrated with relevant environmental data (such as terrain, obstacles, etc.) to create a comprehensive data structure, from which all features and environmental factors in the system can be extracted. Significance achieved: By integrating data from different sources into a unified data structure, the availability and accessibility of the data are improved, which is helpful for subsequent data analysis; the integrated features and environmental information ensure that the operating status and surrounding environment of the cable can be fully reflected in the visualization process, enhancing the accuracy of the assessment; the association between features and environmental data makes subsequent analysis and assessment more coherent, and supports context-based risk judgment and decision-making. Step S232 creates a three-dimensional coordinate system and constructs geometric shapes. According to the extracted cable position data, a complete three-dimensional coordinate system is constructed to ensure that the relative positions of all models are correct and reflect the actual seabed environment; the mesh model and curve generation tool are used to generate corresponding curves and volume models according to the shape characteristics of the cable to show the geometric appearance of the cable; the environmental features (terrain and obstacles, etc.) are integrated into the three-dimensional model to enhance the realism and integrity of the model. Significance achieved: By establishing a three-dimensional model, the actual position and shape of the cable on the seabed can be reproduced more realistically, which improves the intuitiveness of visualization; the three-dimensional model provides an effective tool for subsequent data analysis, supporting users to observe the cable and its surrounding environment from multiple angles and deeply understand potential risks; through spatial matching, the relationship between the cable and the surrounding environment can be better identified, and the comprehensive understanding of the seabed environment can be enhanced, which is convenient for risk assessment. Step S233 Texture and material processing and risk assessment: Use visualization software to add appropriate textures and material properties to the cable and environmental models to enhance the visual effect and improve the realism of the model; Based on the visualization model and the extracted status information, risk assessment and status analysis are performed to identify potential risk factors (such as wear, surrounding obstacles, etc. that may affect the cable). Significance achieved: Appropriate texture and material processing make the visualization effect more realistic, easier for users to understand and judge, and improve user experience; risk assessment and status analysis are combined with visualization models to provide decision support, clearly show potential risks to users, and help formulate preventive measures or emergency response strategies; combined with information on cable status and environmental changes, the model can be dynamically updated, making risk assessment real-time and effective, and adapting to changing environments and conditions.

[0051] In summary, through the synergy of these three steps, the three-dimensional visualization model finally formed in this embodiment can truly reflect the actual status of the cable and the complexity of the surrounding environment. This not only helps engineers and decision makers better understand the current status of cable operation, but also effectively identifies and evaluates potential risks, thereby improving the efficiency and reliability of submarine cable management and maintenance. These steps ensure the scientificity and practicality of the entire process, making cable status monitoring and evaluation more accurate and efficient.

[0052] Furthermore, if Figure 4 As shown, the process of implementing the cable position marking in step S3 specifically includes the following steps: S31: Setting a moving mark to calibrate the analyzed cable position through the sea surface display system; Moving markers will be based on real-time extrapolated cable positions through interpolation; S32: Record all data during the position calibration process, including the real-time position and depth of the mobile marker and its relationship with the cable position, and feed back to the central monitoring system in real time; S33: Based on the results of all detection and position calibration, a detection report is automatically generated, including the specific location, depth and current status of the cable, with 3D visualization results and risk assessment information.

[0053] In step S31 of this embodiment, a mobile marker is set and a position calibration is performed. Through the sea surface display system, a mobile marker is dynamically set to indicate the analyzed cable position, so as to facilitate subsequent data processing and monitoring; the mobile marker combines the real-time estimated cable position with known data through interpolation methods (such as linear interpolation, curve interpolation, etc.), so as to accurately determine the development path and specific position of the cable underwater. Significance achieved: Through real-time calculation and dynamic marking, the accuracy of cable position calibration can be improved, especially in complex underwater environments, the dynamic changes of cables can be better captured; the setting of mobile markers enables the system to adjust according to real-time data, provide flexibility, and adapt to changing hydrological environments and working conditions; accurate cable position calibration reduces the Asian Games cost during subsequent processing and maintenance, and improves overall monitoring efficiency. Step S32 records data and real-time feedback, and records in detail all relevant data in the position calibration process, including the real-time position and depth of the mobile marker, and its real-time relationship with the cable position; the real-time recorded data is fed back to the central system for monitoring and further analysis to ensure that relevant personnel can obtain the latest data and status information. Significance achieved: Recording the position calibration data of each step ensures that the operator can fully understand the status of the cable and improve the authenticity and reliability of monitoring; the real-time recording mechanism makes historical data traceable, provides a basis for subsequent analysis, comparison and decision-making, and supports continuous monitoring and maintenance; real-time feedback to the monitoring system enables operators to grasp the changes in the cable position in a timely manner, respond quickly, and reduce potential risks. Step S33 generates a detection report. According to the recorded detection and position calibration results, a detection report containing the specific location, depth and current status of the cable is automatically generated; the detection report also includes three-dimensional visualization results, intuitively displays the shape and location of the cable, and provides risk assessment information to help operators fully understand the cable status and environmental risks. Significance achieved: The automatically generated report will help to organize and summarize information, so that operators can quickly and accurately understand the cable status without manually arranging data; the comprehensive three-dimensional visualization results and risk assessment provide intuitive visualization information, support decision makers to make more informed decisions in maintenance, operation, etc., and improve safety; the detection report can display the results of cable monitoring in a structured manner, which is convenient for sharing with internal or external stakeholders (such as regulators, customers, etc.) within the team to improve communication efficiency.

[0054] In summary, through the implementation of these three steps, this embodiment not only realizes the cable position calibration, but also ensures the accurate monitoring and management of the cable and provides data support for subsequent maintenance. Through the effective combination of dynamic calibration, real-time recording and report generation, the system can better deal with potential risks, improve the efficiency and safety of cable management, and promote the modernization and intelligence of the entire submarine cable industry.

[0055] like Figure 5 As shown, this embodiment provides an embodiment of an electronic device. In this embodiment, the electronic device 1 includes a processor 11 and a memory 12 coupled to the processor 11.

[0056] The memory 12 stores program instructions for implementing a method for nearshore detection of a submarine cable in any of the above embodiments.

[0057] The processor 11 is used to execute program instructions stored in the memory 12 to perform standard operations for nearshore detection of submarine cables.

[0058] The processor 11 may also be referred to as a CPU (Central Processing Unit). The processor 11 may be an integrated circuit chip having the ability to process signals. The processor 11 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0059] Further, Figure 6 The schematic diagram of the structure of the storage medium of an embodiment of the present application is that the storage medium 2 of the embodiment of the present application stores program instructions 21 capable of implementing all the above methods, wherein the program instructions 21 can be stored in the above storage medium in the form of a software product, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or terminal devices such as a computer, a server, a mobile phone, and a tablet.

[0060] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0061] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of software functional units. The above is only an implementation mode of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly used in other related technical fields, is also included in the patent protection scope of the present invention.

[0062] The specific implementation methods of the invention are described in detail above, but they are only examples, and the invention is not limited to the specific implementation methods described above. For those skilled in the art, any equivalent modification or substitution of the invention is also within the scope of the invention, therefore, the equalization, modification, improvement, etc. made without departing from the spirit and principle of the invention should be included in the scope of the invention.

Claims

1. A method for nearshore detection of submarine cables, characterized in that: The method for nearshore detection of submarine cables comprises: Collect historical submarine cable laying information, geological data and hydrological data in the nearshore area, and conduct preliminary analysis through the geographic information system; deploy a variety of detection equipment to obtain acoustic detection data, laser imaging data and fiber optic sensor data of submarine cables; The acoustic detection data, laser imaging data and fiber optic sensor data are integrated to obtain a fused data set, which is then input into a trained deep learning model to obtain cable layout, depth and status information; a three-dimensional visualization model is constructed based on the obtained cable layout, depth and status information to evaluate the seabed environment and cable status; Through the sea surface display system, mobile markers are placed on the cable and nearby areas to calibrate the cable position; based on the detection and analysis results, a detection report is generated, including the cable position, depth and status.

2. A method for nearshore detection of submarine cables according to claim 1, characterized in that: The preliminary analysis through geographic information system includes the following steps: Identify various available data sources, including government databases, industry reports, scientific research papers, and historical survey databases; obtain relevant historical submarine cable deployment information from each identified data source, as well as geological and hydrological data in the area, including cable type, deployment time, deployment depth, and surrounding soil, rock type, water velocity, and wave height; Standardize all data from the acquired data sources, import the collected data into the geographic information system, and store the data in the spatial database; generate heat maps, perform buffer analysis, and relative position analysis to identify the relationship between cable layout and geographic features; analyze the distribution pattern of historical cable layout and predict the geographic areas and risk factors that may be associated with future cable layout; Through the relative position and distribution pattern of cable laying, various detection equipment are deployed, and acoustic sensors, laser imagers and fiber optic sensor equipment are deployed to the predetermined seabed locations.

3. A method for nearshore detection of submarine cables according to claim 1, characterized in that: Assess the submarine environment and cable status, including the following steps: De-noise the acoustic detection data, laser imaging data, and fiber optic sensor data, convert all types of data into a unified format, synchronize the timestamps of data from different sources, and spatially match the data from different sensors based on geographic coordinates; select a data fusion network, synthesize the data, and obtain a fused data set that contains information on the location, shape, and status of the cable; Select a convolutional neural network, use historical data to train the deep learning model, and predict the layout, depth, and status of the cable based on the input data; input the fused data set into the trained deep learning model, and after reasoning, output the location, depth, and status information of the cable in the form of probability distribution or confidence, indicating the most likely cable characteristics; The cable layout, depth and status features are extracted from the deep learning model output. Using visualization tools, the extracted cable features are converted into a three-dimensional visualization model. The cable's location, shape, depth and corresponding environmental characteristics are all represented. Combined with the three-dimensional visualization model, an in-depth analysis is conducted to evaluate the seabed environment, potential risks and the operating status of the cable.

4. A method for nearshore detection of submarine cables according to claim 3, characterized in that: The data fusion network synthesizes the data, including the following steps: Establish a multimodal input layer of acoustic data, laser imaging data and optical fiber data to retain the time series characteristics of acoustic signals, the spatial characteristics of laser imaging, and the characteristic types of data types of continuous monitoring data of optical fiber sensing; Construct multiple parallel feature extraction networks to automatically extract meaningful features from each type of data, and use the attention mechanism to weight the features so that the attention mechanism can adaptively focus on important information; use the message passing network to perform multi-level feature fusion after feature extraction, combining low-level features with high-level features to maintain contextual relationships; Through the dynamic modeling mechanism, the dynamic model parameters are adjusted through real-time feedback so that it can be re-learned when new data is collected; the fused output will be processed by a graph neural network to capture the relationship between the cable and environmental characteristics, connect each feature with its spatial position, and output the cable position, depth and status information.

5. A method for nearshore detection of submarine cables according to claim 4, characterized in that: Multi-level feature fusion includes the following steps: The low-level and high-level features obtained by the feature extraction network are embedded in the space mapping, the features are mapped to a low-dimensional space through a linear layer, and then a nonlinear transformation is applied for enhancement; Build a graph model that treats features at different levels as nodes in a graph. The connections between features represent their relationships. Each node interacts with its neighbor nodes through a message passing mechanism to learn deeper feature relationships and contextual information. During the fusion process, an adaptive weighting mechanism is applied to dynamically adjust the weights according to the importance of the features and the context. More important features occupy a larger proportion in the final fusion result. The weight calculation is based on automatic learning based on the attention mechanism.

6. A method for nearshore detection of submarine cables according to claim 5, characterized in that: Building a graph model includes the following steps: The acoustic features, laser imaging features and optical fiber sensing features of multi-level feature fusion correspond to one node; the nodes are layered according to the level of the features; Calculate the similarity between different features through distance measurement, and use similarity to build edges; assign attributes to nodes and edges. Node attributes include: feature vector, corresponding to the feature expression of the node; hierarchical information, identifying the hierarchy of features; edge attributes include: edge weight, connection strength or similarity; edge type, indicating what type of relationship; Use the graph processing library to build and manage the graph structure, create an empty graph, add nodes, and add edges; apply the message passing mechanism to interact with information in the graph model, and each node will update its own status based on the information of its neighbor nodes; the node receives information from the neighbor nodes through the edges it connects, and calculates the new feature representation of each node according to the message update rules.

7. A method for nearshore detection of submarine cables according to claim 4, characterized in that: The dynamic modeling mechanism includes the following steps: Establish a real-time feedback system to monitor the performance of the model when new data is input. By evaluating the deviation between the output results and the actual situation in real time, feedback signals are automatically generated to guide the adjustment of model parameters. Regularization techniques are used to impose constraints on model parameters so that they remain balanced when updated. Based on real-time feedback information, the fusion weights of each modal feature are adaptively adjusted; node features are dynamically updated according to the input new data, and during the updating process, the weights and relationships of the edges are dynamically adjusted.

8. A method for nearshore detection of submarine cables according to claim 3, characterized in that: The training of deep learning models includes the following steps: De-noise historical data from acoustic detection, laser imaging, and fiber optic sensing, synchronize data from different sensors based on timestamps, and spatially match data using geographic coordinates; Introduce regularization terms in the loss function, use historical data to train the deep learning model, monitor the learning curve during training and perform early stopping calculations, use adaptive learning rate strategies to optimize parameter update speed, use validation sets to evaluate model performance, analyze the performance of deep learning models in different situations, and adjust network parameters and architectures, number of layers, and convolution kernel size in a timely manner; The performance of the deep learning model was evaluated on an independent test set, and standard indicators including accuracy, recall, and F1 score were used for quantitative evaluation. The ROC curve was drawn to analyze the performance of the deep learning model on different cable states.

9. A method for nearshore detection of submarine cables according to claim 8, characterized in that: Monitoring the learning curve during training and performing early stopping calculations include the following steps: At the end of each training round, record the current training loss, validation loss, and other related indicators, and use data visualization tools to draw learning curves in real time to show the changing trends of the training set and validation set losses; Set an early stopping threshold to decide when to stop training; at the end of each round, check the change in validation loss. If the validation loss does not improve within the set rounds, enter the early stopping mechanism; Verify that the loss reaches a minimum value and save the current deep learning model parameters; if the early stopping condition is met, terminate the training.

10. A method for nearshore detection of submarine cables according to claim 1, characterized in that: The cable location marking includes the following steps: Set moving marks through the sea surface display system to calibrate the analyzed cable positions; By interpolation, the moving marker is adjusted according to the real-time estimated cable position; Record all data during the position calibration process, including the real-time position and depth of the mobile marker and its relationship with the cable position, and feed it back to the central monitoring system in real time; Based on the results of all detection and position calibration, a detection report is automatically generated, including the specific location, depth and current status of the cable, with 3D visualization results and risk assessment information.

Citation Information

Patent Citations

  • Method and device for detecting buried depth of submarine cable in shallow water area

    CN113640808A

  • Submarine cable path detection device and method and cable fault positioning method

    CN114675332A

  • Offshore submarine cable positioning method based on USBL and SINS / GPS system

    CN117406168A

Cited By

  • Optical cable visualization online monitoring method and system based on 5G communication

    CN119382790A

  • Expendable submarine sediment acoustic characteristic in-situ measuring instrument

    CN119414473A

  • Submarine cable acousto-optic monitoring system and submarine cable monitoring method

    CN119471704A

  • Ship navigation environment information acquisition system

    CN119551161A

  • Penetrating end cable wear prediction method and system based on multi-physics field coupling simulation

    CN119849266A