Intelligent Quantification and Evaluation Method for Functional Defects in Drainage Pipelines Based on 3D Point Cloud

By employing an attention-based 3D point cloud semantic segmentation and cylindrical fitting method, the problem of automatic identification and quantitative assessment of functional defects in drainage pipelines was solved, achieving accurate quantification and standardization of functional defects, and supporting maintenance decisions for drainage pipelines.

CN123071231APending Publication Date: 2026-11-10HENAN UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611076550.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-20
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing technologies struggle to automatically identify and quantitatively assess functional defects in drainage pipes, such as deposits, scale, and obstructions, making it difficult to quantify the loss of water flow capacity.

Method used

A 3D point cloud semantic segmentation method based on attention mechanism is adopted, combined with cylinder fitting and voxel-actual measurement calibration, to achieve end-to-end identification and classification of functional defects, quantify deposition thickness, water flow cross-sectional loss rate and longitudinal clearance, and verify its alignment with current standards through full-scale tests.

Benefits of technology

It enables automatic identification and quantitative assessment of functional defects, accurately quantifies water flow capacity loss, complies with current standards, and provides effective support for drainage pipeline maintenance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123071231A_ABST
    Figure CN123071231A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of deep learning and 3D defect detection in drainage pipelines. It discloses an intelligent quantification and evaluation method for functional defects in drainage pipelines based on 3D point clouds, comprising the following steps: acquisition and preprocessing of 3D point clouds on the inner surface of the drainage pipeline; semantic segmentation of 3D point clouds based on attention mechanisms and identification of functional defects; fitting of cylindrical geometric models and establishment of water passage cross-section benchmarks; voxel-actual measurement calibration; quantitative calculation of geometric parameters of functional defects; full-scale experimental verification and standard-compliant intelligent evaluation of functional defects. This invention achieves end-to-end identification and classification of functional defects through semantic segmentation of 3D point clouds based on attention mechanisms. It establishes design water passage cross-section benchmarks using cylindrical fitting, and combines cylindrical projection, triangulation, and voxel-actual measurement calibration to quantify deposition thickness, water passage cross-section loss rate, and longitudinal clearance. Finally, it combines full-scale experimental verification with current standards to form an intelligent quantitative evaluation method for functional defects.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of deep learning and three-dimensional defect detection technology for drainage pipelines, specifically to an intelligent quantification and evaluation method for functional defects in drainage pipelines based on three-dimensional point clouds. Background Technology

[0002] Defects in drainage pipes can be divided into two categories: structural defects and functional defects. Functional defects mainly include sediment, scale, obstructions, remnants of walls and dams, tree roots, and scum. Their direct consequence is a reduction in the cross-sectional area of ​​the pipe and a decrease in drainage capacity. They are the main causes of pipe blockage and sewage overflow, and are also the key focus of daily maintenance of drainage pipes.

[0003] Existing methods for quantifying pipeline defects based on 3D point clouds primarily address structural defects such as pipe wall dents and pits. They typically identify outliers deviating from the cylindrical geometry as defects and quantify their depth, area, and other dimensions. However, functional defects are substances that accumulate inside the pipe, reducing the water flow cross-section. Furthermore, types such as deposits, scale, and obstructions need to be differentiated. Relying solely on geometric outlier identification cannot identify defect types or quantitatively assess water flow capacity loss. Therefore, a technical solution is urgently needed to address the difficulties in automatically identifying functional defects and quantitatively assessing water flow capacity loss. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides an intelligent quantification and evaluation method for functional defects in drainage pipelines based on 3D point clouds. It achieves end-to-end identification and classification of functional defects through 3D point cloud semantic segmentation based on an attention mechanism. A design cross-sectional benchmark is established using cylindrical fitting. Combining cylindrical projection, triangulation, and voxel-actual measurement calibration, the deposition thickness, cross-sectional loss rate, and longitudinal clearance are quantified. Furthermore, full-scale experimental verification and compliance with current standards are incorporated to form an intelligent quantitative evaluation method for functional defects that aligns with the "Technical Specification for Inspection and Evaluation of Urban Drainage Pipelines." This method features automatic defect type identification, quantitative evaluation of cross-sectional loss, and compliance with standards, thus solving the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent quantification and evaluation of functional defects in drainage pipes based on three-dimensional point clouds, comprising the following steps: S1. Acquisition and preprocessing of 3D point cloud of inner surface of drainage pipe; S2. Attention-based semantic segmentation and functional defect identification of 3D point clouds; S3. Fitting the cylindrical geometric model and establishing the reference for the water passage section; S4, Voxel-actual measurement calibration; S5. Quantitative calculation of geometric parameters of functional defects; S6. Full-scale testing verification and standard-compliant intelligent assessment of functional defects.

[0006] Preferably, in step S1, the three-dimensional point cloud of the inner surface of the drainage pipe is obtained by multi-view three-dimensional reconstruction or three-dimensional laser scanning, and denoising preprocessing is performed by statistical outlier removal method.

[0007] Preferably, in step S2, considering the disordered, sparse, and unevenly distributed characteristics of the drainage pipe point cloud data, an encoder-decoder semantic segmentation network with an attention mechanism is constructed to perform multi-class semantic annotation on each point in the point cloud, identifying and distinguishing different types of functional defects such as deposits, scale, and obstacles end-to-end; specifically including the following: S21. Encoding stage: The farthest point sampling algorithm is used to perform hierarchical downsampling on the input point cloud to obtain the sampling center point. The ball query algorithm is used to divide the local region with each sampling center point as the center of the ball. The geometric features of each local region are extracted by the multilayer perceptron and aggregated into local feature vectors by max pooling. Residual connections are introduced in each feature extraction module to alleviate gradient vanishing. S22. Decoding stage: The multi-scale abstract features generated in the encoding stage are upsampled layer by layer using nearest neighbor-based weighted interpolation to gradually restore the original resolution of the point cloud; S23, Cross-layer fusion: Multi-scale features extracted from each layer in the encoding stage are fused with corresponding layer features in the decoding stage via skip connections, and a channel-space dual-path attention mechanism is embedded at each skip connection to adaptively calibrate local feature responses and guide the network to focus on functional defect regions; S24. Point-by-point classification and training: The fused features are mapped to the probability of each point belonging to the category of deposition, scaling, obstacles and normal pipe wall through one-dimensional convolution. The point with the highest probability is taken as the semantic label of the point. Based on the three-dimensional point cloud database of functional defects in drainage pipes, the network is trained and optimized with weighted cross-entropy loss. The optimal combination of hyperparameters and attention mechanism configuration are selected to output the point cloud subsets of various functional defects.

[0008] Preferably, in step S3, a method for fitting the cylindrical geometric model of the inner surface of the drainage pipe based on the Random Sampling Consensus (RANSAC) algorithm is established, the pipe axis direction and radius parameters are extracted, and a plane passing through the origin and perpendicular to the axis is used as the reference for the water passage section, specifically including the following: S31. Fit a cylindrical geometric model to the point cloud of a normal pipe wall using the RANSAC algorithm, and extract the axis direction and radius; wherein, the equation of the cylinder axis is expressed as: ; Where (l, m, n) are the axial direction vectors, the equation of the cylindrical surface is expressed as: ; Where P(x, y, z) represents a point on the surface of the cylinder, P(x0, y0, z0) represents a point on the axis of the cylinder, and r0 represents the radius of the water passage section of the point cloud model; S32. Taking the plane passing through the origin and perpendicular to the axis as the reference for the water passage section, the formula is expressed as: Where A0 is the area of ​​the cross-section of the water passage.

[0009] Preferably, in step S4, a voxel-actual measurement calibration method based on proportional relationships is established. The calibration coefficient is calibrated using the ratio of the known design radius of the pipeline to the reconstructed radius, and the point cloud voxel dimensions are converted into actual dimensions in physical space. The calibration relationship is as follows: ; Where, r real Given the design radius as shown in the design drawings, the coefficient conversion This is the coefficient used to convert the point cloud voxel size to the actual size.

[0010] Preferably, in step S5, for the identified functional defect point cloud, a geometric parameter calculation model based on cylindrical projection and triangulation is proposed to calculate key quantitative indicators such as deposition thickness, water flow cross-sectional loss rate, and longitudinal net distance; specifically including the following: S51. For the identified functional defect point cloud, project the four types of defect points—scale, obstacles, residual walls, and tree roots—onto the cross-section of the defect to construct a triangular mesh. Sum the mesh areas to obtain the voxel water flow cross-sectional loss of the defect. The calculation formula is as follows: ; ; Among them, S Δk The area of ​​each triangular mesh, where K is the total number of triangular meshes. voxel Let area be the voxel area of ​​the cross-section of the water passage. real This refers to the actual area of ​​the cross-section of the water passage. S52. For functional defects such as deposition and scaling, calculate the thickness between the top surface of the defect and the pipe wall radially and characterize the deposition thickness as a percentage of the pipe diameter. The formula for calculating the deposition thickness ratio is as follows: ; ; Where r is the distance from each deposit / scaling functional defect point to the fitted cylinder axis, N is the number of defect points, and t voxel t represents the voxel thickness of the defect. real The actual thickness of the defect; S53. Based on the defect point cloud segmentation results and the cylinder equation, determine the longitudinal center position of each defect instance and obtain the cross-sectional equation passing through the longitudinal center; for multiple defects of the same type in the same pipe section, calculate the distance between the longitudinal centers of each defect, which is the longitudinal net distance of the voxel.

[0011] Preferably, in step S6, a full-scale test model is designed, and a comparative test is conducted on the calculated data and actual measured data of the quantitative indicators of sediment thickness, water flow cross-sectional loss rate, and longitudinal net distance, so as to evaluate the accuracy of the method by relative error.

[0012] The beneficial effects of this invention are as follows: The method of this invention achieves end-to-end identification and classification of functional defects through three-dimensional point cloud semantic segmentation based on attention mechanism, establishes the design water passage cross-section benchmark by cylindrical fitting, and combines cylindrical projection, triangulation and voxel-actual measurement calibration to quantify deposition thickness, water passage cross-section loss rate and longitudinal clearance. Combined with full-scale test verification and current specifications, it forms an intelligent quantitative assessment method for functional defects that is in line with the "Technical Specification for Inspection and Evaluation of Urban Drainage Pipelines", providing effective support for drainage pipeline maintenance decisions. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the intelligent quantification and evaluation method for functional defects in drainage pipelines based on three-dimensional point clouds, as described in this invention. Figure 2 This is a diagram of the 3D point cloud semantic segmentation network structure based on the attention mechanism of the present invention; Figure 3 This is a schematic diagram of the intelligent quantification of functional defects in drainage pipes according to the present invention; Figure 4 A schematic diagram of an intelligent assessment framework for the functional status of urban drainage pipelines. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Terminology Explanation: Functional defects refer to the accumulation of deposits, scale, obstructions, remnants of walls and dams, tree roots, scum, etc. inside drainage pipes, which reduces the cross-sectional area for water flow and decreases drainage capacity. This is different from structural defects such as cracks and deformations.

[0016] Farthest Point Sampling (FPS) is a point cloud downsampling method that iteratively selects the point furthest from the already selected point set as the new sampling point, so that the sampling results are evenly distributed in space and can better preserve the geometric contour of the point cloud.

[0017] Ball Query is a method for constructing local neighborhoods in point clouds. It uses the sampling center point as the center of a sphere and a given radius to construct a sphere. Points falling inside the sphere are considered as local neighborhoods of the center point, which are used to divide local regions of the point cloud.

[0018] Attention mechanism is a method that adaptively weights features by learning weights. Channel attention is used to calibrate the importance of different feature channels, and spatial attention is used to calibrate the importance of different spatial locations. The two are connected to form a channel-space dual-path attention mechanism.

[0019] Random Sample Consensus (RANSAC) is a robust model parameter estimation method that uses random sampling and iterative optimization based on maximizing the number of inliers to robustly fit a geometric model in data containing outliers. This invention uses it to fit a cylinder to establish a reference for the design water passage section.

[0020] The cross-sectional loss rate refers to the ratio of the cross-sectional area occupied by functional defects to the designed full cross-sectional area of ​​the pipeline, which is used to characterize the degree of loss of the pipeline's water carrying capacity.

[0021] This embodiment takes a section of DN800 (800mm inner diameter) concrete drainage pipe as the research object. The pipe contains three typical functional defects: deposits (approximately 15% of the pipe diameter in thickness), scale (circular distribution along the pipe wall), and obstructions (protruding bricks). A high-precision 3D laser scanner (FARO Focus S350) was used to acquire point clouds of the pipe's inner surface at a resolution of 2mm@10m, collecting approximately 8.5 million points. This embodiment provides a technical solution: an intelligent quantification and evaluation method for functional defects in drainage pipes based on 3D point clouds. Specifically, it is an intelligent evaluation method that uses 3D point cloud semantic segmentation to perform end-to-end identification, classification, and quantification of cross-sectional loss of water flow due to functional defects such as deposits, scale, and obstructions, and aligns with current standards. Figure 1 As shown, it includes the following steps: Step S1: Acquisition and preprocessing of 3D point cloud of inner surface of drainage pipe.

[0022] A three-dimensional point cloud model of the inner surface of the drainage pipe was obtained, and noise reduction preprocessing was performed using methods such as statistical outlier removal.

[0023] Specifically, step S1 is as follows: 1.1 Point Cloud Acquisition: This embodiment uses a multi-view 3D reconstruction method to obtain the point cloud of the inner surface of the pipe. Specifically, a dense point cloud is reconstructed using a 3D laser scanner, and the output format is .ply, which includes the coordinates (x, y, z) and color information (R, G, B) of the points.

[0024] 1.2 Statistical Outlier Removal: The original point cloud contains isolated outliers caused by specular reflection, environmental noise, etc. This embodiment uses Statistical Outlier Filtering (SOR) for noise reduction. The parameter settings are as follows: The number of neighboring points k is set to 50, meaning that for each point, its 50 nearest neighbors are searched.

[0025] Standard deviation threshold (std): Set to 2.0. Calculate the distance from each point to its k-neighborhood; all distances follow a Gaussian distribution. If the average distance of a point is greater than the global mean plus 2.0 times the standard deviation, it is identified as an outlier and removed.

[0026] Step S2: 3D point cloud semantic segmentation and functional defect identification based on attention mechanism.

[0027] Develop a semantic segmentation method for 3D point clouds of drainage pipes based on attention mechanism to achieve end-to-end accurate identification and classification of functional defects such as deposits, scale, and obstacles.

[0028] Specifically, in step S2, an encoder-decoder point cloud semantic segmentation network based on an attention mechanism is constructed, such as... Figure 2 As shown, specifically: 2.1 Data Organization: The preprocessed point cloud on the inner surface of the pipe is divided into several overlapping point cloud blocks along the axial direction. 4096 points are randomly sampled from each block as network input. The input channels of each point are coordinates (x, y, z) and color (R, G, B), a total of 6 dimensions. The training set, validation set and test set are divided in a 6:2:2 ratio, and data augmentation such as random rotation, jitter and scaling are performed.

[0029] 2.2 Encoding Stage (Downsampling): Four sets of ensemble abstraction modules are set. The number of sampling center points of the farthest point in layers 1-4 is 1024, 256, 64, and 16 respectively; the sphere query radius is 0.05 m, 0.10 m, 0.20 m, and 0.40 m respectively; and the maximum number of points K in the neighborhood of each sphere is 32, 32, 16, and 16 respectively. For each local neighborhood, the coordinates of the neighborhood points are first converted to local relative coordinates by subtracting the coordinates of the sphere center and then concatenated with its features. Then, the coordinates are gradually increased in dimensionality by a multilayer perceptron (64, 64, and 128 channels respectively), and finally aggregated into the feature vector of the local region by channel-wise max pooling. Residual connections are introduced in each feature extraction module. The module input is aligned with the channels by one-dimensional convolution and then added to the module output to alleviate gradient vanishing caused by network deepening and enhance feature representation ability.

[0030] 2.3 Decoding Stage (Upsampling): Four sets of feature propagation modules are set up symmetrically to those in the encoding stage. For a point to be interpolated, the k=3 closest known points in the previous layer are retrieved, and the features of these 3 points are weighted and interpolated using the reciprocal of the square of the distance as the weight. The point cloud resolution is restored layer by layer until the original number of points is reached.

[0031] 2.4 Cross-layer fusion and attention calibration: Drawing on the feature pyramid network structure, the multi-scale features extracted from each layer in the encoding stage are spliced ​​together with the interpolated features of the corresponding layers in the decoding stage via skip connections in the channel dimension, achieving effective integration of shallow high-resolution geometric details and deep semantic information. A channel-space dual-path attention mechanism is embedded at each skip connection. First, the average pooling and max pooling branches generate channel attention weights through a shared multilayer perceptron to calibrate the features in the channel dimension. Then, the average pooling and max pooling results in the channel dimension are convolved to generate spatial attention weights to calibrate the features in the spatial dimension. This adaptively calibrates the local feature response, suppresses redundant responses in normal pipe wall areas, and guides the network to focus on functional defect areas such as deposition, scaling, and obstacles.

[0032] 2.5 Point-by-point classification: The fused and calibrated point-by-point features are sequentially processed by one-dimensional convolution (128 channels), batch normalization, ReLU activation and Dropout (dropout rate 0.5), and then mapped to the category dimension by one-dimensional convolution. Finally, Softmax outputs the probability of each point belonging to one of the four categories: deposition, scaling, obstruction and normal pipe wall. The one with the highest probability is taken as the semantic label of the point.

[0033] 2.6 Training and Optimization: To address the class imbalance caused by the significantly smaller number of functional defect points compared to normal pipe wall points, a weighted cross-entropy loss function with the reciprocal of the number of points in each class was used. The Adam optimizer was selected, with an initial learning rate of 0.001, which decayed to 0.7 times the original rate every 20 training epochs. The batch size was set to 16, and a total of 200 training epochs were performed. The model with the optimal average intersection-union ratio (IUR) on the validation set was used as the final model. Testing showed that the average IUR for segmenting the three types of functional defects—deposition, scaling, and obstructions—in this embodiment reached over 85%, achieving end-to-end accurate identification and classification of functional defects. Point cloud subsets of various functional defects could be obtained without manual segmentation, providing accurate defect point cloud input for the subsequent geometric parameter quantification calculation in step S5.

[0034] Step S3: Fitting the cylindrical geometric model and establishing the reference for the water passage section.

[0035] The cylindrical geometric model of the inner surface of the pipe is fitted based on the random sampling consensus algorithm, the axial direction and radius are extracted, and the design water passage cross-section benchmark is established.

[0036] Specifically, in step S3, RANSAC is used to fit a cylindrical geometric model to the point cloud of the normal pipe wall, and the axial direction and radius are extracted. The equation of the cylinder axis is expressed as: ; Where (l, m, n) are the axial direction vectors, the equation of the cylindrical surface is expressed as: ; Where P(x, y, z) represents a point on the surface of the cylinder, P(x0, y0, z0) represents a point on the axis of the cylinder, and r0 represents the radius of the water-crossing section of the point cloud model.

[0037] A plane passing through the origin and perpendicular to the axis is used as the reference for the water passage cross-section, providing a geometric reference for quantifying the water passage cross-sectional loss; the formula is expressed as: Where A0 is the area of ​​the cross-section of the water passage.

[0038] Step S4: Voxel-actual measurement calibration.

[0039] A voxel-to-physical measurement calibration method based on proportional relationships is established to convert point cloud voxel dimensions into actual physical dimensions.

[0040] Specifically, in step S4, the calibration coefficient is calibrated using the ratio of the known design radius of the pipeline to the reconstructed radius, and a voxel-actual measurement calibration relationship is established: This converts the point cloud voxel dimensions into actual physical space dimensions.

[0041] Step S5: Quantitative calculation of geometric parameters of functional defects.

[0042] For the identified functional defect point cloud, key quantitative indicators such as deposition thickness, water flow cross-sectional loss rate, and longitudinal clearance are calculated based on cylindrical projection and triangulation. Figure 3 As shown.

[0043] Specifically, step S5 is as follows: 5.1 Quantification of Cross-sectional Loss: The identified functional defect point cloud is cylindrically projected along the axis and triangulated to calculate the cross-sectional area occupied by the defect, thereby obtaining the water flow cross-sectional loss rate. The calculation formula is as follows: ; ; Among them, S Δk The area of ​​each triangular mesh, where K is the total number of triangular meshes. voxel Let area be the voxel area of ​​the cross-section of the water passage. real This refers to the actual area of ​​the cross-section of the water passage. 5.2 Size Quantification: For deposit and scaling defects, the thickness between the top surface of the defect and the pipe wall is calculated radially and the deposition thickness ratio is characterized as a percentage of the pipe diameter. The calculation formula is as follows: ; ; Where r is the distance from each deposit / scaling functional defect point to the fitted cylinder axis, N is the number of defect points, and t voxel t represents the voxel thickness of the defect. real This represents the actual thickness of the defect.

[0044] 5.3 Clearance Quantification: Calculate the longitudinal clearance of the defect section along the pipeline axis and output the key quantitative indicators of each functional defect.

[0045] Step S6: Full-scale testing to verify and align with standard-compliant intelligent assessment of functional defects.

[0046] Design a full-scale test model and conduct comparative tests between calculated data and actual data; based on the functional defect level classification standard of the "Technical Specification for Inspection and Evaluation of Urban Drainage Pipelines", integrate the calculation formula of pipe section functional defect parameters and maintenance index to form an intelligent quantitative evaluation method for functional defects that is in line with the current specifications.

[0047] Specifically, step S6 is as follows: 6.1 Full-scale verification: Design a full-scale test model and conduct comparative tests on the calculated data and actual measured data for quantitative indicators such as sediment thickness, cross-sectional loss rate, and longitudinal net distance, and evaluate the accuracy of the method with relative error; 6.2 Grading: Based on the functional defect grading standards of the "Technical Specification for Inspection and Evaluation of Urban Drainage Pipelines", the quantitative indicators are graded. 6.3 Exponential Integration: such as Figure 4 As shown, the integrated pipe section functional defect parameters and maintenance index calculation formulas form an intelligent quantitative assessment method for drainage pipeline functional defects that is in line with current standards, and output functional defect levels and maintenance decision recommendations.

[0048] This invention achieves end-to-end identification and classification of functional defects through three-dimensional point cloud semantic segmentation based on attention mechanism. It establishes a design water passage cross-section benchmark by cylinder fitting, and combines cylinder projection, triangulation and voxel-actual measurement calibration to quantify deposition thickness, water passage cross-section loss rate and longitudinal clearance. Combined with full-scale test verification and current specifications, it forms an intelligent quantitative assessment method for functional defects that is in line with the "Technical Specification for Inspection and Evaluation of Urban Drainage Pipelines".

[0049] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0050] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0051] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0052] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0053] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent quantification and evaluation of functional defects in drainage pipes based on three-dimensional point clouds, characterized in that, Includes the following steps: S1. Acquisition and preprocessing of 3D point cloud of inner surface of drainage pipe; S2. Attention-based semantic segmentation and functional defect identification of 3D point clouds; S3. Fitting the cylindrical geometric model and establishing the reference for the water passage section; S4, Voxel-actual measurement calibration; S5. Quantitative calculation of geometric parameters of functional defects; S6. Full-scale testing verification and standard-compliant intelligent assessment of functional defects.

2. The intelligent quantification and evaluation method for functional defects of drainage pipelines based on three-dimensional point clouds according to claim 1, characterized in that: In step S1, the three-dimensional point cloud of the inner surface of the drainage pipe is obtained by multi-view three-dimensional reconstruction or three-dimensional laser scanning, and denoising preprocessing is performed by statistical outlier removal method.

3. The intelligent quantification and evaluation method for functional defects of drainage pipelines based on three-dimensional point clouds according to claim 1, characterized in that: In step S2, an encoder-decoder semantic segmentation network with an attention mechanism is constructed to perform multi-class semantic annotation on each point in the point cloud, and to identify and distinguish different types of functional defects such as deposits, scale, and obstacles end-to-end; specifically, it includes the following: S21. Encoding stage: The farthest point sampling algorithm is used to perform hierarchical downsampling on the input point cloud to obtain the sampling center point. The ball query algorithm is used to divide the local region with each sampling center point as the center of the ball. The geometric features of each local region are extracted by the multilayer perceptron and aggregated into local feature vectors by max pooling. Residual connections are introduced in each feature extraction module to alleviate gradient vanishing. S22. Decoding stage: The multi-scale abstract features generated in the encoding stage are upsampled layer by layer using nearest neighbor-based weighted interpolation to gradually restore the original resolution of the point cloud; S23, Cross-layer fusion: Multi-scale features extracted from each layer in the encoding stage are fused with corresponding layer features in the decoding stage via skip connections, and a channel-space dual-path attention mechanism is embedded at each skip connection to adaptively calibrate local feature responses and guide the network to focus on functional defect regions; S24. Point-by-point classification and training: The fused features are mapped to the probability of each point belonging to the category of deposition, scaling, obstacles and normal pipe wall through one-dimensional convolution. The point with the highest probability is taken as the semantic label of the point. Based on the three-dimensional point cloud database of functional defects in drainage pipes, the network is trained and optimized with weighted cross-entropy loss. The optimal combination of hyperparameters and attention mechanism configuration are selected to output the point cloud subsets of various functional defects.

4. The intelligent quantification and evaluation method for functional defects of drainage pipelines based on three-dimensional point clouds according to claim 1, characterized in that: In step S3, a method for fitting the cylindrical geometric model of the inner surface of the drainage pipe based on the random sampling consensus RANSAC algorithm is established. The pipe axis direction and radius parameters are extracted, and a plane passing through the origin and perpendicular to the axis is used as the reference for the water passage section. Specifically, the method includes the following: S31. Fit a cylindrical geometric model to the point cloud of a normal pipe wall using the RANSAC algorithm, and extract the axis direction and radius; wherein, the equation of the cylinder axis is expressed as: ; Where (l, m, n) are the axial direction vectors, the equation of the cylindrical surface is expressed as: ; Where P(x, y, z) represents a point on the surface of the cylinder, P(x0, y0, z0) represents a point on the axis of the cylinder, and r0 represents the radius of the water passage section of the point cloud model; S32. Taking the plane passing through the origin and perpendicular to the axis as the reference for the water passage section, the formula is expressed as: Where A0 is the area of ​​the cross-section of the water passage.

5. The intelligent quantification and evaluation method for functional defects of drainage pipelines based on three-dimensional point clouds according to claim 1, characterized in that: In step S4, a voxel-actual measurement calibration method based on proportional relationships is established. The calibration coefficient is calibrated using the ratio of the known design radius of the pipeline to the reconstructed radius, converting the point cloud voxel dimensions into actual dimensions in physical space. The calibration relationship is as follows: ; Where, r real Given the design radius as shown in the design drawings, the coefficient conversion This is the coefficient used to convert the point cloud voxel size to the actual size.

6. The intelligent quantification and evaluation method for functional defects of drainage pipelines based on three-dimensional point clouds according to claim 1, characterized in that: In step S5, for the identified functional defect point cloud, a geometric parameter calculation model based on cylindrical projection and triangulation is proposed to calculate key quantitative indicators such as deposition thickness, water flow cross-sectional loss rate, and longitudinal net distance; specifically including the following: S51. For the identified functional defect point cloud, project the four types of defect points—scale, obstacles, residual walls, and tree roots—onto the cross-section of the defect to construct a triangular mesh. Sum the mesh areas to obtain the voxel water flow cross-sectional loss of the defect. The calculation formula is as follows: ; ; Among them, S Δk The area of ​​each triangular mesh, where K is the total number of triangular meshes. voxel Let area be the voxel area of ​​the cross-section of the water passage. real This refers to the actual area of ​​the cross-section of the water passage. S52. For functional defects such as deposition and scaling, calculate the thickness between the top surface of the defect and the pipe wall radially and characterize the deposition thickness as a percentage of the pipe diameter. The formula for calculating the deposition thickness ratio is as follows: ; ; Where r is the distance from each deposit / scaling functional defect point to the fitted cylinder axis, N is the number of defect points, and t voxel t represents the voxel thickness of the defect. real The actual thickness of the defect; S53. Based on the defect point cloud segmentation results and the cylinder equation, determine the longitudinal center position of each defect instance and obtain the cross-sectional equation passing through the longitudinal center; for multiple defects of the same type in the same pipe section, calculate the distance between the longitudinal centers of each defect, which is the longitudinal net distance of the voxel.

7. The intelligent quantification and evaluation method for functional defects of drainage pipelines based on three-dimensional point clouds according to claim 1, characterized in that: In step S6, a full-scale test model is designed to conduct comparative tests on the calculated data and actual measured data of quantitative indicators such as sediment thickness, cross-sectional loss rate, and longitudinal net distance, and the accuracy of the method is evaluated by relative error.