Shield tunnel leakage water automatic identification method based on point cloud data

Abstract

The invention discloses a shield tunnel leakage water automatic identification method based on point cloud data. The method comprises the following steps: preprocessing point cloud data; denoising the point cloud; generating a grey-scale map and establishing a data set; model training and evaluation; and identifying a water leakage area and calculating the area. By the adoption of the method, point cloud data processing and the YOLO deep learning network, the position of water leakage can be accurately positioned, accurate leakage area calculation is provided, and compared with other methods, the method is not affected by tunnel illumination conditions and different environments and has higher adaptability and robustness.

Description

Technical Field

[0001] The invention relates to a method for analyzing water leakage in a tunnel, and in particular to a method for automatically identifying water leakage in a shield tunnel based on point cloud data. Background Art

[0002] In recent years, the urbanization process has accelerated, and a large number of shield tunnel projects have been built in dense urban environments. However, with this, tunnels face a variety of risks, including segment leakage during construction and operation. Therefore, it is crucial to conduct effective physical property detection and analysis of tunnel leakage. To solve this problem, the latest measurement technology and methods are applied, point cloud data is captured through 3D laser scanning technology, and high-precision automatic detection of tunnel leakage is achieved through new algorithms.

[0003] Current tunnel water leakage detection and analysis methods usually rely on manual inspection methods. Currently, most leakage detection methods rely on manual inspection, which is hampered by inefficiency and subjective judgment of inspectors. This method cannot meet the growing demand for fast and accurate leakage identification analysis. Leak identification is not only time-consuming, but also prone to inaccuracies.

[0004] Through the search of prior art literature, it was found that Chinese invention patent publication number CN118799791A describes a method for detecting water leakage in tunnels. The patent proposes to establish a water leakage dataset and build a water leakage semantic segmentation model based on the EfficientNet B0 network. The model is trained based on the dataset, and the optimized model is finally saved to detect water leakage in tunnels. Although this method has certain application prospects in water leakage detection, its adaptability under different lighting and different tunnel environments needs to be verified. In addition, this method cannot calculate the leakage area of a specific section of the tunnel, and has certain limitations. Summary of the invention

[0005] The technical problem to be solved by the embodiments of the present invention is to provide a method for automatically identifying water leakage in a shield tunnel based on point cloud data, which can be used to detect and analyze the physical properties of water leakage in the tunnel.

[0006] In order to solve the above technical problems, an embodiment of the present invention provides a method for automatically identifying water leakage in a shield tunnel based on point cloud data, which is characterized by comprising the following steps: S1: Use a 3D laser scanner to stitch and downsample according to the set station distance, and segment the resulting tunnel point cloud at the same interval and unfold it into a plane; S2: denoising by using CSF and SOR methods in sequence; S3: Convert the point cloud data into grayscale images using a rasterization-based method, map the coordinates and intensity information of the 3D point cloud onto a 2D image, and construct a dataset from the grayscale images generated from the point clouds of various tunnels; S4: Train a YOLO-based leakage detection model using the dataset; S5: Process the target tunnel point cloud image according to the steps of S1 to S3, divide the tunnel segment image into sub-images of a size suitable for model detection, analyze the sub-images using the trained leakage detection model, segment the leakage area, and calculate the leakage area using the following formula: ; where: M l is the leakage area, m is the pixels of the leakage segmentation mask in the sub-image of the tunnel segment, p0 is the number of pixel points of the grayscale image, n is the number of sub-images, divided according to the inner wall perimeter of the tunnel, D is the actual width of the tunnel segment, l is the length pixels of the grayscale image, and w is the width pixels in the grayscale image.

[0007] Furthermore, the S1 further includes the steps of: Convert the segmented tunnel point cloud from the Cartesian coordinate system to the cylindrical coordinate system. After the conversion, the segmented point cloud can be unfolded. The unfolding method is:

[0008] wherein, the tunnel segmented point cloud is unfolded along the Z-axis. During the unfolding process, the position of point P(θ, r, z) is projected onto a plane to become point P'(x', y', z').

[0009] Furthermore, the denoising step using CSF in the S2 includes: S21: Flip the original point cloud and let the virtual cloth fall onto the inverted point cloud. The cloth is modeled by a mass-spring system, where the cloth consists of particles and is connected by springs to simulate the behavior of real cloth. The displacement of each particle under the action of gravity is calculated according to the following formula

[0010] where, m is the mass of the particle, X(t) is the position of the particle at time t, ∆t is the time step, and G is the gravitational constant; S22: After the cloth is under the action of gravity, the position information of each particle is adjusted according to the force exerted by neighboring particles. Calculate the height difference of each particle relative to its neighbors, and calculate the displacement vector through the particles connected by springs. The calculation formula is:

[0011] where, is the displacement vector of the particle. When b = 1, the particle can move; otherwise, it is 0. is the position of the current particle, is the position of the neighboring particle connected to it, = (0, 0, 1) T is the vertical direction vector; Set the height threshold according to the tunnel environment, and regard the particles with a vertical distance exceeding this threshold as non-ground points and remove them from the point cloud.

[0012] Further, the denoising step using SOR in S2 includes: S23: For each point in the point cloud p i , find its k nearest neighbor points { p i1 ,p i2 …p ij }, and calculate the Euclidean distance p i between the target point p ij and each neighbor point d ij :

[0013] S24: For each point p i , calculate the average distance μ i and the standard deviation σ i in its neighborhood. The formulas are as follows: ; ; S25: Calculate the segmentation threshold T, , λ is the multiple set according to the point cloud density. If the average distance of a certain point exceeds the threshold T, then this point is regarded as an outlier and removed from the point cloud.

[0014] Further, the step of converting the point cloud data into a grayscale image in S3 includes: S31: Image resolution calculation. Calculate the range of the point cloud to determine the number of rows and columns in the generated image:

[0015] where z min and z max respectively represent the minimum and maximum values in the Z-axis direction of the point cloud, and y minand y max represent the minimum and maximum values in the Y-axis direction, and s is the pixel resolution, which is set according to the point cloud density; S32: Linearly normalize the intensity value to the gray scale range of 0 to 255, and the calculation formula is:

[0016] where the intensity is the intensity value of each point in the point cloud, I min is the minimum intensity value, I max is the maximum intensity value, I mapped is the normalized gray value, γ is a non-linear parameter used to control the normalization distribution, and the contrast effect of the image is adjusted through gamma correction, which is set according to the characteristics of different tunnel point clouds; S33: Map the 2D coordinates z and y of each point to the corresponding positions in the image matrix, and assign the normalized intensity value to the pixels at these positions.

[0017] Implementing the embodiments of the present invention has the following beneficial effects: The method of the present invention combines point cloud data processing with the YOLO deep learning network, can accurately locate the position of water leakage, and provides accurate calculation of the leakage area. Compared with other methods, the present invention is not affected by the tunnel lighting conditions and different environments, and has stronger adaptability and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of the present invention; Figure 2 is a schematic diagram of the unfolded point cloud of the tunnel section of the present invention; Figure 3 is the denoising result of the point cloud of a certain tunnel section in an embodiment of the present invention; Figure 4 is the recognition result diagram of the leakage area in a certain tunnel section in an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0019] To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings.

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. The examples are only used to explain the present invention and are not used to limit the scope of the present invention. In the following paragraphs, the present invention will be described more specifically by way of example with reference to the accompanying drawings. The advantages and features of the present invention will be clearer according to the following description and the claims. It should be noted that the drawings are all in a very simplified form and use non-precise scales, and are only used to facilitate and clearly assist in explaining the objectives of the embodiments of the present invention.

[0021] AsFigure 1 As shown in Figure 1 , an automatic identification method for leakage in shield tunnels based on point cloud data according to an embodiment of the present invention mainly includes the following steps: S1: Preprocessing of point cloud data; S2: Denoising of point cloud; S3: Generation of grayscale images and establishment of data sets; S4: Model training and evaluation; S5: Identification of leakage areas and calculation of areas.

[0022] In S1, the point cloud data obtained by three-dimensional laser scanning is preprocessed according to the following steps: First, point cloud stitching registration and denoising are performed. First, the Trimble RealWorks software supporting the three-dimensional laser scanner is used for point cloud station stitching and downsampling, and finally the tunnel point cloud is segmented at the same interval.

[0023] After that, the segmented tunnel point cloud needs to be converted from a rectangular coordinate system to a cylindrical coordinate system. After the conversion is completed, the segmented point cloud can be unfolded. The unfolding method is shown in formula (1). Among them, the tunnel segmented point cloud is unfolded along the Z axis. During the unfolding process, the position of point P( θ, r，z ) is projected onto a plane to become point P'( x', y', z' ), (see Figure 2 ).

[0024] (1) Where the original point is in the cylindrical coordinate system, θ, r, and z respectively represent the azimuth angle, radial distance, and height value of the point, and x', y', and z' are the three-dimensional coordinates of the corresponding point after unfolding.

[0025] Specifically, the tunnel under test is constructed by the shield method, and the segment width is 1.8 m. In this embodiment, the point cloud data is obtained by a Trimble stationary three-dimensional laser scanner, the station spacing is 15 m, and 3 white target marks are set on the left and right sides and the bottom of the tunnel between stations as coordinate control points. The original tunnel point cloud is downsampled by the above steps, and the sampling interval is set to 0.01 m. After downsampling, the tunnel point cloud is segmented, and each segment is segmented at a fixed interval of five tunnel segments, numbered sequentially from 1, and then each segment of the tunnel point cloud is unfolded into a plane by the above method.

[0026] In S2, the CSF and SOR methods are used for denoising in sequence, and it is specifically implemented through two steps: S21: Denoising of point cloud based on CSF; S22: Denoising of point cloud based on SOR; In S21, the tunnel slice point cloud is denoised based on CSF (cloth simulation filtering) according to the following steps: (1) Point cloud flipping First, the original point cloud is flipped, and the virtual cloth is dropped onto the inverted point cloud. The cloth is modeled by a mass-spring system, where the cloth consists of particles and is connected by springs to simulate the behavior of real cloth. The displacement of each particle under the action of gravity is calculated according to the following formula.

[0027] (2) (2) Position adjustment After the cloth is under the action of gravity, the position information of each particle is adjusted according to the forces exerted by neighboring particles to control the movement of the particles in the void area of the inverted surface. Calculate the height difference of each particle relative to its neighbors, and calculate the displacement vector through the particles connected by springs, as follows: (3) Among them, is the displacement vector of the particle, b = 1 when the particle is movable, otherwise 0; is the position of the current particle, is the position of the neighboring particle connected to it, = (0,0,1) T is the vertical direction vector. By analyzing the particle positions after cloth simulation, CSF can determine which particles are stably on the "ground" and which particles are in the "air" or lifted. Finally, by setting a threshold (based on the height of the particles), CSF regards the particles above this threshold as non-ground points (noise) and removes them from the point cloud.

[0028] In S22, since CSF cannot completely remove all noise points, the SOR method further eliminates the remaining noise points. The point cloud denoising based on SOR (statistical outlier removal) is determined as follows: (1) Neighborhood search For each point p i in the point cloud, find its k nearest neighbor points { p i1 ,p i2 …p ij}, and calculate the Euclidean distance between the target point p i and each neighbor point p ij ​d ij : (4) (2) Calculate the average distance and standard deviation For each point p i , calculate the average distance within its neighborhood μ i and standard deviation σ i , as follows respectively: (5) (6) (3) Outlier judgment Finally, calculate the segmentation threshold T, which is usually a certain multiple of the average distance plus the standard deviation λ .

[0029] (7) λ is the multiple set for the point cloud density. If the average distance of a certain point exceeds the threshold T , then this point is regarded as an outlier and removed from the point cloud. The SOR method also needs to set the number of neighborhood points k and multiple λ .

[0030] In the above, in the steps of the CSF algorithm for filtering noise points, the cloth resolution is set to 0.1, the maximum number of iterations is set to 500, and the classification threshold is set to 0.02. Therefore, the expanded point cloud is divided into ground and non - ground points. Although the CSF successfully filters out most of the noise (non - ground points), including pipelines, shadows, bolt holes, and segment ring joints, etc. The remaining noise points are further filtered using the SOR method. In order to further eliminate the remaining noise, the SOR is applied. This method requires two key parameters: the number of neighborhood points k is set to 1000 because this value provides better filtering performance. The multiple λ of the standard deviation is set to 5. The denoising process is as Figure 3 shown.

[0031] In S3, the generation of the grayscale image and the establishment of the dataset include the following steps: S31: Generation of the grayscale image; S32: Establishment of the dataset.

[0032] In S31, the grayscale image is generated by the following steps: The leakage area has a stronger absorption ability for laser, and the reflection intensity is a reliable criterion for water leakage detection. Therefore, a rasterization-based method is used to convert the PCD into a grayscale image, mapping the coordinates and intensity information of the three-dimensional point cloud onto a two-dimensional image. The main process of this method includes three aspects: 1) calculation of image resolution, 2) grayscale intensity mapping, and 3) rasterization.

[0033] (1) Calculation of image resolution. The number of rows and columns in the generated image is determined by calculating the range of the point cloud according to Equation 8. Where z min and z max represent the minimum and maximum values in the Z-axis direction of the point cloud respectively. Similarly, y min and y max represent the minimum and maximum values in the Y-axis direction. s is the pixel resolution. An image matrix is created.

[0034] (8) (2) Grayscale intensity mapping. The intensity value is linearly normalized to the grayscale range of 0 to 255 using Equation 9. Where intensity is the intensity value of each point in the point cloud. I min is the minimum intensity value. I max is the maximum intensity value. I mapped is the normalized grayscale value, γ is a non-linear parameter used to control the normalization distribution, and the contrast effect of the image is adjusted through gamma correction, which is set according to the characteristics of different tunnel point clouds.

[0035] (9) (3) Rasterization. The 2D coordinates z and y of each point are mapped to the corresponding positions in the image matrix, and the normalized intensity value is assigned to the pixel at that position.

[0036] In S32, the dataset is established as follows: A dataset is constructed from the grayscale images generated from the point clouds of various tunnels. As mentioned above, a high-resolution 3D laser scanner is used to capture the point clouds of the tunnels, and then the above steps are used to process and convert them into 2D grayscale images. The leakage areas in the pictures are manually annotated to ensure that the model has high-quality labeled data for training.

[0037] Using the above method, the denoised tunnel segmented point cloud is converted into a grayscale image. The pixel resolution of the generated grayscale image is set to 0.01. γ is set to 1. The leakage areas in the generated grayscale image are manually marked to establish a dataset for model training.

[0038] In S4, a YOLO-based leakage detection model is trained using the above dataset. The YOLO algorithm is a deep learning model for object detection in computer vision, and a YOLO-based leakage detection model is trained using the constructed dataset. The YOLO framework is chosen because of its real-time object detection ability and accuracy. (1) During training, hyperparameters such as the learning rate, batch size, and number of iterations need to be adjusted to optimize the model performance. (2) To improve the diversity and robustness of the dataset, data augmentation techniques such as translation, scaling, and flipping in deep learning are adopted, thus improving the generalization ability of the model under different tunnel conditions. (3) Metrics such as the mean average precision (mAP), precision, and recall obtained after training are used to continuously evaluate the model performance.

[0039] Training is carried out on a Windows 11 system equipped with a 12th generation Intel (R) Core (TM) i7-12700F processor and an Nvidia GeForce RTX3060Ti graphics card. The model is trained using the above dataset under these conditions. In this study, the setting of hyperparameters is divided into two categories: training and data augmentation (see Table 1).

[0040] Table 1 Model training parameter settings

[0041] In the output results after model training, the model training results of this model show good performance. The precision steadily increases as training progresses and finally exceeds 0.8. At the same time, the recall also shows a positive trend and finally reaches above 0.8, demonstrating a strong target capture ability. In addition, the mean average precision (mAP50) continuously improves during training and finally stabilizes above 0.8. In summary, as training progresses, the model shows a strong improvement in all key metrics and finally stabilizes above 0.8. The overall prediction and segmentation ability increases as training progresses, making the model suitable for high-precision tasks.

[0042] In S5, to detect the leakage area in an actual tunnel project, first, the target tunnel point cloud is processed using the above steps and converted into a grayscale image. Then each image is divided into three sub-images to match the target size (640*640) used during model training, thus ensuring a more accurate segmentation of the leakage area. Then the trained leakage detection model is used to analyze the sub-images to effectively segment the leakage area. The segmentation result of the leakage area determines the number of pixels covered by the leakage area mask. In addition, the number of pixels in the width of the grayscale image of a tunnel segment corresponds to the value of the user-defined fixed interval mentioned in step S1, enabling the calculation of the actual measurement value of the leakage area in each tunnel segment. The calculation is shown in Equation 10. Where Ml is the leakage area. m is the pixel of the leakage segmentation mask in the sub-image of the tunnel segment, n is 3, representing the number of sub-images. p0 is the number of pixel points in the grayscale image. D is the actual width of a slice of the tunnel segment. In addition, l is the number of length pixels of the grayscale image, and w is the number of width pixels in the grayscale image.

[0043] (10) After evaluating the robustness of the model, it was applied to the leakage detection of a shield tunnel in Guangzhou. By comparing the leakage areas detected by the model with the actual leakage areas, the accuracy of the method was verified. As Figure 4 shown, the model effectively detected and segmented all leakage areas in one of the tunnel segments and was not affected by complex backgrounds. In addition, the leakage area of the tunnel segment can be calculated using Equation 10. The leakage areas were segmented by the proposed model and the leakage areas were calculated, and then the results were compared with the measured values of the manually marked leakage areas. The results are shown in Table 2. The results show that the accuracy of the model reached 92%, further demonstrating its effectiveness in practical applications.

[0044] Table 2 Model Detection Accuracy

[0045] As described above, it is only the preferred embodiment of the present invention, and there is no any form of limitation to the present invention; any ordinary technician in the industry can smoothly implement the present invention according to what is shown in the accompanying drawings of the specification and the above description; however, any equivalent changes made by those skilled in the art within the scope of the technical solution of the present invention by using the technical content disclosed above, such as some modifications, decorations and evolutions, are all equivalent embodiments of the present invention; at the same time, any equivalent changes, modifications and evolutions made to the above embodiments based on the essential technology of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. An automatic identification method for leakage in shield tunnels based on point cloud data, characterized in that, Including the following steps: S1: Use a 3D laser scanner to perform stitching and downsampling according to the set station distance, segment the obtained tunnel point cloud at the same interval, and unfold it into a plane; S2: Denoise successively using the CSF and SOR methods in sequence; S3: Use a rasterization-based method to convert the point cloud data into a grayscale image, map the coordinates and intensity information of the 3D point cloud onto the 2D image, and construct a dataset from the grayscale images generated from the point clouds of various tunnels; S4: Use the dataset to train a leak detection model based on YOLO; S5: Process the target tunnel point cloud image according to the steps of S1 to S3, divide the tunnel segment image into sub-images suitable for model detection, analyze the sub-images using the trained leak detection model, segment the leak area, and calculate the leak area using the following formula: ； Where: M l is the leakage area, m is the pixel of the leakage segmentation mask in the sub-image of the tunnel section, p0 is the number of pixel points of the grayscale image, n is the number of sub-images, which is divided according to the inner wall perimeter of the tunnel, D is the actual width of the tunnel section, l is the length pixel of the grayscale image, and w is the width pixel in the grayscale image.

2. The automatic identification method for leakage of shield tunnel based on point cloud data according to claim 1, characterized in that The S1 further includes the steps: Convert the segmented tunnel point cloud from the Cartesian coordinate system to the cylindrical coordinate system, and after the conversion, unfold the segmented point cloud. The unfolding method is: Among them, the tunnel segmented point cloud is unfolded along the Z axis. During the unfolding process, the position of point P(θ, r, z) is projected onto the plane to become point P'(x', y', z').

3. The automatic identification method for seepage and leakage of shield tunnels based on point cloud data according to claim 1, wherein, The steps of using CSF for denoising in S2 include: S21: Flip the original point cloud, and let the virtual cloth fall onto the inverted point cloud. The cloth is modeled by a mass-spring system, where the cloth consists of particles and is connected by springs to simulate the behavior of real cloth. The displacement of each particle under the action of gravity is calculated according to the following formula Among them, m is the mass of the particle, X(t) is the position of the particle at time t, ∆t is the time step, and G is the gravitational constant; S22: After the cloth is under the action of gravity, the position information of each particle is adjusted according to the forces exerted by neighboring particles. Calculate the height difference of each particle relative to its neighbors, and calculate the displacement vector through the particles connected by springs. The calculation formula is: Among them, is the displacement vector of the particle. When b = 1, the particle can move; otherwise, it is 0; is the position of the current particle, is the position of the neighboring particle connected to it, = (0, 0, 1) T is the vertical direction vector; Set a height threshold according to the tunnel environment, and regard the particles with a vertical distance exceeding this threshold as non-ground points and remove them from the point cloud.

4. The automatic identification method for leakage of shield tunnel based on point cloud data according to claim 3, wherein, The steps of using SOR for denoising in S2 include: S23: For each point in the point cloud p i , find its k nearest neighbor points { p i1 ,p i2 …p ij}, and calculate the Euclidean distance p i between the target point p ij and each neighbor point d ij : S24: For each point p i , calculate the average distance within its neighborhood μ i and the standard deviation σ i , and the formulas are as follows: ； ； S25: Calculate the segmentation threshold T, , λ which is a multiple set for the point cloud density. If the average distance of a certain point exceeds the threshold T, then this point is regarded as an outlier and removed from the point cloud.

5. The automatic identification method for leakage of shield tunnel based on point cloud data according to claim 1, characterized in that, The steps of converting the point cloud data into a grayscale image in S3 include: S31: Image resolution calculation, calculate the range of the point cloud to determine the number of rows and columns in the generated image: where z min and z max represent the minimum and maximum values in the Z-axis direction of the point cloud respectively, y min and y max represent the minimum and maximum values in the Y-axis direction, and s is the pixel resolution; S32: Linearly normalize the intensity value to the grayscale range of 0 to 255. The calculation formula is: where the intensity is the intensity value of each point in the point cloud, I min is the minimum intensity value, I max is the maximum intensity value, I mapped is the normalized grayscale value, γ is a non-linear parameter used to control the normalized distribution, and the contrast effect of the image is adjusted through gamma correction, which is set according to the characteristics of different tunnel point clouds; S33: The 2D coordinates z and y of each point are mapped to the corresponding positions in the image matrix, and the normalized intensity value is assigned to the pixel at that position.

Images