Submarine oil and gas pipeline burying state identification method

Abstract

The invention belongs to the field of submarine pipeline risk early warning, and relates to a submarine oil and gas pipeline burying state identification method based on unsupervised machine learning and 3D Hough transformation. The method comprises three parts of pipeline data acquisition, pipeline point cloud data clustering, and pipeline state detection and identification, wherein the pipeline point cloud data clustering part comprises the following steps: firstly, removing most noise points in point cloud data by using cloth filtering and statistical filtering, and then clustering the point cloud data by using a DBSCAN algorithm of a self-adaptive threshold value; the higher the density is, the higher the class interestingness is. According to the pipeline state detection and identification part, 3D Hough transformation is used for carrying out cylinder and circular truncated cone detection on interest classes, and the exposed state of the pipeline is determined according to different detected results.

Description

Technical Field

[0001] The present invention relates to the technical field of underwater acoustic detection of submarine pipelines and provides a method for identifying the buried state of submarine oil and gas pipelines. Background Art

[0002] In recent years, countries have been competing to develop offshore oil and gas industries. Submarine pipelines operate in complex environments, making them susceptible to corrosion when exposed. Leaks can cause significant economic losses. Consequently, research on the monitoring and integrity assessment of exposed and suspended submarine pipelines has intensified in recent years. However, global research on exposed submarine pipeline detection technology remains underdeveloped.

[0003] The current detection of exposed submarine pipelines has the following problems: (1) It focuses on detecting exposed pipelines after a leak; (2) It mostly uses side-scan sonar for detection, which has disadvantages such as low positioning accuracy, high noise, and inability to perform three-dimensional visualization analysis; (3) There is a lack of intelligent systems for automatic early warning using relevant parameters obtained in the ocean, and manual analysis is generally used, which has problems such as misjudgment, missed judgment, and low efficiency. Therefore, it is necessary to develop automated pipeline exposed detection technology based on current detection technology to achieve intelligent and high-precision control of pipeline status detection.

[0004] Submarine pipeline exposure detection technology determines whether a pipeline is suspended or exposed. Unintended exposure or suspended oil and gas pipelines pose serious safety risks. Natural environmental factors such as waves and tides exert forces on the pipeline, causing vortex-induced vibrations. If the pipeline is suspended beyond its allowable length or the vibration reaches a certain level, damage and leakage may occur.

[0005] Submarine pipeline inspections typically rely on manual analysis of two-dimensional seafloor images obtained by side-scan sonar. In recent years, computer processing has been gradually introduced to process side-scan sonar images. However, due to the difficulty in distinguishing the pipeline from the seafloor environment in side-scan sonar images, the results are often suboptimal. Therefore, it is necessary to research and develop an intelligent system for detecting exposed submarine pipelines based on multibeam sonar. This system could, to some extent, address the inefficiency, mechanical nature, and low precision of current manual observation of side-scan sonar images and manual identification of exposed pipelines.

[0006] Currently, existing pipeline inspection technologies primarily focus on side-scan sonar image recognition and pipeline inspection based on side-scan sonar combined with neural networks. In side-scan sonar image recognition research, images are typically captured directly from the side-scan sonar data display software as the data source. However, the images displayed by the display software do not objectively reflect the size of the sampled data. In a neural network-based approach, two classification and recognition algorithms, neural networks and sparse representation, are used to establish submarine pipeline status recognition and classification models. In the network recognition and classification model, the grayscale matrix converted from the sampled data undergoes image preprocessing and is then used to generate training and test samples. A network-based submarine pipeline side-scan sonar image status recognition and classification model is then established to classify pipeline status images. In the sparse representation recognition and classification model, the grayscale matrix is ​​converted into a grayscale-gradient co-occurrence matrix. A sparse dictionary constructed from the digital features of the sample grayscale-gradient co-occurrence matrix is ​​used to recognize and classify submarine pipeline status images. However, these methods still rely on image processing and cannot effectively utilize the information contained in the pipeline coordinates. Consequently, recognition results are unstable when the submarine pipeline is exposed to a small area or when the seabed conditions are complex. Further research is needed on comprehensive pipeline exposure recognition systems. Summary of the Invention

[0007] The purpose of the present invention is to address the shortcomings of the existing technology and propose a method for identifying the buried status of submarine oil and gas pipelines, namely a method for identifying the buried status of submarine oil and gas pipelines based on unsupervised machine learning and 3D Hough transform, to solve the problem of low recognition accuracy of existing pipeline status recognition based on side-scan sonar.

[0008] Solve the following technical problems:

[0009] By analyzing 3D point cloud data collected by multi-beam sonar, it is possible to separate submarine pipelines from the seabed topography and determine the pipeline's status. This, to a certain extent, addresses the mechanical, accidental, and inefficient nature of current manual judgments of pipeline exposure status based on side-scan sonar images.

[0010] In order to achieve the above object, the present invention adopts the following technical solutions:

[0011] A method for identifying the buried state of a submarine oil and gas pipeline comprises the following steps:

[0012] Step 1: Data collection, obtaining seabed point cloud data;

[0013] Step 2: point cloud clustering, using the adaptive threshold DBSCAN algorithm combined with statistical filtering to determine the optimal value of the threshold MinPts;

[0014] Step 3: Detect and identify the pipeline point cloud and obtain the detection result by feeding back Hough transform;

[0015] Step 4: Determine the pipeline's exposed and leaky state. Determine the pipeline's suspended, exposed, non-suspended, and non-exposed states based on the test results.

[0016] Furthermore, the point cloud clustering includes:

[0017] (1) Cloth filtering, used to separate the submarine ground point cloud from the ground pipeline point cloud containing the pipeline;

[0018] (2) Ground object point cloud denoising: statistical filtering is performed on the non-seabed ground point cloud separated by cloth filtering;

[0019] (3) Adaptive unsupervised clustering:

[0020] After filtering, the points belonging to the pipeline in the remaining point cloud are relatively dense, so the density-based DBSCAN unsupervised clustering method is used to cluster the point cloud; the Eps and MinPts parameters are selected by the adaptive threshold method based on data statistical principles to determine the optimal value of the threshold MinPts.

[0021] Furthermore, the ground object point cloud denoising process includes: setting the search point number K of the statistical filter, calculating each point x in the target non-ground point cloud i The average distance to K points in the neighborhood is as shown in formula (1):

[0022]

[0023] Set the standard deviation multiple a, calculate the average value μ and standard deviation σ of the average distance of all data points, obtain the screening range of discrete points, and provide the corresponding threshold for discrete point screening, as shown in formulas (2) and (3):

[0024]

[0025]

[0026] Where n is the number of point clouds;

[0027] Determine the distance threshold d threhold , as shown in formula (4):

[0028] d threshold =μ+k·σ(4);

[0029] Enter big d successively threhold and Xiao D threhold A statistical filter is used to remove distant noise points and finely remove noise points near the pipeline.

[0030] Furthermore, the adaptive unsupervised clustering includes:

[0031] Calculate the distance between each point in the data set; distinguish abnormal data based on the distance value, and select the Euclidean distance, as shown in formula (5):

[0032]

[0033] Calculate the distance between each data object in the data set, record it as d(i,j), and finally construct all the calculated d(i,j) into a Dist n×n The matrix is ​​expressed as shown in formula (6):

[0034] Dist n×n ={d(i,j)|1≤i≤n,1≤j≤n} (6);

[0035] The range with the largest Euclidean distance is selected as the optimal value range of Eps. After using the above-described method to determine the optimal value of the neighborhood radius Eps, the density of data objects in the Eps neighborhood of data object i in the dataset is counted. Finally, the distance δ between data object i and the data object with higher density is calculated to determine the optimal value of the threshold MinPts:

[0036]

[0037] Furthermore, the detecting and identifying pipeline point clouds includes:

[0038] (1) Estimating the normal vector of the 3D point cloud;

[0039] (2) Perform axial estimation based on 3D Hough transform;

[0040] (3) Perform axial correction on the cylinder and the cone respectively and establish the objective function of the correction.

[0041] Furthermore, the normal vector estimation for the three-dimensional point cloud is based on an octree search method to convert the point cloud data into a data structure required for obtaining the normal vector.

[0042] Furthermore, the performing axial estimation based on 3D Hough transform includes:

[0043] Obtain the normal vector for each point, generate a Gaussian map of the cylinder or frustum, map all points on the Gaussian map to Hough space, use an accumulator to count the foci of all mapped circles on the Hough Gaussian, find the intersection point greater than the threshold, and determine the estimated cylinder or frustum axis n based on the vector of the line connecting the intersection point and the center of the sphere.

[0044] Furthermore, the modified objective function is as follows:

[0045] δ=ab (8);

[0046] The semi-major axis of the elliptical cross section a=R', the semi-minor axis b=R, When δ is less than a preset threshold, the corresponding axis can be considered to be the optimal axis n.

[0047] Furthermore, the axial correction of the cylinder and the frustum includes obtaining the cross section of the cylinder and determining the direction of the axial error:

[0048] Use Hough transform method to iteratively calculate the ellipse parameters and finally get the ellipse rotation angle and major and minor axes, rotation angle The corresponding two-dimensional vector The direction of the error is obtained, and the semi-major axis and semi-minor axis of the ellipse are calculated by efficient Hough transform. The theoretical value α of the error angle is calculated according to formula (9):

[0049]

[0050] Furthermore, the axial correction of the cylinder and the frustum also includes:

[0051] The axial direction of the cylinder is iteratively corrected according to the theoretical value of the error angle α. The corrected axial direction is Take n' as the initial value, cyclically determine the direction of the axial error, calculate the length difference δ between the semi-major axis and the semi-minor axis, and determine whether the iteration termination condition is met. If not, continue to correct the axial direction until the objective function value is optimal. Finally, the optimal axial direction n' is obtained by inverse transformation of the matrix M to obtain the optimal axial direction n after correction of the cylinder.

[0052] The axial correction of the frustum is based on the projection mechanism. The outer contour ellipse of the projection area is selected as the target object, and the error of the major and minor axis length of the ellipse δ = ab is used as the target function. According to the rotation direction of the outer ellipse The relationship between the major and minor axis parameters of formula (8) is used to correct the axial direction of the frustum, and the iterative correction makes the outer ellipse approach a circle.

[0053] The beneficial effects of the present invention are as follows:

[0054] The present invention uses multi-beam sonar to scan the seabed topography, filters the collected three-dimensional point cloud data, and removes noise points that interfere with subsequent processing. Unsupervised learning methods are used for clustering, combined with adaptive threshold methods to optimize the clustering effect, and 3D Hough transform is used to perform cylinder and frustum detection on the obtained interest class. According to the different detection results, the exposed status of the pipeline is finally output. The present invention combines unsupervised learning with 3D Hough transform to greatly improve the recognition effect of the buried status of submarine oil and gas pipelines. To a certain extent, it solves the lag, inefficiency and mechanical nature of the existing manual observation of seabed images. Compared with traditional image analysis based on side-scan sonar images, it makes more full use of pipeline information and improves the accuracy of identifying the buried status of submarine oil and gas pipelines to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a workflow diagram of the buried state identification method for submarine oil and gas pipelines based on unsupervised machine learning and 3D Hough transform.

[0056] Figure 2 Flowchart for point cloud filtering and unsupervised clustering with adaptive thresholding.

[0057] Figure 3 In order to detect and identify the structure diagram of the pipeline point cloud through feedback Hough transform. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0059] Example 1:

[0060] A method for identifying the buried state of submarine oil and gas pipelines first uses multibeam sonar to collect seabed point cloud data to form a seabed point cloud dataset. Then, using point cloud filtering and unsupervised clustering, noise is removed from the seabed point cloud dataset and adaptive threshold unsupervised clustering is performed. The dataset is sorted by intra-cluster density to determine the optimal threshold (MinPts). Finally, a feedback Hough transform is used to perform axial correction on cylinders and frustums. This correction process yields the optimal axis and radius R. The cylinder's axis must also be determined based on the cross-section of the cylinder after axial correction. For a frustum, only the axis and outer radius are known, so the axis and inner radius of the frustum must also be determined. Based on the estimated optimal axis, a cross-section of the cylinder is obtained perpendicular to the axis. This cross-section is a standard circle, which is fitted using the Hough transform. The center of the circle is the cylinder's axis. Similarly, the circular ring formed by the frustum along the optimal axis is detected. The center and inner radius of the ring are the axis and another radius parameter of the frustum. Ultimately, detection and identification of cylinders and frustums is achieved.

[0061] Point cloud clustering includes:

[0062] (1) Cloth filtering, used to separate the submarine ground point cloud from the ground pipeline point cloud containing the pipeline;

[0063] (2) Ground object point cloud denoising: statistical filtering is performed on the non-seabed ground point cloud separated by cloth filtering;

[0064] (3) Adaptive unsupervised clustering:

[0065] After filtering, the points belonging to the pipeline in the remaining point cloud are relatively dense, so the density-based DBSCAN unsupervised clustering method is used to cluster the point cloud; the Eps and MinPts parameters are selected by the adaptive threshold method based on data statistical principles to determine the optimal value of the threshold MinPts.

[0066] The noise reduction process of ground object point cloud includes: setting the search point number K of the statistical filter, calculating the x value of each point in the target non-ground point cloud, and i The average distance to K points in the neighborhood is as shown in formula (1):

[0067]

[0068] Set the standard deviation multiple a, calculate the average value μ and standard deviation σ of the average distance of all data points, obtain the screening range of discrete points, and provide the corresponding threshold for discrete point screening, as shown in formulas (2) and (3):

[0069]

[0070]

[0071] Where n is the number of point clouds;

[0072] Determine the distance threshold d threhold , as shown in formula (4):

[0073] d threshold =μ+k·σ(4);

[0074] Enter big d successively threhold and Xiao D threhold A statistical filter is used to remove distant noise points and finely remove noise points near the pipeline.

[0075] Example 2:

[0076] like Figures 1 to 3 A method for identifying the buried state of submarine oil and gas pipelines mainly includes the following steps:

[0077] 1) Data Collection

[0078] Seafloor point cloud data collected using multibeam sonar.

[0079] 2) Point cloud clustering using the adaptive threshold DBSCAN algorithm combined with statistical filtering

[0080] (1) Performing cloth filtering on the input point cloud data to separate the submarine ground point cloud from the ground pipeline point cloud containing the pipeline;

[0081] (2) Ground object point cloud denoising: Perform statistical filtering on the non-seabed ground point cloud separated by cloth filtering. Set the search point number K of the statistical filter and calculate the x value of each point in the target non-ground point cloud. i The average distance to K points in the neighborhood is as shown in formula (1):

[0082]

[0083] Set the standard deviation multiple a, calculate the average value μ and standard deviation σ of the average distance of all data points, obtain the screening range of discrete points, and provide the corresponding threshold for discrete point screening, as shown in formulas (2) and (3):

[0084]

[0085]

[0086] Where n is the number of point clouds.

[0087] For points in the non-ground point cloud, all points whose distance to their neighbors is not within the range of (μ-aσ,μ+aσ) are considered outliers; in this algorithm, the k value and the standard deviation multiple can be set. The former controls the neighborhood size, and the latter controls the condition range, that is, the severity of the screening.

[0088] Determine the distance threshold d threhold , as shown in formula (4):

[0089] d threshold =μ+k·σ(4);

[0090] Enter big d successively threhold and Xiao D threhold Statistical filter to remove distant noise points and finely remove noise points near pipelines;

[0091] (3) Adaptive unsupervised clustering:

[0092] After filtering, the remaining point cloud contains relatively dense points belonging to the pipeline. Therefore, the density-based DBSCAN unsupervised clustering method is used to cluster the point cloud. To address the problem that the fixed-threshold DBSCAN algorithm is difficult to accurately segment all local areas in the entire data space, the Eps and MinPts parameters are selected using an adaptive threshold method based on data statistics.

[0093] The distance between each point in the data set is calculated to obtain the distance value between each point; based on the characteristics that abnormal data objects are distributed discretely and irregularly and the data volume is less than that of meaningful data objects, and according to data statistical principles, a reasonable EPS parameter is found to distinguish meaningful data objects from abnormal data objects, thereby reducing the error rate of abnormal data objects marked by the clustering algorithm;

[0094] The distance is selected as Euclidean distance, and the formula is shown in formula (5):

[0095]

[0096] Calculate the distance between each data object in the data set, record it as d(i,j), and finally construct all the calculated d(i,j) into a Dist n×n The matrix is ​​expressed as shown in formula (6):

[0097] Dist n×n ={d(i,j)|1≤i≤n,1≤j≤n} (6);

[0098] The range with the largest Euclidean distance is selected as the optimal value range of Eps; after the DBSCAN algorithm completes clustering, multiple clusters are formed, each cluster has a cluster center, and the area around the cluster center is a region with high local density, while the area around the non-cluster area has low local density. After using the above-described method to determine the optimal value of the neighborhood radius Eps, the data object density of the Eps neighborhood of data object i in the data set is statistically analyzed, and finally formula (7) is used to calculate the distance δ between data object i and the data object with higher density, so as to determine the optimal value of the threshold MinPts:

[0099]

[0100] 3) Detect and identify pipeline point clouds through feedback Hough transform

[0101] Based on the 3D Hough transform, each cluster obtained by clustering is detected and identified in turn according to the density of the point cloud within the class. The main steps are as follows:

[0102] (1) Estimating the normal vector of a 3D point cloud. The octree-based search method can quickly convert the point cloud data into the data structure required for normal vector calculation, providing a good prerequisite for the subsequent axial estimation of the cylinder or frustum.

[0103] (2) Perform axial estimation based on 3D Hough transform:

[0104] Obtain the normal vector for each point to generate a Gaussian map of the cylinder or frustum. Map all points on the Gaussian map to Hough space. Use an accumulator to count the foci of all mapped circles on the Hough Gaussian map. Find the intersection point greater than a threshold. Determine the estimated axis n of the cylinder or frustum based on the vector connecting the intersection point and the center of the sphere.

[0105] (3) Perform axial correction on the cylinder and the cone respectively:

[0106] The modified objective function is established as shown in formula (8):

[0107] δ=ab(8);

[0108] The semi-major axis of the elliptical cross section a=R', the semi-minor axis b=R, When δ is less than a preset threshold, the corresponding axis can be considered to be the optimal axis n.

[0109] Obtain the cross section of the cylinder and determine the direction of the axial error: Use the Hough transform method to iteratively calculate the ellipse parameters and finally obtain the ellipse rotation angle and major and minor axes, rotation angle The corresponding two-dimensional vector The direction of the error is obtained, and the semi-major axis and semi-minor axis of the ellipse are calculated by efficient Hough transform. The theoretical value α of the error angle is calculated according to formula (9):

[0110]

[0111] The axial direction of the cylinder is iteratively corrected according to the theoretical value of the error angle α, and the corrected axial direction is recorded as Take n' as the initial value, cyclically determine the axial error direction, calculate the length difference δ between the semi-major axis and the semi-minor axis, and determine whether the iteration termination condition is met. If not, continue to correct the axial direction until the objective function value is optimal. The axial direction n' at this time is the optimal axial direction of the cylinder. Finally, the optimal axial direction n' is obtained by inverse transformation of the matrix M to obtain the optimal axial direction n after correction of the cylinder.

[0112] The axial correction of the frustum is based on the projection mechanism. The outer contour ellipse of the projection area is selected as the target object, and the error of the major and minor axis length of the ellipse δ = ab is used as the objective function. According to the rotation direction of the outer ellipse The relationship between the major and minor axis parameters of formula (8) is used to correct the axial direction of the frustum, and the iterative correction is made to make the outer ellipse approach a circle, thereby achieving the axial correction of the frustum;

[0113] (4) Radius and position estimation based on Hough transform:

[0114] After obtaining the optimal axial direction and radius R through the correction process, the axis of the cylinder must be determined based on the cross-section after the axial correction of the cylinder. For a frustum, only the axial direction and outer radius are known, so the axis and inner radius of the frustum must also be determined. Based on the optimal axial direction of the cylinder estimated above, the cross-section of the cylinder is obtained perpendicular to the axial direction. The cross-section is a standard circle, and the circle is fitted using the Hough transform method. The center of the circle is the axis of the cylinder. Similarly, the ring projected along the optimal axial direction of the frustum is detected. The center of the ring and the inner radius are the axis and another radius parameter of the frustum. Ultimately, the detection and recognition of cylinders and frustums are achieved.

[0115] 4) Determination of pipeline exposure status:

[0116] According to the detection and recognition results obtained in (4), if a complete cylinder or frustum can be identified for the point cloud class, it means that there is a completely exposed suspended pipe, and the pipe state is determined to be suspended; if only part of the cylinder or frustum arc surface can be identified, it means that the pipe is partially exposed, and the state is determined to be exposed non-suspended; if part of the continuous cylinder or frustum arc surface cannot be detected, the pipe state is determined to be not exposed.

[0117] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for identifying the buried state of a submarine oil and gas pipeline, characterized in that: The following steps are involved: Step 1: Data collection, obtaining seabed point cloud data; Step 2: point cloud clustering, using the adaptive threshold DBSCAN algorithm combined with statistical filtering to determine the optimal value of the threshold MinPts; Step 3: Detect and identify the pipeline point cloud and obtain the detection result by feeding back Hough transform; Step 4: Determine the pipeline's exposed and leaky state. Determine the pipeline's suspended, exposed, non-suspended, and non-exposed states based on the test results.

2. The method for identifying the buried state of a submarine oil and gas pipeline according to claim 1, characterized in that: The point cloud clustering includes: (1) Cloth filtering, used to separate the submarine ground point cloud from the ground pipeline point cloud containing the pipeline; (2) Ground object point cloud denoising: statistical filtering is performed on the non-seabed ground point cloud separated by cloth filtering; (3) Adaptive unsupervised clustering: After filtering, the points belonging to the pipeline in the remaining point cloud are relatively dense, so the density-based DBSCAN unsupervised clustering method is used to cluster the point cloud; the Eps and MinPts parameters are selected by the adaptive threshold method based on data statistical principles to determine the optimal value of the threshold MinPts.

3. The method for identifying the buried state of a submarine oil and gas pipeline according to claim 2, characterized in that: The ground object point cloud denoising process includes: setting the search point number K of the statistical filter, calculating each point x in the target non-ground point cloud i The average distance to K points in the neighborhood is as shown in formula (1): Set the standard deviation multiple a, calculate the average value μ and standard deviation σ of the average distance of all data points, obtain the screening range of discrete points, and provide the corresponding threshold for discrete point screening, as shown in formulas (2) and (3): Where n is the number of point clouds; Determine the distance threshold d threhold , as shown in formula (4): d threshold =μ+k·σ(4); Enter big d successively threhold and Xiao D threhold A statistical filter is used to remove distant noise points and finely remove noise points near the pipeline.

4. The method for identifying the buried state of a submarine oil and gas pipeline according to claim 3, characterized in that: The adaptive unsupervised clustering includes: Calculate the distance between each point in the data set; distinguish abnormal data based on the distance value, and select the Euclidean distance, as shown in formula (5): Calculate the distance between each data object in the data set, record it as di,j, and finally construct all the calculated di,j into a Dist n×n The matrix is ​​expressed as shown in formula (6): Dist n×n ={d(i,j)|1≤i≤n,1≤j≤n}(6); The range with the largest Euclidean distance is selected as the optimal value range of Eps. After using the above-described method to determine the optimal value of the neighborhood radius Eps, the density of data objects in the Eps neighborhood of data object i in the dataset is counted. Finally, the distance δ between data object i and the data object with higher density is calculated to determine the optimal value of the threshold MinPts:

5. The method for identifying the buried state of a submarine oil and gas pipeline according to claim 4, characterized in that: The detecting and identifying pipeline point clouds includes: (1) Estimating the normal vector of the 3D point cloud; (2) Perform axial estimation based on 3D Hough transform; (3) Perform axial correction on the cylinder and the cone respectively and establish the objective function of the correction.

6. The method for identifying the buried state of a submarine oil and gas pipeline according to claim 5, characterized in that: The normal vector estimation for the three-dimensional point cloud is based on an octree search method to convert the point cloud data into a data structure required for obtaining the normal vector.

7. The method for identifying the buried state of a submarine oil and gas pipeline according to claim 6, characterized in that: The axial estimation based on 3D Hough transform includes: Obtain the normal vector for each point, generate a Gaussian map of the cylinder or frustum, map all points on the Gaussian map to Hough space, use an accumulator to count the foci of all mapped circles on the Hough Gaussian, find the intersection point greater than the threshold, and determine the estimated cylinder or frustum axis n based on the vector of the line connecting the intersection point and the center of the sphere.

8. The method for identifying the buried state of a submarine oil and gas pipeline according to claim 7, characterized in that: The modified objective function is as follows: δ=ab(8); The semi-major axis of the elliptical cross section a=R', the semi-minor axis b=R, When δ is less than a preset threshold, the corresponding axis can be considered to be the optimal axis n.

9. The method for identifying the buried state of a submarine oil and gas pipeline according to claim 8, characterized in that: The axial correction of the cylinder and the frustum includes obtaining the cross section of the cylinder and determining the direction of the axial error: Use Hough transform method to iteratively calculate the ellipse parameters and finally get the ellipse rotation angle and major and minor axes, rotation angle The corresponding two-dimensional vector The direction of the error is obtained, and the semi-major axis and semi-minor axis of the ellipse are calculated by efficient Hough transform. The theoretical value α of the error angle is calculated according to formula (9):

10. The method for identifying the buried state of a submarine oil and gas pipeline according to claim 9, characterized in that: The axial correction of the cylinder and the frustum also includes: The axial direction of the cylinder is iteratively corrected according to the theoretical value α of the error angle. The corrected axial direction is n' = (sinαcosφ, sinαsinφ, cosα). Taking n' as the initial value, the axial error direction is determined cyclically, and the length difference δ between the semi-major axis and the semi-minor axis is calculated. It is judged whether the iterative termination condition is met. If not, the axial direction is continuously corrected until the objective function value is optimal. Finally, the optimal axial direction n' is obtained by inverse transformation of the matrix M to obtain the optimal axial direction n of the cylinder after correction. The axial correction of the frustum is based on the projection mechanism. The outer contour ellipse of the projection area is selected as the target object, and the length error of the major and minor axes of the ellipse δ = ab is used as the objective function. The axial direction of the frustum is corrected according to the rotation direction of the outer ellipse (cosφ, sinφ) and the major and minor axis parameter relationship of formula (8). The iterative correction makes the outer ellipse approach a circle.

Images