An Anti-interference Segmentation Method for Exposed Subsea Pipelines Based on Multibeam Bathymetry Data

Abstract

This invention relates to an anti-interference segmentation method for exposed subsea pipelines based on multibeam bathymetry data, belonging to the field of marine engineering exploration technology. The method includes: calculating the depth gradient magnitude and elevation anomaly of preprocessed multibeam bathymetry data to select candidate points with pipeline characteristics, forming a candidate region set; calculating the micro-scale surface features, macro-scale profile morphology features, and relative elevation features of the candidate region set to construct feature vectors; inputting the feature vectors into a gradient boosting decision tree model to output a refined segmentation result for the exposed pipeline; and inputting the refined segmentation result into a trained bi-branch feature fusion network to obtain an optimized segmentation result. This invention effectively solves the problem of distinguishing exposed pipelines from various interference objects in complex seabed environments by combining multi-scale geometric morphological analysis, physically guided deep learning, and multi-source geographic information verification, significantly improving the accuracy, robustness, and engineering practicality of the segmentation.

Description

Technical Field

[0001] This invention relates to the field of marine engineering exploration technology, specifically to a method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data. Background Technology

[0002] Submarine pipelines, as the core facilities for transporting marine oil and gas resources, are hailed as the lifeline of marine oil and gas transportation. Due to their long-term exposure to the complex marine environment, submarine pipelines are susceptible to various factors such as ocean currents, geological changes, fishing activities, and ship anchor damage, which can lead to defects such as leakage, overhang, and corrosion. These defects pose significant safety hazards to the safe transportation of marine oil and gas, and also constitute serious environmental risks to the marine ecosystem.

[0003] Multibeam bathymetry (MBB) systems, with their high efficiency, can acquire high-precision, high-density three-dimensional topographic and acoustic imagery of the seabed, making them a core technology for pipeline condition surveying. However, in actual pipeline routing areas, various interfering objects are often densely distributed, such as abandoned fishing nets, construction trenches, reefs, and natural sand waves. These interfering objects are highly similar to the characteristics of exposed pipelines in terms of morphology and physical response: for example, abandoned fishing nets appear as discrete mesh-like reflections in acoustic images, trenches exhibit linear concave features, and reefs are mostly point-like strong reflectors. This complex interference leads to traditional segmentation methods (such as global thresholding, edge gradient, etc.) that rely on single features (such as global thresholding, region growing, edge detection, etc.) generally facing extremely high false detection rates, making it difficult to meet the urgent need for accurate identification in engineering applications.

[0004] Existing pipeline inspection methods based on multibeam data have several significant shortcomings, as follows:

[0005] 1. Severe interference from various objects: Within the pipeline routing area, there are various objects causing interference, such as abandoned fishing nets, pipe trenches, reefs, and natural sand waves. Among them, the acoustic reflection of abandoned fishing nets is discrete and net-like, pipe trenches are linear depressions, and reefs are point-like strong reflectors. The characteristics of these objects in terms of depth and intensity data are very similar to those of the exposed pipeline. This leads to a very high false detection rate for traditional segmentation methods, such as global thresholding, region growing, and edge detection.

[0006] 2. Poor feature robustness: Suspended matter in the water, noise inherent in the data itself, and complex seabed topography can all affect pipeline point cloud data, often resulting in blurred, sparse, or locally fragmented features. Traditional algorithms are poorly adapted to these situations, leading to discontinuous segmentation results and severe missed detections.

[0007] 3. Insufficient automation and generalization capabilities: Traditional algorithms and early machine learning models heavily rely on manual feature design and parameter tuning during application. These methods have poor universality for different sea areas, seabed conditions, and pipelines of different diameters, resulting in low automation, requiring significant manual intervention, and low detection efficiency. Summary of the Invention

[0008] To address the aforementioned problems, the purpose of this invention is to provide a method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data.

[0009] A method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data includes:

[0010] Step 1: Acquire multibeam bathymetry data of the target seabed; the multibeam bathymetry data includes raw depth data and raw echo intensity data;

[0011] Step 2: Preprocess the multibeam echo sounding data to obtain preprocessed multibeam echo sounding data;

[0012] Step 3: Calculate the depth gradient magnitude and elevation anomaly of the preprocessed multibeam echo sounding data, and select candidate points with pipeline characteristics to form a candidate region set;

[0013] Step 4: Calculate the microscale surface features, macroscale profile morphology features, and relative elevation features of the pipeline candidate region on the seabed to construct feature vectors;

[0014] Step 5: Input the feature vector into the gradient boosting decision tree model to output the refined segmentation result of the pipeline leakage;

[0015] Step 6: Train the dual-branch feature fusion network using labeled data to obtain the trained model; wherein, the main branch of the dual-branch feature fusion network adopts a U-Net structure, and the auxiliary branch adopts a lightweight network, and the two perform feature fusion in the decoder;

[0016] Step 7: Input the refined segmentation results into the trained model to obtain the optimized segmentation results.

[0017] Preferably, step 2: preprocessing the multibeam bathymetry data to obtain preprocessed multibeam bathymetry data includes:

[0018] Step 2.1: Correct and denoise the original depth data to obtain corrected depth data; the correction process is as follows:

[0019]

[0020] in, For the corrected depth, Original depth As the reference speed of sound, This is the measured sound velocity profile;

[0021] Step 2.2: Perform adaptive normalization and Gaussian denoising on the original echo intensity data to obtain the normalized echo intensity; wherein, the adaptive normalization process is as follows:

[0022]

[0023] in, The normalized echo intensity The original echo intensity, For the local minimum echo intensity, For the local maximum echo intensity, and This is the contrast stretching parameter.

[0024] Preferably, step 3: calculating the depth gradient magnitude and elevation anomaly of the preprocessed multibeam echo sounding data, and selecting candidate points with pipeline characteristics to form a candidate region set, includes:

[0025] Step 3.1: Calculate the partial derivatives of the corrected depth data in the horizontal and vertical directions, and take the Euclidean modulus of the partial derivatives in both directions to obtain the depth gradient magnitude.

[0026] Step 3.2: Use the least squares method to fit the local seabed background surface, and subtract the depth value of each data point of the corrected depth data from the corresponding local seabed background surface to obtain the elevation anomaly;

[0027] Step 3.3: Using the depth gradient amplitude, elevation anomaly, and normalized echo intensity, establish joint judgment conditions to screen out candidate points with pipeline characteristics and form a binary candidate map;

[0028] Step 3.4: Use morphological closing operations to process the binary candidate image to obtain a candidate region set.

[0029] Preferably, in step 3.3, the joint determination condition is:

[0030]

[0031] in, For gradient threshold, The range of intensity thresholds. Elevation threshold For depth gradient magnitude, To normalize the echo intensity, This indicates an elevation anomaly.

[0032] Preferably, in step 4, the microscale surface features include: surface curvature distribution features and the angular deviation between the neighborhood normal vector and the average normal vector; the macroscale profile morphology features include: profile symmetry index and profile Gaussian curvature integral; the relative elevation features of the pipeline candidate region on the seabed are:

[0033]

[0034] in, This refers to the actual elevation value of the pipeline candidate point or pipeline candidate area on the seabed. This represents the fitted local seabed background surface. Indicates relative elevation.

[0035] Preferably, in step 6, a convolutional block attention module is embedded in the dual-branch feature fusion network, and attention weights are calculated sequentially in the channel and spatial dimensions to extract pipeline features; wherein, the calculation method of the convolutional block attention module is as follows:

[0036]

[0037] in, This represents the feature map optimized by the Convolutional Block Attention Module (CBAM), which fuses channel and spatial attention weights to focus on key pipeline features. This represents a channel attention mechanism that calculates weights for each channel dimension of the feature map, highlighting channel features valuable for pipeline segmentation. This represents a spatial attention mechanism that calculates weights for the spatial dimensions of the feature map, focuses on the spatial region where the pipeline is located, and suppresses background / interference objects.

[0038] Preferably, during training, the loss function is:

[0039]

[0040] in, , , For the weights of the loss function, Represents the dice loss function. Represents the focus loss function. This represents the geometric consistency loss.

[0041] Preferably, after step 7, the method further includes:

[0042] The optimized segmentation results and their corresponding geographical locations are transformed to the geodetic coordinate system. The shortest perpendicular distance between each connected component in the optimized segmentation results and the pipeline design route is calculated, and the segmentation results with the shortest perpendicular distance within the preset tolerance threshold are retained. The AI-optimized segmentation results are compared and analyzed with historical detection data to calculate the overlap rate and morphological similarity between the two. The pipeline bare leakage sections with both overlap rate and morphological similarity greater than the preset threshold are used as the final pipeline bare leakage distribution map.

[0043] The present invention also provides an electronic device, including a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor. The transceiver, the memory, and the processor are connected via the bus. The computer program, when executed by the processor, implements the steps in the above-described method for anti-interference segmentation of exposed submarine pipelines based on multibeam echo sounding data.

[0044] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps in the above-described method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data.

[0045] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0046] This invention relates to an anti-interference segmentation method for exposed subsea pipelines based on multibeam echo sounding data. Compared with existing technologies, this invention constructs a cascaded technical framework of "data preprocessing → coarse segmentation → geometric fine segmentation → deep learning optimization → geographic information verification," which deeply integrates multi-scale geometric morphology analysis, physical-guided deep learning, and multi-source geographic information. This results in a false positive rate of less than 3%, 5%, and 2% for fishing nets, pipe trenches, and reefs, respectively, and an overall segmentation accuracy rate of over 95%, effectively solving the problem of false detection caused by interference in existing methods.

[0047] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1A flowchart of an anti-interference segmentation method for exposed subsea pipelines based on multibeam bathymetry data provided by this invention;

[0050] Figure 2 A schematic diagram of a method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data provided by the present invention;

[0051] Figure 3 The images provided by this invention are a comparison of the effects of pipeline leakage detection, where the green image is a multi-beam sonar image and the orange image is a side-scan sonar image. Detailed Implementation

[0052] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0053] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0054] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0055] Please see Figure 1 An anti-interference segmentation method for exposed subsea pipelines based on multibeam bathymetry data, comprising:

[0056] Step 1: Acquire multibeam bathymetry data of the target seabed; the multibeam bathymetry data includes raw depth data and raw echo intensity data;

[0057] Step 2: Preprocess the multibeam echo sounding data to obtain preprocessed multibeam echo sounding data;

[0058] Step 3: Calculate the depth gradient magnitude and elevation anomaly of the preprocessed multibeam echo sounding data, and select candidate points with pipeline characteristics to form a candidate region set;

[0059] Step 4: Calculate the microscale surface features, macroscale profile morphology features, and relative elevation features of the pipeline candidate region on the seabed to construct feature vectors;

[0060] Step 5: Input the feature vector into the gradient boosting decision tree model to output the refined segmentation result of the pipeline leakage;

[0061] Step 6: Train the dual-branch feature fusion network using labeled data to obtain the trained model; wherein, the main branch of the dual-branch feature fusion network adopts a U-Net structure, and the auxiliary branch adopts a lightweight network, and the two perform feature fusion in the decoder;

[0062] Step 7: Input the refined segmentation results into the trained model to obtain the optimized segmentation results.

[0063] Please see Figure 2 The principles of the present invention will be explained below with reference to specific embodiments:

[0064] I. Preprocessing and Feature Enhancement.

[0065] This step is fundamental to all subsequent processing and aims to improve the quality of the raw data and enhance the characteristics of the pipeline target, creating favorable conditions for subsequent segmentation. The inputs for this step include: raw depth data and echo intensity data directly acquired by the multibeam echo sounder.

[0066] The core operations include the following three:

[0067] (1) Sound velocity profile depth correction: Since the sound velocity in seawater varies with depth, temperature, and salinity, the original depth data may contain errors. This invention utilizes field-measured sound velocity profiles (CTD data) to uniformly correct the original depth data to the depth corresponding to a preset reference sound velocity. The calculation formula is as follows:

[0068]

[0069] in, For the corrected depth, Original depth The reference speed of sound is 1500 m / s. This is a sound velocity profile measured by a CTD instrument. This formula can be used to correct the raw depth data to the depth corresponding to the reference sound velocity, improving the accuracy of the depth data.

[0070] (2) Adaptive normalization of echo intensity: To overcome the problem of large differences in echo intensity in different regions, local adaptive normalization is performed on the denoised echo intensity data. This method performs contrast stretching based on the statistical characteristics (such as minimum and maximum values) in the local neighborhood of each pixel, making the acoustic reflection features of the pipe target more prominent in the image and forming a sharp contrast with the background.

[0071] The calculation formula is:

[0072]

[0073] in, The normalized echo intensity The original echo intensity, For the local minimum echo intensity, For the local maximum echo intensity, and This is a contrast stretching parameter. Contrast stretching is performed using local statistics to make the echo intensity characteristics of the target area more prominent.

[0074] (3) Gaussian filtering for noise removal: Noise is inevitably present in multibeam echo sounding data, and Gaussian filtering is required for noise removal. This method adaptively adjusts the direction and shape of the Gaussian filter kernel according to the local gradient direction, effectively suppressing noise while preserving key features such as pipe edges to the maximum extent, and avoiding the loss of pipe features due to filtering operations.

[0075] Through this step, the raw, noisy, and biased data is transformed into high-precision 3D seabed point clouds and feature-enhanced acoustic images, laying a solid foundation for accurate segmentation.

[0076] II. Coarse segmentation - rapid location of candidate regions.

[0077] This step aims to quickly identify all possible pipeline candidate regions from massive amounts of data, ensuring high recall and laying the foundation for subsequent fine-grained segmentation. The input for this step is preprocessed and feature-enhanced augmented data.

[0078] The core operations include the following three:

[0079] (1) Multi-feature joint extraction: Simultaneous calculation of three key features.

[0080] Depth gradient magnitude Used to quantify the drastic spatial variation of seabed depth, highlighting the difference between the edge of a channel (where depth changes abruptly) and the smooth background. It can reflect the severity of depth changes.

[0081] Normalized echo intensity The output of preprocessing and feature enhancement is used directly to highlight the intensity features of the target region.

[0082] Elevation Anomaly The background surface is obtained by fitting local curved surfaces, and the elevation anomaly of data points relative to the background surface is calculated. This is used to distinguish between exposed pipelines (bulges) and depression-type interference (such as pipe trenches). The elevation deviation of the target relative to the background can be quantified, which can reflect the degree of protrusion of the pipeline relative to the seabed background.

[0083] This will be described in detail in the following examples. and The specific calculation process.

[0084] (2) Adaptive threshold segmentation.

[0085] A joint decision criterion is set to filter each data point and generate a binary candidate image. The joint decision criterion is:

[0086]

[0087] in, Indicates whether the point at coordinates (x, y) is a candidate point for the pipeline: a value of 1 indicates that it is a candidate point, and 0 indicates that it is not. It is determined by a combination of gradient, intensity, and elevation thresholds. For gradient threshold, The range of intensity thresholds. The elevation threshold is used. This joint decision criterion allows for the selection of candidate points with pipeline characteristics, forming a binary candidate map.

[0088] (3) Morphological connectivity optimization. Due to factors such as noise, pipe segments in the binary candidate image may be interrupted or contain voids. Morphological closing operations (dilation followed by erosion) are used to process the binary candidate image, connecting pipe segments interrupted by noise, filling voids, and forming a set of connected candidate regions. .

[0089] This step quickly identified all "suspected targets" and eliminated most background areas, such as... Figure 2 The "coarse segmentation" module points to the arrow indicating "exclude most of the background." Its output is a significantly narrowed candidate region for subsequent fine processing, greatly improving the overall processing efficiency of the algorithm.

[0090] III. Refined segmentation - multi-scale geometric morphological filtering.

[0091] This step aims to accurately distinguish between actual exposed pipes and interfering elements, such as fishing nets, trenches, and rocks, introduced during coarse segmentation, from a geometric perspective. The input to this step is the set of candidate pipe regions output from the previous step. .

[0092] The core operation is multi-scale geometric feature calculation. For each candidate region, its detailed geometric features are calculated at different spatial scales.

[0093] (1) Microscopic surface feature analysis. Specifically, this includes:

[0094] (a) Surface curvature distribution. Specifically, based on PCA of the local neighborhood of the point cloud, the surface curvature is calculated:

[0095]

[0096] Data meaning: For the local neighborhood point cloud of each point in the pipeline candidate region set, the variance of the three-dimensional coordinates of the neighborhood points is decomposed by principal component analysis (PCA) to obtain three eigenvalues. (Corresponding to the three-axis dimensions of the ellipsoid), used to calculate surface curvature. PCA is a method for extracting microscopic geometric features of point clouds within a candidate region set, filtering out points whose curvature conforms to the smooth characteristics of a pipe, and eliminating interference points with abnormal curvature such as fishing nets / reefs, thus narrowing the precise segmentation range.

[0097] in The dimension (maximum variance) represents the direction of the longest axis of the ellipsoid. The dimension representing the direction of the secondary axis of the ellipsoid (moderate variance). The dimension (minimum variance) represents the direction of the shortest axis of the ellipsoid. The surface of the pipe is relatively smooth, and its curvature distribution is relatively concentrated; while the surfaces of interfering objects such as fishing nets and reefs are uneven, with drastic changes in curvature and a more dispersed distribution. The surface curvature distribution can effectively distinguish the pipe from some interfering objects.

[0098] (b) Normal Vector Consistency Measurement. Specifically, it calculates the angle deviation between the neighborhood normal vector and the average normal vector. The surface morphology of the pipe is regular, and the direction of its normal vector changes in a certain regularity. The value is relatively small; however, the surface morphology of the interfering object is irregular, and the direction of the normal vector is chaotic. The value is relatively large, and this feature can be used to further distinguish pipes from interference objects.

[0099] Data meaning: For each point in the candidate region set, calculate the angle between the normal vector of each point and the average normal vector of the neighborhood. The consistency of the neighborhood normal vectors is quantified; the direction of the pipe surface normal vectors is stable. Small, disruptive objects (such as reefs). Large. As a geometric screening criterion for the candidate region set, it is used to eliminate... Interference points exceeding the threshold are further refined to concentrate the effective pipeline points in the candidate region.

[0100] (2) Quantification of macroscopic-scale profile morphology. Specifically, this includes:

[0101] (a) Profile symmetry index. Specifically, extract vertical profiles along the pipeline route and calculate left-right symmetry. The cross-section of the pipe is approximately symmetrical. The value is close to 1; however, the cross-sectional asymmetry of the interfering object is relatively high. If the deviation from 1 is significant, the symmetry index of the cross-section can be used to distinguish between the pipeline and the interference.

[0102] (b) Integral of Gaussian curvature in the cross section. Its calculation formula is:

[0103]

[0104] in This represents the seabed profile extracted from multibeam data, i.e., depth (height) as a function of horizontal distance. The changing curve. The exposed, bulging pipe corresponds to... The corresponding recessed pipe trench This integral value can effectively distinguish between exposed pipes and pipe trenches.

[0105] (3) Local seabed context modeling. Specifically:

[0106] Robustly fitting the local seabed background surface using the RANSAC algorithm Calculate the relative elevation:

[0107]

[0108] in, This refers to the actual elevation of the pipeline candidate point or pipeline candidate area on the seabed. (Exposed pipeline) trench By combining relative elevation characteristics, areas of exposed pipelines can be further screened out.

[0109] (4) Integrated geometric classifier.

[0110] Construct feature vectors from the above multi-scale features. The input is a gradient boosting decision tree model, which can automatically learn the ability of different features to distinguish between exposed pipes and interfering objects, thereby automatically differentiating exposed pipes from categories such as fishing nets, trenches, and reefs, and outputting refined segmentation results. .

[0111] The output of this step is the refined segmentation result. It effectively eliminates most of the interfering elements such as fishing nets and pipe trenches that are difficult to distinguish in coarse segmentation. At this point, the accuracy of the segmentation results has been greatly improved.

[0112] IV. AI-optimized segmentation - deep learning intelligent identification.

[0113] This step aims to leverage the powerful feature learning capabilities of deep learning models to solve difficult targets with similar geometric features that are hard to distinguish, and to repair incomplete pipeline segments.

[0114] 1. The deep learning model's network architecture employs a dual-branch feature fusion network. The main branch is U-Net, which boasts excellent performance in pixel-level segmentation tasks, accurately segmenting target regions. The auxiliary branch is a lightweight network primarily used to extract global contextual features, providing more comprehensive information for the segmentation task. Both are fused at the decoder, fully combining local detail features with global contextual features to improve segmentation accuracy.

[0115] 2. Attention mechanism.

[0116] An embedded convolutional block attention module (CBAM) is used to calculate attention weights sequentially in the channel and spatial dimensions, enabling the model to focus on key features of the pipeline. The calculation method is as follows:

[0117]

[0118] in, The feature map F is processed to obtain a feature map F' optimized in the channel dimension, which can highlight the channel features that are important for the segmentation task; Further processing of F' further focuses the spatial area on the pipeline region, improving the model's ability to identify pipeline features.

[0119] (3) Physically guided loss function.

[0120] Using a composite loss function:

[0121]

[0122] in, , , These are the weights of the loss function. It is mainly used to optimize the overlap of segmentation regions, address the class imbalance problem caused by the small number of pixels in the pipeline, and ensure that the model can find and cover the entire pipeline region to prevent missed detections. The main focus is on difficult samples, avoiding easy samples dominating the training process, ensuring that the model can accurately distinguish between pipes and interference objects, preventing false detections, and improving boundary accuracy; As a geometric consistency loss, the segmentation results of the model are forced to satisfy the continuity of the pipeline and the constraints of cylindrical / regular geometry in three-dimensional space, and isolated, unfit fragments are penalized, so that the segmentation results are more consistent with the actual shape of the pipeline.

[0123] (4) Model training and reasoning.

[0124] The network is trained using labeled data. Regional data is input into the trained model to obtain an optimized segmentation probability map, which is then thresholded to obtain the AI-optimized segmentation result. .

[0125] The output of this step is the AI-optimized segmentation result. It can further distinguish difficult targets that are hard to determine by geometric morphology, and effectively repair pipe fractures, improving the continuity and integrity of the segmentation results, such as... Figure 2 The "AI Optimized Segmentation" module in the middle points to the arrow indicating "Distinguish Difficult Targets".

[0126] V. Location Verification - Geographic Information Fusion Confirmation.

[0127] This step is the final quality control stage. It utilizes multi-source geographic information to cross-validate the AI ​​segmentation results, ensuring the accuracy and historical authenticity of their geographical locations. The input to this step is the AI-optimized segmentation result output from the previous step. .

[0128] The core operations of this step include the following:

[0129] 1. Coordinate unification and registration. All data and segmentation results are uniformly converted to a standard geodetic coordinate system (such as WGS-84 UTM) to ensure the accuracy of the geographical location information of the segmentation results, providing precise location references for subsequent pipeline maintenance operations.

[0130] 2. Route consistency analysis. Calculation. The shortest perpendicular distance between each connected component and the designed pipeline route is used to retain results where the distance is within a tolerance threshold. This analysis can eliminate false segmentation regions that deviate too far from the designed pipeline route, thus improving the reliability of the segmentation results.

[0131] 3. Spatiotemporal continuity verification. By comparing with historical detection data, the overlap rate and morphological similarity are calculated, and persistent bare segments are assigned a high confidence level. This verification step can further confirm the authenticity of bare segments and avoid misjudgment due to anomalies in a single detection data.

[0132] Output: Generates a final map of exposed pipe leaks with precise geographic coordinates. It outputs quantitative statistical reports including information such as exposed length, exposed height, and geographical location, providing comprehensive and accurate data support for pipeline maintenance decisions.

[0133] The output of this step is the final, reliable "Final Pipeline Exposed Leakage Segmentation Result and Report." It integrates the analysis results of the inspection data with prior engineering geographic information, providing the most authoritative and reliable data support for subsea pipeline maintenance decisions.

[0134] The following example, using the inspection of an 813mm diameter subsea oil pipeline at the Suizhong terminal in Liaodong Bay, will be used to illustrate the invention in detail:

[0135] 1. Data Acquisition: The Reson Seabat T20-P multibeam echo sounder system was used, operating at a frequency of 400kHz and with a line spacing of 50m, to acquire raw data including depth and backscatter intensity. This system met the data accuracy and density requirements for this subsea oil pipeline inspection.

[0136] 2. Preprocessing: Depth correction is performed using measured sound velocity profiles to correct the raw depth data to accurate depth data; local adaptive normalization is performed on the echo intensity, and contrast stretching parameters are set. , According to the formula:

[0137]

[0138] Calculations are performed; anisotropic Gaussian filtering is used ( Noise removal preserves pipe edge features while suppressing noise.

[0139] 3. Coarse segmentation: Set gradient threshold Intensity threshold Elevation threshold According to the joint judgment conditions:

[0140]

[0141] Initial segmentation is performed to generate a binary candidate map; then, connectivity optimization is performed through morphological closing operations to form a set of connected candidate regions.

[0142] 4. Refined Segmentation: For each candidate region, its multi-scale geometric features (curvature, symmetry, skeleton smoothness, etc.) and relative elevation are calculated. These features are used to construct feature vectors, which are then input into a pre-trained GBDT classifier for identification. This effectively eliminates more than 90% of interference from fishing nets and pipe trenches, resulting in refined segmentation results. .

[0143] 5. AI-optimized segmentation: The refined segmentation result is cropped into 512x512 pixel blocks and input into the trained dual-branch attention network. The learning rate is set to 0.001, the batch size to 8, and the loss function weights are set accordingly. , , The loss function formula is:

[0144]

[0145] The model further reduced the missegmentation rate and repaired locally broken pipe segments, resulting in AI-optimized segmentation results. .

[0146] 6. Geographic Verification and Output: The results were converted to the WGS-84 UTM standard geodetic coordinate system and compared with the designed route. The average vertical distance was calculated to be 2.1m, which is less than the tolerance threshold of 5m. The overlap rate was also calculated with historical data. The value was greater than the set value of 0.6. Finally, a bare leakage distribution map of the pipeline section was generated, confirming the existence of 2 bare leakage sections with a total length of 25.6 meters and a maximum exposed height of 0.45 meters, providing accurate data support for the maintenance work of the subsea oil pipeline.

[0147] The practical application of this invention in the Liaodong Bay Suizhong Terminal subsea oil pipeline fully demonstrates the superior effectiveness of the technical solution. Specifically, the invention achieves the following beneficial effects:

[0148] 1. Excellent anti-interference performance: Through multi-level anti-interference mechanisms, including fine segmentation under geometric constraints, attention mechanism and physical guidance deep learning optimization, and multi-source geographic information fusion verification, the misclassification rate of fishing nets, pipe trenches and reefs is less than 3%, 5% and 2% respectively, and the overall segmentation accuracy is improved to more than 95%, effectively solving the problem of false detection caused by interference objects in existing methods.

[0149] 2. Extremely high segmentation accuracy: This invention integrates multi-scale geometric morphology analysis and deep learning technology, resulting in clear segmentation boundaries, a positioning error of less than 0.3 meters, strong ability to recover weak and broken pipe features, and the ability to accurately capture the actual shape and location of exposed pipe sections.

[0150] 3. Strong environmental adaptability: Through multi-source feature fusion and physical constraints, the algorithm exhibits good robustness to different seabed sediments, pipe diameters, and data noise levels. It has strong generalization ability and is suitable for various complex marine environments and pipeline inspection scenarios.

[0151] 4. Significant Engineering Benefits: The entire process is highly automated, with a processing speed exceeding 250MB / minute, more than four times more efficient than traditional manual interpretation, significantly reducing labor costs and inspection time. The output results can be directly used to guide pipeline maintenance operations, effectively ensuring the safe operation of subsea pipelines and possessing significant engineering application value.

[0152] Furthermore, to more intuitively demonstrate the technical effects of the present invention, Figure 3A comparison of the results of pipeline leakage detection is provided. In this image, the multibeam sonar image rendered in green pseudo-color clearly shows the three-dimensional morphological features of the seabed topography, including the pipeline's geometry, exposed height, and surrounding topographic undulations; the side-scan sonar image rendered in orange pseudo-color provides the acoustic reflection characteristics of the pipeline and its surrounding environment, effectively reflecting the surface condition of the pipeline and the surrounding seabed conditions.

[0153] By spatially registering and fusing the two types of data, it can be observed that:

[0154] (1) In the multibeam imaging (green), the pipeline exhibits a continuous linear convex feature, with its exposed section forming a significant height difference with the seabed background;

[0155] (2) In the side-scan sonar image (orange), the exposed section of the pipe exhibits strong acoustic reflection characteristics, which contrasts sharply with the sediment-covered section;

[0156] (3) The collaborative display of the two types of data verifies the consistency of the method of the present invention under different detection principles, and demonstrates the advantages of multi-source data fusion in pipeline status identification.

[0157] This visualization method not only intuitively displays the detection results of pipeline leaks, but also demonstrates the effectiveness of the method of this invention in processing multi-source marine exploration data, providing engineers with a reliable basis for interpretation.

[0158] The present invention also provides an electronic device, including a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor. The transceiver, the memory, and the processor are connected via the bus. The computer program, when executed by the processor, implements the steps in the aforementioned method for anti-interference segmentation of exposed subsea pipelines based on multibeam echo sounding data. Compared with the prior art, the beneficial effects of the electronic device provided by the present invention are the same as those of the aforementioned method for anti-interference segmentation of exposed subsea pipelines based on multibeam echo sounding data, and will not be elaborated upon here.

[0159] The present invention also provides a computer-readable storage medium storing a computer program thereon, characterized in that, when the computer program is executed by a processor, it implements the steps in the above-described method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present invention are the same as the beneficial effects of the above-described method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data, and will not be elaborated here.

[0160] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data, characterized in that, include: Step 1: Acquire multibeam bathymetry data of the target seabed; the multibeam bathymetry data includes raw depth data and raw echo intensity data; Step 2: Preprocess the multibeam echo sounding data to obtain preprocessed multibeam echo sounding data; Step 3: Calculate the depth gradient magnitude and elevation anomaly of the preprocessed multibeam echo sounding data, and select candidate points with pipeline characteristics to form a candidate region set; Step 4: Calculate the microscale surface features, macroscale profile morphology features, and relative elevation features of the pipeline candidate region on the seabed to construct feature vectors; The microscale surface feature analysis includes: Surface curvature distribution, specifically, based on PCA of the local neighborhood of the point cloud, to calculate surface curvature: in This represents the dimension along the longest axis of the ellipsoid. The dimension representing the direction of the secondary axis of the ellipsoid. The dimension representing the direction of the shortest axis of the ellipsoid; Step 5: Input the feature vector into the gradient boosting decision tree model to output the refined segmentation result of the pipeline leakage; Step 6: Train the dual-branch feature fusion network using labeled data to obtain the trained model; wherein, the main branch of the dual-branch feature fusion network adopts a U-Net structure, and the auxiliary branch adopts a lightweight network, and the two perform feature fusion in the decoder; Step 7: Input the refined segmentation results into the trained model to obtain the optimized segmentation results.

2. The method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data according to claim 1, characterized in that, Step 2: Preprocessing the multibeam echo sounding data to obtain preprocessed multibeam echo sounding data, including: Step 2.1: Correct and denoise the original depth data to obtain corrected depth data; the correction process is as follows: in, For the corrected depth, Original depth As the reference speed of sound, This is the measured sound velocity profile; Step 2.2: Perform adaptive normalization and Gaussian denoising on the original echo intensity data to obtain the normalized echo intensity; wherein, the adaptive normalization process is as follows: in, The normalized echo intensity The original echo intensity, For the local minimum echo intensity, For the local maximum echo intensity, and This is the contrast stretching parameter.

3. The method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data according to claim 2, characterized in that, Step 3: Calculate the depth gradient magnitude and elevation anomaly of the preprocessed multibeam echo sounding data, and select candidate points with pipeline characteristics to form a candidate region set, including: Step 3.1: Calculate the partial derivatives of the corrected depth data in the horizontal and vertical directions, and take the Euclidean modulus of the partial derivatives in both directions to obtain the depth gradient magnitude. Step 3.2: Use the least squares method to fit the local seabed background surface, and subtract the depth value of each data point of the corrected depth data from the corresponding local seabed background surface to obtain the elevation anomaly; Step 3.3: Using the depth gradient amplitude, elevation anomaly, and normalized echo intensity, establish joint judgment conditions to screen out candidate points with pipeline characteristics and form a binary candidate map; Step 3.4: Use morphological closing operations to process the binary candidate image to obtain a candidate region set.

4. The method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data according to claim 3, characterized in that, In step 3.3, the joint determination condition is: in, This indicates whether the point at coordinates (x, y) is a candidate point for the pipeline; a value of 1 indicates it is a candidate point, and 0 indicates it is not. For gradient threshold, The range of intensity thresholds. Elevation threshold For depth gradient magnitude, To normalize the echo intensity, This indicates an elevation anomaly.

5. A method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data according to claim 4, characterized in that, In step 4, the microscale surface features include: surface curvature distribution characteristics and the angular deviation between the neighborhood normal vector and the average normal vector; the macroscale profile morphology features include: profile symmetry index and profile Gaussian curvature integral; the relative elevation characteristics of the pipeline candidate region on the seabed are: in, This refers to the actual elevation value of the pipeline candidate point or pipeline candidate area on the seabed. This represents the fitted local seabed background surface. Indicates relative elevation.

6. The method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data according to claim 5, characterized in that, In step 6, a convolutional block attention module is embedded in the dual-branch feature fusion network, and attention weights are calculated sequentially in the channel and spatial dimensions to extract pipeline features; wherein, the calculation method of the convolutional block attention module is as follows: in, This represents the feature map after optimization by the attention module of the convolutional block. This indicates a channel attention mechanism used for feature maps. Calculate weights based on channel dimension. This represents a spatial attention mechanism used for feature maps. Calculate weights based on spatial dimensions.

7. The method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data according to claim 6, characterized in that, During training, the loss function is: in, , , For the weights of the loss function, Represents the dice loss function. Represents the focus loss function. This represents the geometric consistency loss.

8. A method for anti-interference segmentation of exposed subsea pipelines based on multibeam bathymetry data according to claim 7, characterized in that, Following step 7, the following is also included: The optimized segmentation results and their corresponding geographical locations are transformed to the geodetic coordinate system. The shortest perpendicular distance between each connected component in the optimized segmentation results and the pipeline design route is calculated, and the segmentation results with the shortest perpendicular distance within the preset tolerance threshold are retained. The AI-optimized segmentation results are compared and analyzed with historical detection data to calculate the overlap rate and morphological similarity between the two. The pipeline bare leakage sections with both overlap rate and morphological similarity greater than the preset threshold are used as the final pipeline bare leakage distribution map.

9. An electronic device comprising a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the transceiver, the memory, and the processor are connected via the bus, characterized in that, When the computer program is executed by the processor, it implements the steps in the anti-interference segmentation method for exposed submarine pipelines based on multibeam bathymetry data as described in any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps in the anti-interference segmentation method for exposed submarine pipelines based on multibeam bathymetry data as described in any one of claims 1-8.

Images