Three-view underground pipeline intelligent detection method based on three-dimensional ground penetrating radar

Abstract

The invention discloses a three-view underground pipeline intelligent detection method based on a three-dimensional ground penetrating radar. According to a traditional method, only a universal model is used for recognizing a radar image of a single B-scan longitudinal section feature image, and an existing model does not perform targeted adjustment on ground penetrating radar pipeline image features, so that the image features of cavity diseases and pipelines cannot be accurately distinguished, and the recognition accuracy is low. According to the three-view underground pipeline intelligent detection method, after acquired underground three-dimensional data is combined with a DCO-YOLO model, the space overlapping degree of marking frames in three views and matched pipeline feature information are obtained based on a three-dimensional distance intersection-to-union ratio algorithm; and completing the rapid identification of pipeline features and the matching process of the three views by using the space overlapping degree of the marking frames in the three views and the matching pipeline feature information.

Description

Technical Field

[0001] The present invention specifically relates to an intelligent underground pipeline detection method based on three-dimensional ground penetrating radar and three-view images, belonging to the technical field of underground pipeline detection. Background Art

[0002] As core infrastructure for maintaining water, drainage, gas, and communications, the structural integrity of urban underground pipelines is directly linked to the safe operation of cities. However, due to factors such as pipe aging, design flaws, complex environmental loads, and inadequate maintenance, pipeline systems are prone to structural damage such as leakage and displacement, which in turn can lead to secondary disasters such as road collapse and urban waterlogging.

[0003] Traditional pipeline detection methods primarily rely on electromagnetic induction and are suitable for detecting metallic pipelines. While traditional pipeline-internal inspection technologies have some detection capabilities, they are not directly applicable to non-metallic pipelines, such as plastic and concrete pipes, due to their lack of conductivity. This increases the difficulty of detection. Furthermore, traditional pipeline detection suffers from inherent drawbacks such as poor adaptability to operating environments and low detection efficiency, making it difficult to meet the inspection needs of large-scale pipeline networks in modern cities. With technological advancements, more and more pipelines are being laid using trenchless methods, such as directional drilling and pipe jacking. While this method reduces the damage caused by ground excavation, it also makes it difficult to determine the exact location of the pipeline using traditional ground survey methods. Furthermore, urban environments are plagued by numerous electromagnetic interference sources, such as high-voltage power lines and radio towers, which can interfere with pipeline detection signals and affect the accuracy of detection results. Ground conditions, such as metal guardrails and traffic conditions, can also affect pipeline detection. Metallic objects, such as metal guardrails, can generate electromagnetic interference, while heavy traffic areas can cause unstable detection signals due to factors such as vehicle vibration. In view of this situation, the development of efficient pipeline detection technology is of great engineering significance for evaluating the health status of the system and formulating maintenance plans.

[0004] Ground-penetrating radar technology, with its non-contact electromagnetic detection characteristics, can achieve the organic unity of spatial positioning and structural assessment of underground pipelines by analyzing the reflected signals formed by differences in the dielectric constants of the strata. However, its single-antenna structure has obvious limitations in the three-dimensional analysis of complex pipeline networks.

[0005] Three-dimensional ground-penetrating radar (GPR) uses a multi-channel array antenna system to simultaneously acquire B-scan (longitudinal profile), C-scan (horizontal slice), and D-scan (transverse profile) data, significantly improving the three-dimensional characterization capabilities of pipeline networks. However, the massive amount of data leads to increasingly prominent problems such as low efficiency and subjectivity in manual interpretation. To overcome these limitations, the application of existing deep learning technologies in GPR image recognition is gaining momentum. However, existing technologies suffer from low efficiency in three-dimensional data interpretation and insufficient fusion of multi-view features. Traditional methods only use general models to recognize single B-scan (longitudinal profile) radar images. Existing models lack specific adjustments to GPR pipeline image features, making it impossible to accurately distinguish between cavity defects and pipeline image features, resulting in low recognition accuracy. Currently, three-dimensional GPR underutilizes the three-view features for underground pipeline inspection. Using B-scan images alone is ineffective for effectively identifying the three-dimensional direction characteristics of underground pipelines. This leaves room for improvement, and detection range blind spots affect the quality and reliability of inspection results. Summary of the Invention

[0006] In order to overcome the defects of the existing technology, a three-dimensional ground penetrating radar-based three-view underground pipeline intelligent detection method is provided to solve the above problems.

[0007] A three-view intelligent detection method for underground pipelines based on 3D ground-penetrating radar is proposed. The method combines the acquired underground 3D data with the DCO-YOLO model and then uses the 3D distance intersection-to-union algorithm to derive the spatial overlap of the annotated boxes in the three views and the matching pipeline feature information. The spatial overlap of the annotated boxes in the three views and the matching pipeline feature information are then used to complete the rapid identification of pipeline features and the three-view matching process.

[0008] As a preferred solution: underground three-dimensional data of the actual road is collected by three-dimensional ground penetrating radar, and a preprocessing process is performed after obtaining the underground three-dimensional data. The preprocessing process includes inverse selective discrete Fourier transform, background removal filtering, progressive low-pass filtering, and gain adjustment; the preprocessing process also includes homogenization adjustment of the three-dimensional radar data. The homogenization adjustment process is to obtain a three-perspective image set of B-scan longitudinal profile feature image, C-scan horizontal slice feature image, and D-scan transverse profile feature image by multi-perspective decomposition of the radar data. The image parameters of the three-perspective image set are depth 3m, span 30m, and vertical span direction 1.32m. The three-perspective image set is adjusted to adjust the image size, optimize the contrast, remove irrelevant patterns, and remove text annotations through standardization processing to form homogenized data, and the homogenized data is used as input image data.

[0009] As a preferred solution: after preprocessing the underground three-dimensional data, the DCO-YOLO model that matches the underground three-dimensional data is adapted. The adaptation process is to improve the DCO-YOLO model to form an initial model that is adapted to the preprocessed homogenized data, and the model is processed according to the spatial position of the underground pipeline corresponding to the actual road. Specifically, based on the direction of the underground pipeline, the data set is divided into different annotation categories during the initial model training. Among them, the pipeline buried depth is calculated based on the pipeline B-scan longitudinal profile feature image, and the pipeline buried depth is obtained as the distance from the pipeline B-scan longitudinal profile feature image to the ground.

[0010] As a preferred solution, the input image data is restored before calculating the 3D-DIoU. The restoration process is as follows:

[0011] Based on the positions of the three views in the input image data, a finite three-dimensional space R with a length of L, a width of W, and a height of H is constructed. The relative positions of the predicted recognition boxes in the corresponding views are kept unchanged. Each recognition box contains three-dimensional data (x, y, z) in the three-dimensional space R. When the input image data is the x data of the pipeline feature recognition box 1-B on the B-Scan view, the corresponding conversion formula is:

[0012]

[0013] In the above formula, x B1 and x B2 Respectively represent the coordinate values ​​of the left and right edges of the 1-B annotation box converted to space R; X B1 and X B2 They represent the coordinate values ​​of the left and right edges of the 1-B annotation box in the input image respectively; main_view(X,Y,width,height) represents the coordinates and size of the upper left vertex of the B-Scan.

[0014] As a preferred solution: each identification frame ensures the corresponding supplementation of one dimension of data when mapped to space R, and the supplementation processing process is: when the identification frame is the pipeline B-scan longitudinal section feature image identification frame, the supplementation process is that the pipeline B-scan longitudinal section feature image identification frame contains the spatial information of the pipeline width x and height z, and lacks the spatial information of the length y, and the corresponding y information is supplemented. The corresponding processing and supplementation are analogous, that is, the z information of the B-scan longitudinal section feature image is given to the C-scan horizontal slice feature image, and the x information of the C-scan horizontal slice feature image is supplemented to the D-scan transverse section feature image.

[0015] As a preferred solution, the process of obtaining the spatial overlap of the annotation boxes in the three views and matching the pipeline feature information based on the 3D distance intersection ratio algorithm is as follows:

[0016] The three-view matching process is judged by calculating and comparing 3D-DIoU. 3D-DIoU is used to reflect the spatial correlation after the annotation boxes of the three view features of the pipeline feature are mapped to the three-dimensional space R. The calculation formula of 3D-DIoU is:

[0017]

[0018] In the above formula, 3D-DIoU is the ratio of the intersection volume and the union volume of two 3D boxes; d is the Euclidean distance between the center points of the two 3D boxes; c is the diagonal length of the minimum circumscribed cube covering the two 3D boxes;

[0019] The 3D-DIoU value of each annotation frame mapped to the three-dimensional space is calculated respectively, and the 3D-DIoU value of each annotation frame mapped to the three-dimensional space is compared with its corresponding set threshold. When the threshold conditions are met, it indicates that the three identification frames are the B-scan longitudinal section feature image, C-scan horizontal slice feature image, and D-scan transverse section feature image corresponding to the same pipeline, that is, the three view frames are in a successful matching state.

[0020] The beneficial effects of the present invention are:

[0021] The present invention combines the acquired underground 3D data with the DCO-YOLO model and then uses the 3D distance intersection-to-union ratio algorithm to derive the spatial overlap of the annotated boxes in the three views and the matching pipeline feature information, forming a detailed and complete processing process for the underground 3D data. The spatial overlap of the annotated boxes in the three views and the matching pipeline feature information are used to complete the rapid identification of pipeline features and the three-view matching process, completing the process of accurately distinguishing the image features of cavity diseases and pipelines with high recognition accuracy. The present invention's prediction and feature matching of 3D radar data can accurately and efficiently detect the distribution and operating status of underground pipelines, thereby providing a basis for their health assessment and maintenance planning. The specific advantages of the present invention are:

[0022] First, the model of this application improves YOLOv11 by replacing the original uu.Upsample module with Dysample to improve the model's sensitivity to edge features in ground-penetrating radar images. CGLU is introduced to improve the C2PSA module to enhance the local details of pipeline features in ground-penetrating radar images. Outlook Attention is introduced to improve the C3K2 module, allowing the model to pay more attention to the surrounding linear changes of the pipeline waveform and suppress background noise unrelated to the central features, ultimately achieving the goal of improving detection accuracy.

[0023] Second: This application proposes the idea of ​​mapping the three-view features of the B-scan longitudinal section feature image, the C-scan horizontal slice feature image, and the D-scan transverse section feature image into three-dimensional space, especially introducing 3D-DIoU to estimate the relationship between the spatial positions of each view, which can achieve precise matching of the features of each view, and integrate the features of the three views to determine the direction of the pipeline, making the detection results more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a flowchart of an intelligent underground pipeline detection method based on three-dimensional ground penetrating radar.

[0025] Figure 2 This is a schematic diagram of the DCO-YOLO underground pipeline three-view intelligent recognition algorithm structure;

[0026] Figure 3 This is a schematic diagram of the Dysample structure;

[0027] Figure 4 Schematic diagram of the structure of GLU and CGLU;

[0028] Figure 5 This is a diagram of the OutlookAttention structure;

[0029] Figure 6 3D-DIoU schematic diagram;

[0030] Figure 7 This is a visual comparison diagram of the DCO-YOLO underground pipeline identification network of the present invention and other methods;

[0031] Figure 8 This is a visualization diagram of pipeline characteristics and spatial position prediction according to the present invention. DETAILED DESCRIPTION

[0032] The following describes the embodiments of the present invention through specific examples. Those skilled in the art will readily understand the other advantages and benefits of the present invention from the disclosure herein. The present invention may also be implemented or applied through various other specific embodiments, and the details in this specification may be modified or altered based on different viewpoints and applications without departing from the spirit of the present invention.

[0033] Specific implementation method 1: Combination Figures 1 to 8This embodiment describes the intelligent detection method for underground pipelines in three views in this embodiment. After the acquired underground three-dimensional data is combined with the DCO-YOLO model, the spatial overlap of the annotation boxes in the three views and the matching pipeline feature information are obtained based on the three-dimensional distance intersection-over-union algorithm. The spatial overlap of the annotation boxes in the three views and the matching pipeline feature information are used to complete the rapid identification of pipeline features and the three-view matching process. The three-dimensional distance intersection-over-union algorithm in this embodiment is the 3D-DIoU algorithm.

[0034] Combine Figure 1 As shown, the specific operation process of the present invention is:

[0035] First, the data comes from the underground 3D data under the actual road. The collected 3D ground penetrating radar data is preprocessed, specifically including background removal filtering, progressive low-pass filtering, gain adjustment, and other necessary preprocessing processes.

[0036] Secondly, based on pipeline direction characteristics, the preprocessed data is annotated and divided into training and validation sets. The core C3k2, C2PSA, and Upsample modules in the YOLOv11 framework are improved, and the improved model is used for training and prediction.

[0037] Then, the two-dimensional view of the predicted three-dimensional ground penetrating radar image is restored to three-dimensional space, and the three-view feature matching of the pipeline is achieved based on 3D-DIoU;

[0038] Finally, the predicted matching pipeline features and the remaining pipeline features are drawn with different color identification boxes, and the corresponding pipeline type and burial depth are output.

[0039] The present invention collects underground data through three-dimensional ground penetrating radar, and generates a three-view image set of B-Scan, C-Scan, and D-Scan through preprocessing; adopts the DCO-YOLO model for feature recognition, maps the three-view recognition frame to three-dimensional space based on the three-dimensional distance intersection-to-union algorithm, calculates the spatial overlap to match the pipeline features, and realizes the pipeline direction classification and burial depth calculation process.

[0040] The present invention has been demonstrated through experiments, which show that in the specific implementation of the present invention, in 100 kilometers of measured data, the detection accuracy P, recall rate R and average accuracy reached 96.2%, 93.3% and 96.7% respectively. The data information is significantly better than the traditional model, and can provide an efficient and high-precision corresponding solution for the intelligent detection of underground pipelines.

[0041] Specific embodiment 2: This embodiment is a further limitation of specific embodiment 1. In this embodiment, the three-view underground pipeline intelligent detection method is specifically divided into the following steps:

[0042] Step 1: Use 3D ground-penetrating radar to collect 3D underground data of the actual road. Preprocess the three-dimensional data to generate a three-view image set. Specifically, preprocess the raw data collected by the frequency-stepped 3D ground-penetrating radar, including inverse discrete Fourier transform, background removal filtering, progressive low-pass filtering, and gain adjustment.

[0043] Among them, according to the frequency domain signal characteristics of the frequency-stepping 3D radar, the frequency domain signal is converted into a time domain signal through the inverse discrete Fourier transform to optimize the time domain resolution. The formula is:

[0044]

[0045] In the above formula, S(f k ) is the kth frequency component, N is the number of frequency sampling points, and s(t) is the time domain signal.

[0046] Background removal filtering eliminates interference caused by antenna direct coupling, surface reflections, and system noise, highlighting effective reflection features. Progressive low-pass filtering suppresses high-frequency electronic noise and low-frequency surface reflections, eliminating periodic noise and smoothing random noise. Gain adjustment compensates for signal attenuation, enhancing visibility of deep, weak targets and ensuring comparability of reflected energy from different media in the image.

[0047] Step 2: Label the preprocessed data and divide it into training set and validation set, specifically:

[0048] The dataset is divided into different annotation categories based on the pipeline direction. The B-scan longitudinal profile feature image corresponding to vertical pipelines is 1-B, the C-scan horizontal slice feature image is 1-C, and the D-scan transverse profile feature image is 1-D. The B-scan longitudinal profile feature image corresponding to horizontally inclined pipelines is 1-B, the C-scan horizontal slice feature image is 2-C, and the D-scan transverse profile feature image is 2-D. The B-scan longitudinal profile feature image corresponding to depth-oblique pipelines is 1-B, the C-scan horizontal slice feature image is 3-C, and the D-scan transverse profile feature image is 3-D.

[0049] Since there is no obvious difference in the corresponding B-Scan hyperbolic features regardless of the pipeline direction, the corresponding B-scan longitudinal section feature images of the three types of pipelines are all 1-B.

[0050] LabelImg software was used to annotate multi-view GPR images and extract pipeline geometric features and spatial attributes. Pre-training was used to verify annotation quality and optimize feature recognition accuracy. The dataset was divided into training and validation sets in an 8:2 ratio. Annotation files were uniformly converted to .txt format to facilitate subsequent training of the target recognition model.

[0051] Step 3: Combine Figure 2 As shown, training and prediction are performed based on the DCO-YOLO model. The core Upsample, C2PSA, and C3k2 modules in the original YOLOv11 framework are improved by introducing DySample dynamic upsampling, CGLU local feature perception, and OutlookAttention cross-dimensional correlation mechanisms. This significantly improves the ability to represent small-scale target edge features, allowing the model to focus more on local details and edge features of pipeline waveforms. Training and prediction using the improved DCO-YOLO model consists of the following steps:

[0052] Part I: Combination Figure 3 As shown in the figure, the DySample dynamic upsampling module is used to replace the traditional Upsample upsampling, and content-aware feature reconstruction is achieved through two-stage offset generation and pixel rearrangement. For the input low-resolution feature map X∈R H×W×C DySample first directly generates the position offset of each target sampling point through a lightweight linear layer (LinearLayer) Where s is the upsampling multiple. The dimension of the offset is 2s 2 A dynamic adjustment mechanism is introduced to generate a dynamic range factor through a linear layer and a Sigmoid function. The formula for O is:

[0053] O=0.5sigmoid(linear1(X)Linear2(X)) (1)

[0054] In the above formula, the sigmoid function, σ in the figure, can map any real input to the range of (0,1) to control the range of sampling points.

[0055] Pixel Shuffle redistributes information from the channel dimension into the spatial dimension, achieving upsampling. The offset O is superimposed on the standard upsampling grid G ​​to form a dynamic sampling grid S. The original grid G ​​represents the fixed sampling locations for bilinear interpolation. The grid_sample function is then used to resample the input feature X according to the dynamic grid S, yielding the upsampled feature X′.

[0056] X'=grid_sample(X,S) (2)

[0057] Part II: Combination Figure 4The structural flow shown in the figure introduces a convolutional gated linear unit (CGLU) in the C2PSA module. It extracts neighborhood features through deep convolution to generate position-dependent channel weights. It includes a bilinear transformation branch: a value branch, where the input undergoes a linear transformation to generate primary features, preserving global information; and a gate branch, where the input undergoes a linear transformation and is then activated by a GELU (Gaussian Error Linear Unit). However, the weights of the gate branch are derived from the input itself, resulting in a limited receptive field. The improved CGLU (Convolutional GLU), based on the GLU, adds a 3×3 deep convolution before the GLU's gate branch, leveraging neighboring pixel features to generate a locally perceptual gating signal. Deep convolution extracts features surrounding the token, assigning each position a unique channel attention weight, addressing the coarse-grained nature of SE.

[0058] The output of the gating branch (local features) is multiplied element-by-element with the output of the value branch (global features) to generate fine-grained channel attention weights. The specific formula is as follows.

[0059] ConvGLU(X)=Linear1(X)⊙GELU(DepthwiseC onv(Linear2(X)))(3)

[0060] In the above formula, DepthwiseCONV (DWConv) represents a 3×3 depthwise convolution, and ⊙ represents element-wise multiplication. CGLU not only focuses on the global features of the input image but also observes the local features surrounding each pixel, dynamically adjusting channel weights at different locations. CGLU introduces a 3×3 depthwise convolution in the gating branch to dynamically extract neighborhood features and generate position-dependent channel weights. This module can enhance edge and phase transition features in different regions of the radar signal's time-frequency image.

[0061] Part III: Combination Figure 5 As shown, the OutlookAttention mechanism is embedded in the C3K2 module, which aggregates multi-granularity features based on the local window attention and performs the input feature map X∈R H×W×C The value V and attention weight A are generated by linear projection, and the local window (K×K) of each center token is extracted using the unfold operation to obtain The central token is generated through the linear layer After reshaping, it is normalized by the Softmax function and local features are aggregated by matrix multiplication.

[0062]

[0063] In the above formula, Y Δi,jRepresents the local features after aggregation, and Matmul is a matrix multiplication function that sums the attention weight and value.

[0064] The aggregation result of the local window is restored to the global feature map through the folding operation. The formula is as follows.

[0065]

[0066] In the above formula, the right side input is the aggregation result Y of (i, j) Δi,j The final output is obtained by summing the aggregation results of all local windows covering position (i, j)

[0067] The accuracy, recall and average precision are used to evaluate the detection effect of the model.

[0068] Step 4: Implement the recognition and processing of pipeline features in the three-view image based on the DCO-YOLO model. Specifically, use the DCO-YOLO trained in step 3 to accurately identify targets in the three-view image of underground pipelines obtained by the 3D ground penetrating radar.

[0069] Step 5: Restore the three 2D views of the recognized 3D GPR image to 3D space and match the pipeline's three-view features based on 3D-DIoU. Specifically:

[0070] Based on the positions of the three views in the input image (the horizontal coordinate X of the upper left vertex of the view, the vertical coordinate Y of the upper left corner of the view, the width of the view, and the height of the view), a finite three-dimensional space R with a length of L, a width of W, and a height of H is constructed. The relative positions of the predicted recognition boxes in the corresponding views are kept unchanged, and each recognition box contains three-dimensional data (x, y, z) in the three-dimensional space R. Taking the x data of the pipeline feature recognition box 1-B on the B-Scan view as an example, the conversion formula is:

[0071]

[0072] In the above formula, x B1 ,x B2 , respectively represent the coordinate values ​​of the left and right edges of the 1-B annotation box converted to space R. B1 , X B2 They represent the coordinate values ​​of the left and right edges of the 1-B annotation box in the input image respectively. main_view(X,Y,width,height) stores the coordinates and size of the upper left vertex of the B-Scan.

[0073] The identification box 1-B on the main view contains the spatial information of the pipe's width (x) and height (z). It lacks the spatial information of its length (y). This information can be used to supplement the y information of nD. Similarly, the z information of 1-B can be supplemented for nC, and the x information of nC can be supplemented for nD.

[0074] Furthermore, the process of determining the matching of three views by calculating and comparing 3D-DIoU includes:

[0075] Combine Figure 6 As shown in Figure 2, 3D-DIoU is used to reflect the spatial correlation after 1-B, nC, and nD are mapped to the three-dimensional space R. The calculation formula of 3D-DIoU is:

[0076]

[0077] In the above formula, 3D IoU is the ratio of the volume intersection of two 3D boxes to their volume union, d is the Euclidean distance between the center points of the two 3D boxes, and c is the diagonal length of the smallest circumscribed cube covering the two 3D boxes.

[0078] The 3D-DIoU values ​​of 1-B and nC, and 1-B and nD mapped into three-dimensional space are calculated respectively and compared with the set threshold of 0.4. When the threshold conditions are met, it means that the three identification frames are the B-scan longitudinal section feature image, C-scan horizontal section feature image, and D-scan transverse section feature image corresponding to the same pipeline, that is, the three view frames match.

[0079] Step 6: Visualize the detection results. When a set of matching pipe features is identified in the input image, the buried depth is calculated based on the position of the 1-B annotation box. The buried depth is represented by the actual distance from the upper edge of the 1-B annotation box to the ground. The predicted matching pipe features are drawn with different colored identification boxes from the remaining pipe features. Based on the category of the matching pipe features, the corresponding pipe type and buried depth are output.

[0080] Specific embodiment three: This embodiment is a further limitation of specific embodiment one or two. In this embodiment, underground three-dimensional data of the actual road is collected by a three-dimensional ground penetrating radar. After obtaining the underground three-dimensional data, it undergoes a preprocessing process, which includes inverse selective discrete Fourier transform, background removal filtering, progressive low-pass filtering, and gain adjustment; the preprocessing process also includes homogenizing the three-dimensional radar data. The homogenization adjustment process is to obtain a three-perspective image set of B-scan longitudinal profile feature image, C-scan horizontal slice feature image, and D-scan transverse profile feature image by multi-perspective decomposition of the radar data. The image parameters of the three-perspective image set are depth 3m, span 30m, and vertical span direction 1.32m. The three-perspective image set is standardized to adjust the image size, optimize the contrast, remove irrelevant patterns, and remove text annotations to form homogenized data, and the homogenized data is used as input image data.

[0081] Specific implementation method four: This implementation method is a further limitation of specific implementation methods one, two or three. In this implementation method, after preprocessing the underground three-dimensional data, the DCO-YOLO model that matches the underground three-dimensional data is adapted. The adaptation process is to improve the DCO-YOLO model to form an initial model that is adapted to the preprocessed homogenized data, and the model is processed according to the spatial position of the underground pipeline corresponding to the actual road. Specifically, based on the direction of the underground pipeline, the data set is divided into different annotation categories during the initial model training, wherein the pipeline buried depth is calculated based on the pipeline B-scan longitudinal profile feature image, and the pipeline buried depth is obtained as the distance from the pipeline B-scan longitudinal profile feature image to the ground.

[0082] Specific embodiment five: This embodiment is a further limitation of specific embodiments one, two, three or four. In this embodiment, the input image data is restored before calculating the 3D-DIoU. The restoration process is to restore the three two-dimensional views of the input image data to a three-dimensional space. The processing process is to construct a finite three-dimensional space R with a length L, a width W, and a height H based on the positions of the three views in the input image, keep the relative position of the predicted recognition box in the corresponding view unchanged, supplement one dimension of information, and convert it into a three-dimensional annotation box in the space R.

[0083] The specific restoration process is as follows:

[0084] Based on the positions of the three views in the input image data, a finite three-dimensional space R with a length of L, a width of W, and a height of H is constructed. The relative positions of the predicted recognition boxes in the corresponding views are kept unchanged. Each recognition box contains three-dimensional data (x, y, z) in the three-dimensional space R. When the input image data is the x data of the pipeline feature recognition box 1-B on the B-Scan view, the corresponding conversion formula is:

[0085]

[0086] In the above formula, x B1 and x B2 Respectively represent the coordinate values ​​of the left and right edges of the 1-B annotation box converted to space R; X B1 and X B2 They represent the coordinate values ​​of the left and right edges of the 1-B annotation box in the input image respectively; main_view(X,Y,width,height) represents the coordinates and size of the upper left vertex of the B-Scan.

[0087] Specific embodiment six: This embodiment is a further limitation of specific embodiments one, two, three, four or five. In this embodiment, each identification frame ensures the corresponding supplementation of one dimension of data when mapped to space R. The supplementation processing process is: when the identification frame is the pipeline B-scan longitudinal section feature image identification frame, the supplementation process is that the pipeline B-scan longitudinal section feature image identification frame contains the spatial information of the pipeline width x and height z, and lacks the spatial information of the length y, and the corresponding y information of the D-Scan is supplemented. The corresponding processing and supplementation are carried out in this way, that is, the z information of the B-scan longitudinal section feature image is given to the C-scan horizontal slice feature image, and the x information of the C-scan horizontal slice feature image is supplemented to the D-scan transverse section feature image.

[0088] Specific embodiment seven: This embodiment is a further limitation of specific embodiments one, two, three, four, five or six. In this embodiment, the process of obtaining the spatial overlap of the annotation boxes in the three views and matching the pipeline feature information based on the three-dimensional distance intersection ratio algorithm is as follows:

[0089] The three-view matching process is judged by calculating and comparing 3D-DIoU. 3D-DIoU is used to reflect the spatial correlation after the annotation boxes of the three view features of the pipeline feature are mapped to the three-dimensional space R. The calculation formula of 3D-DIoU is:

[0090]

[0091] In the above formula, 3D-DIoU is the ratio of the intersection volume and the union volume of two 3D boxes; d is the Euclidean distance between the center points of the two 3D boxes; c is the diagonal length of the minimum circumscribed cube covering the two 3D boxes;

[0092] The 3D-DIoU value of each annotation frame mapped to the three-dimensional space is calculated respectively, and the 3D-DIoU value of each annotation frame mapped to the three-dimensional space is compared with its corresponding set threshold. When the threshold conditions are met, it indicates that the three identification frames are the B-scan longitudinal section feature image, C-scan horizontal slice feature image, and D-scan transverse section feature image corresponding to the same pipeline, that is, the three view frames are in a successful matching state.

[0093] When the present invention is applied to the detection of underground pipelines under roads, a frequency-stepped 3D ground-penetrating radar is used to detect underground pipelines over 100 km of urban roads, covering the typical distribution area of ​​urban underground pipeline networks. Pre-processing is performed according to the above steps, and model tests and comparative analysis are carried out.

[0094] Combine Figure 7 The recognition effect is compared and analyzed with the model test results in Table 1, specifically:

[0095] Table 1 Model comparison test results

[0096]

[0097] According to the data in the table above, the DCO-YOLO model used in this paper has significant advantages over other deep learning models in the experiment. It improves the accuracy by 2%, the recall by 2.1%, and the average precision by 0.9 compared with the original YOLOv11 model, verifying the effectiveness of the multi-dimensional feature enhancement strategy.

[0098] Combine Figure 8 As shown in the figure, the pipeline characteristics and spatial position prediction visualization results of the present invention can accurately identify the pipeline direction and burial depth, verifying the feasibility of the present invention.

[0099] Drawing on extensive practical experience and expertise in this field, this paper designs a three-view intelligent detection method for underground pipelines based on 3D ground-penetrating radar (GPR). This method addresses key technical challenges in existing 3D GPR data processing, such as insufficient spatiotemporal feature fusion, inadequate utilization of multi-view correlations, and low recognition accuracy in complex scenes. This method innovatively constructs a dual-strategy collaborative detection framework. Through bidirectional optimization of multi-view feature fusion of 3D data and an improved DCO-YOLO object detection model, it achieves accurate identification and spatial localization of underground pipelines.

Claims

1. A three-dimensional ground penetrating radar-based intelligent underground pipeline detection method, characterized by: The three-view intelligent detection method for underground pipelines is to combine the acquired underground 3D data with the DCO-YOLO model, and then use the 3D distance intersection-to-union algorithm to obtain the spatial overlap of the annotation boxes in the three views and the matching pipeline feature information. The spatial overlap of the annotation boxes in the three views and the matching pipeline feature information are used to complete the rapid identification of pipeline features and the three-view matching process.

2. The intelligent underground pipeline detection method based on three-dimensional ground penetrating radar three-view images according to claim 1, characterized in that: The underground three-dimensional data of the actual road is collected by three-dimensional ground penetrating radar. After obtaining the underground three-dimensional data, it undergoes a preprocessing process, which includes inverse selective discrete Fourier transform, background removal filtering, progressive low-pass filtering, and gain adjustment. The preprocessing process also includes homogenization adjustment of the three-dimensional radar data. The homogenization adjustment process is to obtain a three-view image set of B-scan longitudinal profile feature image, C-scan horizontal slice feature image, and D-scan transverse profile feature image by multi-view decomposition of the radar data. The image parameters of the three-view image set are depth 3m, span 30m, and vertical span direction 1.32m. The three-view image set is standardized to adjust the image size, optimize the contrast, remove irrelevant patterns, and remove text annotations to form homogenized data, which is used as input image data.

3. The intelligent underground pipeline detection method based on three-dimensional ground penetrating radar three-view images according to claim 2, characterized in that: After preprocessing the underground three-dimensional data, the DCO-YOLO model that matches the underground three-dimensional data is adapted. The adaptation process is to improve the DCO-YOLO model to form an initial model that is adapted to the preprocessed homogenized data. The model is processed according to the spatial position of the underground pipeline corresponding to the actual road. Specifically, based on the direction of the underground pipeline, the dataset is divided into different annotation categories during the initial model training. Among them, the pipeline buried depth is calculated based on the pipeline B-scan longitudinal profile feature image, and the pipeline buried depth is obtained as the distance from the pipeline B-scan longitudinal profile feature image to the ground.

4. The intelligent underground pipeline detection method based on three-dimensional ground penetrating radar three-view images according to claim 2, characterized in that: The input image data is restored before calculating 3D-DIoU. The restoration process is as follows: Based on the positions of the three views in the input image data, a finite three-dimensional space R with a length of L, a width of W, and a height of H is constructed, and the relative positions of the predicted recognition boxes in the corresponding views are kept unchanged. Each recognition box in the three-dimensional space R contains three-dimensional data (x, y, z). When the input image data is the x data of the pipeline feature recognition box 1-B on the B-Scan view, the corresponding conversion formula is: In the above formula, x B1 and x B2 Respectively represent the coordinate values ​​of the left and right edges of the 1-B annotation box converted to space R; X B1 and X B2 They represent the coordinate values ​​of the left and right edges of the 1-B annotation box in the input image respectively; main_view(X,Y,width,height) represents the coordinates and size of the upper left vertex of the B-Scan.

5. The intelligent underground pipeline detection method based on three-dimensional ground penetrating radar three-view images according to claim 4 is characterized by: Each identification frame ensures the corresponding supplementation of one dimension of data when mapped to space R. The supplementation process is: when the identification frame is the pipeline B-scan longitudinal section feature image identification frame, the supplementation process is that the pipeline B-scan longitudinal section feature image identification frame contains the spatial information of the pipeline width x and height z, and lacks the spatial information of the length y, and the corresponding supplementation is the y information of the D-Scan. The corresponding processing and supplementation are analogous, that is, the z information of the B-scan longitudinal section feature image is given to the C-scan horizontal slice feature image, and the x information of the C-scan horizontal slice feature image is supplemented to the D-scan transverse section feature image.

6. The intelligent underground pipeline detection method based on three-dimensional ground penetrating radar three-view images according to claim 5, characterized in that: The process of obtaining the spatial overlap of the annotation boxes in the three views and matching the pipeline feature information based on the 3D distance intersection ratio algorithm is as follows: The three-view matching process is judged by calculating and comparing 3D-DIoU. 3D-DIoU is used to reflect the spatial correlation after the annotation boxes of the three view features of the pipeline feature are mapped to the three-dimensional space R. The calculation formula of 3D-DIoU is: In the above formula, 3D-DIoU is the ratio of the intersection volume and the union volume of two 3D boxes; d is the Euclidean distance between the center points of the two 3D boxes; c is the diagonal length of the minimum circumscribed cube covering the two 3D boxes; The 3D-DIoU value of each annotation frame mapped to the three-dimensional space is calculated respectively, and the 3D-DIoU value of each annotation frame mapped to the three-dimensional space is compared with its corresponding set threshold. When the threshold conditions are met, it indicates that the three identification frames are the B-scan longitudinal section feature image, C-scan horizontal slice feature image, and D-scan transverse section feature image corresponding to the same pipeline, that is, the three view frames are in a successful matching state.

Images