Mobile tunnel detection method and system

Abstract

The invention discloses a mobile tunnel detection method and system, and relates to the field of tunnel detection, and the method comprises the steps: dividing a tunnel into a shield tunnel and a mine tunnel; for a shield tunnel, obtaining a three-dimensional point cloud of the shield tunnel, and extracting bolt holes by adopting a CSF algorithm and a DBSCAN algorithm; adopting a mean-shift algorithm to obtain a central point of the bolt hole, and performing ring-dividing and block-dividing on the three-dimensional point cloud of the shield tunnel according to the central point of the bolt hole to obtain a shield tunnel point cloud after ring-dividing and block-dividing; the method comprises the following steps of: for a mine tunnel, arranging one target every preset distance along the axial direction of the tunnel by using mobile tunnel laser equipment for target arrangement to obtain a mine tunnel point cloud with target position information; generating a mine tunnel lining diagram according to the mine tunnel point cloud with the target position information; establishing an index relationship between the mine tunnel point cloud and the mine tunnel lining graph; and positioning the mine tunnel point cloud according to the index relationship to obtain the positioned mine tunnel point cloud and the like, so that the tunnel defect detection efficiency is improved.

Description

Technical Field

[0001] The present application relates to the field of tunnel detection, and in particular to a mobile tunnel detection method and system. Background Art

[0002] Tunnels are an important part of transportation infrastructure such as railways and highways, and are of great significance to economic and social development. With the rapid development of transportation, tunnel projects are increasing. In order to ensure the safety of tunnel structures and their use functions, operating tunnels need to be regularly inspected and maintained.

[0003] Machine vision is an important intelligent detection method. By analyzing the target image, it can automatically identify defects and anomalies. Introducing machine vision into the field of tunnel detection can significantly improve detection efficiency and objectivity. However, the tunnel environment is complex and changeable, and ordinary two-dimensional imaging is difficult to fully reflect the tunnel structure. 3D laser scanning technology can obtain high-precision 3D point cloud data of the tunnel, providing richer spatial information for tunnel detection.

[0004] However, the original tunnel point cloud data is huge and messy, containing a lot of redundancy and noise, and it is very difficult to analyze directly. Therefore, it is necessary to preprocess the point cloud data, extract key information, and establish the association between the point cloud and the two-dimensional image in order to fully utilize the three-dimensional information for detection. In addition, tunnel defects are of various types and shapes, and it is difficult to design a robust detection algorithm using traditional feature engineering. Summary of the invention

[0005] In response to the problem of low tunnel detection efficiency in the prior art, the present application provides a mobile tunnel detection method and system, which improves the detection efficiency of tunnel defects by dividing tunnels into shield tunnels and mining tunnels, collecting tunnel point clouds for preprocessing, establishing an index relationship between point clouds and lining images, and using machine learning to perform tunnel defect detection.

[0006] The purpose of this application is achieved through the following technical solutions.

[0007] One aspect of the present application provides a mobile tunnel detection method, comprising: according to the tunnel construction method, dividing the tunnel into a shield tunnel and a mine tunnel; for the shield tunnel, obtaining a three-dimensional point cloud of the shield tunnel, and extracting bolt holes by using the CSF algorithm and the DBSCAN algorithm; obtaining the center point of the bolt hole by using the mean-shift algorithm, dividing the three-dimensional point cloud of the shield tunnel into rings and blocks according to the center point of the bolt hole, and obtaining the shield tunnel point cloud after dividing into rings and blocks; for the mine tunnel, using a mobile tunnel laser device with target layout, a target is laid out at a preset distance along the axial direction of the tunnel to obtain a mine tunnel point cloud with target position information; generating a mine tunnel lining map according to the mine tunnel point cloud with target position information; establishing an index relationship between the mine tunnel point cloud and the mine tunnel lining map; locating the mine tunnel point cloud according to the index relationship to obtain the located mine tunnel point cloud; constructing a tunnel detection model based on deep learning, the tunnel detection model includes a backbone network and a neck network; the backbone network adopts a convolutional neural network, and ConvNeXt is set in the backbone network Block network is used to extract tunnel water leakage characteristics. The neck network adopts CBAM attention mechanism to enhance the weight of tunnel water leakage characteristics in channel dimension and spatial dimension. The tunnel detection model is used to detect the shield tunnel point cloud after ring segmentation or the mine tunnel point cloud after positioning to obtain the detection results of the shield tunnel or mine tunnel, which include tunnel leakage, cable looseness and equipment changes.

[0008] Among them, shield tunnel: a tunnel constructed by using a shield machine. The shield machine is a special equipment used for tunnel construction in soft soil strata. It excavates and sets up the segment lining in the soil layer to form a tunnel structure. Mine tunnel: a tunnel constructed by the drilling and blasting method. The drilling and blasting method is a common method for constructing tunnels in rock strata. Through the cycle of drilling, charging, blasting, and slag removal, the tunnel is gradually excavated and concrete is sprayed to form a lining. CSF algorithm: Cloth Simulation Filter, cloth simulation filter algorithm. This algorithm adaptively extracts ground points by simulating the process of virtual cloth laid on a three-dimensional point cloud sagging due to gravity. It is used to segment the lining surface in shield tunnels. Mine tunnel lining map: the inner surface of the mine tunnel is unfolded into a two-dimensional plane image. Through a specific projection transformation method, it is convenient to process and analyze images while maintaining the texture characteristics of the inner wall of the tunnel.

[0009] Among them, the backbone network: the main part of the neural network, used to extract high-dimensional features of the input data. The backbone network generally adopts mature convolutional neural network structures, such as ResNet, VGG, etc., and is appropriately modified according to task requirements. The neck network: the intermediate network connecting the backbone network and the output layer. The neck network plays the role of integrating the features extracted by the backbone network and adjusting the scale and dimension of the feature map to prepare for subsequent classification, detection and other tasks. ConvNeXt Block network: an improved convolutional neural network module. ConvNeXt Block introduces deep separable convolution and inverse residual structure, which reduces the number of model parameters while improving the feature extraction capability, making the network more efficient. CBAM attention mechanism: ConvolutionalBlockAttention Module, convolutional block attention module. CBAM adaptively enhances the feature expression ability of the network by learning feature weights in the spatial dimension and channel dimension. Channel dimension: the depth direction of the feature map. In a convolutional neural network, each feature map corresponds to the output of a certain convolution kernel, reflecting the activation degree of the input in the specific mode defined by the convolution kernel. Spatial dimension: the plane direction of the feature map. The spatial dimension reflects the feature distribution of the image at different positions in the two-dimensional plane, and reflects information such as the spatial structure and texture of the target.

[0010] Furthermore, for a shield tunnel, a three-dimensional point cloud of the shield tunnel is obtained, including: obtaining three-dimensional point cloud data of the shield tunnel; establishing a tunnel section coordinate system with the geometric center of the tunnel as the origin, the vertically upward Z axis as the positive direction, and the X axis perpendicular to the Z axis and pointing to the right, forming the tunnel section coordinate system; calculating the angle θ of the tunnel wall point relative to the section coordinate system; according to the tunnel design parameters and the angle θ, converting the tunnel wall point from the section coordinate system to coordinates in a plane rectangular coordinate system to obtain the unfolded tunnel wall point cloud.

[0011] Furthermore, the angle θ between the tunnel wall point and the cross-section coordinate system is calculated using the following formula:

[0012]

[0013] Where: x i is the horizontal coordinate of the point on the cross section, z i is the vertical coordinate of the point on the cross section, x0 is the horizontal coordinate of the geometric center of the tunnel, and z0 is the vertical coordinate of the geometric center of the tunnel.

[0014] Furthermore, the CSF algorithm and the DBSCAN algorithm are used to extract bolt holes, including: performing coordinate transformation on the unfolded tunnel wall point cloud to obtain a tunnel wall point cloud with the same direction as the initially collected three-dimensional point cloud; constructing a CSF model, initializing the cloth grid, and setting the grid resolution to D times the average point spacing of the unfolded tunnel wall point cloud; the initial cloth position is located within a preset range above the highest point of the point cloud; using the CSF model to obtain the classification threshold hcc of the tunnel wall point cloud, and the classification threshold hcc is used to distinguish between tunnel wall points and bolt hole points; iteratively calculating the height difference between each point in the tunnel wall point cloud and the CSF model cloth particles; marking the points with a height difference greater than the classification threshold hcc as potential bolt hole point clouds, and vice versa; clustering the points marked as potential bolt hole point clouds using the DBSCAN algorithm to obtain the final bolt hole point cloud.

[0015] Among them, cloth mesh (ClothMesh): In the CSF algorithm, the cloth mesh is a virtual triangular mesh plane used to simulate flexible cloth; the cloth mesh is composed of a series of interconnected triangles, and each triangle fixed point is called a particle; the resolution of the cloth mesh is determined by the size of the triangle. The higher the resolution, the denser the mesh and the finer the simulated cloth surface; in this algorithm, the resolution of the cloth mesh is set to D times the average point spacing of the point cloud to balance the computational efficiency and simulation accuracy; the initial position of the cloth mesh is set above the highest point of the point cloud to avoid the cloth passing directly through the point cloud surface.

[0016] Cloth Particle: Cloth particles are the basic units that make up the cloth mesh, corresponding to each triangular vertex in the mesh; each cloth particle has its own specific physical properties, such as mass, speed, force, etc. Cloth particles are connected by virtual springs to simulate the internal tension and shear force of the cloth; during the simulation, cloth particles move under the action of gravity and internal tension, causing the cloth mesh to sag continuously and contact the point cloud surface; by solving the motion equations of cloth particles, the deformation and position of the cloth mesh at each moment can be obtained.

[0017] CSF model (Cloth Simulation Filter Model): The CSF model is a point cloud filtering algorithm based on physical simulation. It separates ground points from non-ground points by simulating the interaction between cloth and point cloud surface. The CSF model consists of three main parts: point cloud data, cloth mesh and physical constraints. The point cloud data is the input original three-dimensional point cloud, which provides the objects to be segmented. The cloth mesh is a virtual flexible plane generated according to the point cloud resolution, which interacts with the point cloud surface through physical simulation. The physical constraints define the material properties (such as stiffness and density) and external environment (such as gravity and damping) of the cloth mesh. By iteratively solving the motion equations of the cloth mesh, it is kept in contact with the point cloud surface under physical constraints, and the filtering and segmentation of the point cloud is achieved through the distance relationship between the cloth particles and the point cloud. The CSF model can adapt to point cloud data with different characteristics by adjusting the cloth properties and classification thresholds, and achieve robust and efficient point cloud processing.

[0018] Furthermore, the mean-shift algorithm is used to obtain the center points of the bolt holes, and the three-dimensional point cloud of the shield tunnel is divided into rings and blocks according to the center points of the bolt holes to obtain the shield tunnel point cloud after the division into rings and blocks, including: according to the collected three-dimensional point cloud of the shield tunnel, using the final bolt hole point cloud, using the mean-shift clustering algorithm to cluster each bolt hole and obtain the center points of the corresponding bolt holes; according to the preset splicing block design data, defining the recognition template of the center of the splicing block bolt hole, matching the obtained center points of the bolt holes with the recognition template, and obtaining the bolt holes of different splicing blocks. The center point of the bolt hole; according to the relative position relationship of the center points of the bolt holes of different splicing blocks, the center point of the bolt holes of each splicing block is determined; the center points of the bolt holes of adjacent splicing blocks are used to locate the longitudinal seams between adjacent splicing blocks through interpolation and linear fitting; the center points of the bolt holes in the same splicing block are used to locate the longitudinal seams between adjacent splicing blocks through linear fitting and translation; the center points of the bolt holes of two adjacent rings are used to locate the annular seams through interpolation and linear fitting; the located seams are used to divide the collected 3D point cloud of the shield tunnel into rings and blocks to obtain the shield tunnel point cloud after ring division and block division.

[0019] Among them, the identification template of the center of the bolt hole of the splicing block: in the construction of the shield tunnel, the tunnel lining is assembled by prefabricated splicing blocks (such as pipe segments). Each splicing block is reserved with regularly arranged bolt holes for connecting adjacent splicing blocks. The identification template of the center of the bolt hole of the splicing block is a virtual geometric model that describes the theoretical position relationship of the center of the bolt hole on a single splicing block. The template is usually created according to the design drawings of the splicing block and contains the relative coordinates of the center point of each bolt hole. By matching the actual extracted center point of the bolt hole with the identification template, the specific position and number of each splicing block can be determined. The size of the identification template and the arrangement of the bolt holes will vary according to different shield tunnel designs and need to be adapted according to the engineering information.

[0020] Interpolation and Line Fitting: Interpolation and Line Fitting are two commonly used mathematical methods in data processing. Interpolation is a method for estimating the value of a function between known data points. Common interpolation methods include linear interpolation and spline interpolation. In this ring and block method, interpolation is used to estimate the joint position between the center points of two adjacent bolt holes. Through linear interpolation, a series of equally spaced points can be generated between the center points of the two bolt holes, which approximately represent the edge of the splicing block. Line fitting is a method for fitting a straight line equation based on known data points. Common fitting methods include least squares method, RANSAC, etc. In this method, line fitting is used to estimate the line connecting the center points of bolt holes on the same splicing block or adjacent rings, which represents the longitudinal edge or circumferential seam of the splicing block. By fitting the straight line equation of the center points of the bolt holes, the positioning error of a single point can be reduced, and a more stable edge or seam position can be obtained.

[0021] Furthermore, different splicing blocks include: capping blocks, adjacent blocks and standard blocks; the capping block is located at the tunnel vault position, and is assembled from multiple splicing blocks. The capping block is provided with multiple bolt holes, and the center points of the bolt holes in the capping block are distributed in a ring shape. The spatial position of the tunnel vault is determined by identifying the center points of the bolt holes in the capping block; the adjacent blocks are located on both sides of the capping block, adjacent to and connected to the capping block, and the adjacent blocks are provided with multiple bolt holes, and the center points of the bolt holes in the adjacent blocks are distributed in the same horizontal plane; the spatial position of both sides of the tunnel is determined by identifying the center points of the bolt holes in the adjacent blocks; the standard blocks are located below the capping blocks and the adjacent blocks, and are arranged longitudinally and circumferentially along the tunnel to form the main body of the tunnel lining structure. The standard blocks are provided with multiple bolt holes, and the center points of the bolt holes in the standard blocks are distributed in the same vertical plane; the longitudinal and circumferential joints of the tunnel lining are determined by identifying the center points of the bolt holes in the standard blocks.

[0022] Furthermore, an index relationship between the mine tunnel point cloud and the mine tunnel lining diagram is established, including: using a scanner to collect two-dimensional cross-sectional point clouds along the design axis direction of the mine tunnel; converting the collected two-dimensional cross-sectional point clouds into three-dimensional point clouds according to the mileage data of the scanner; taking the design axis direction of the mine tunnel as the positive direction of the y-axis, taking the normal direction of the section where the center point of the scanner is located as the positive direction of the z-axis, scanning with the center point as the origin o, and establishing a three-dimensional coordinate system of the tunnel; along the positive direction of the y-axis, numbering the sections perpendicular to the y-axis in the order of collection, and establishing a column index; taking the intersection point P of the negative direction of the z-axis and the section as the starting point, numbering the point cloud in the section in a clockwise direction, and establishing a row index; according to the row index and the column index, writing the three-dimensional coordinates and point cloud information of the corresponding point cloud into the pixel matrix, and establishing an index relationship between the mine tunnel point cloud and the mine tunnel lining diagram.

[0023] Among them, the design axis direction: the design axis refers to the center line of the tunnel in the tunnel design drawing, which represents the theoretical direction of the tunnel. In tunnel construction and inspection, the design axis provides a global reference direction to guide tunnel excavation, lining, measurement and other work. The design axis is usually defined by a series of three-dimensional coordinate points or linear elements (such as straight lines, circular curves, transition curves, etc.). In this indexing method, the design axis direction of the mining tunnel is defined as the positive direction of the y-axis of the tunnel's three-dimensional coordinate system. With the design axis direction as a reference, the tunnel point cloud and lining drawing can be unified into a common coordinate framework, which is convenient for establishing the corresponding relationship between them.

[0024] Mileage data: Mileage data refers to the distance from a certain position in the tunnel to the starting point of the tunnel, usually measured along the design axis of the tunnel. In tunnel construction and inspection, mileage data is used to mark the spatial position of each section or component in the tunnel. It is an important positioning method. Mileage data can be obtained through measuring equipment such as total stations and laser rangefinders, or it can be calculated based on the movement speed and time of the scanner. In this method, the mileage data recorded when the scanner collects two-dimensional cross-sectional point clouds is used to convert the cross-sectional point clouds into three-dimensional point clouds. Through mileage data, the cross-sectional point clouds collected at different positions can be spliced ​​together according to the actual spatial relationship to form a complete three-dimensional point cloud model of the tunnel.

[0025] Point cloud information: Point cloud information refers to the attribute data contained in each point in the point cloud, such as three-dimensional coordinates, color, reflection intensity, etc. In addition to basic geometric information, point clouds can also carry rich semantic information, such as material type, damage status, etc. Point cloud information can be directly collected by a scanner, or extracted or appended from the original point cloud through post-processing methods. In this indexing method, the three-dimensional coordinates and other attribute information of the point cloud are written into the pixel matrix, and a one-to-one correspondence is established with the pixels of the lining map. Through point cloud information, the geometric and semantic features of the tunnel can be mapped from the three-dimensional space to the two-dimensional image space, which is convenient for analysis using image processing and deep learning algorithms.

[0026] Pixel matrix: A pixel matrix refers to a two-dimensional array indexed by row and column numbers, where each element corresponds to a pixel in the image. Each element in the pixel matrix can store attribute information such as color, grayscale, depth, etc. In digital image processing, the pixel matrix is ​​the basic storage and representation form of the image. Most image algorithms operate on the pixel matrix. In this indexing method, the tunnel lining map is represented as a pixel matrix. The row and column indexes of the matrix correspond to the tunnel section and axial position. By mapping the tunnel point cloud to the pixel matrix, a pixel-level correspondence between the point cloud and the lining map can be established, converting the three-dimensional information into two-dimensional information, which is convenient for subsequent image analysis and defect detection.

[0027] Furthermore, the mine tunnel point cloud is positioned according to the index relationship to obtain the positioned mine tunnel point cloud, including: constructing a training set according to the preset layout rule of the targets in the mine tunnel lining map; the preset layout rule includes the equidistant layout of the targets along the tunnel axis and the black and white distance of the target image; training the YOLOv7 model with the training set to obtain a target recognition model; using the target recognition model to obtain the pixel coordinates of each target in the mine tunnel lining map; according to the pixel coordinates of the target, using the index relationship between the mine tunnel point cloud and the mine tunnel lining map, obtaining the three-dimensional coordinates of each target in the mine tunnel point cloud; according to the three-dimensional coordinates of the target, performing coordinate transformation on the mine tunnel point cloud to align the mine tunnel point cloud with the mine tunnel lining map to obtain the positioned mine tunnel point cloud.

[0028] Among them, the black-and-white distance: In this method, the black-and-white distance refers to the width of the black-and-white rectangular stripes in the target pattern; the target is usually composed of several black-and-white rectangular stripes arranged alternately, forming a special visual pattern. The black-and-white rectangular stripes produce obvious grayscale changes and gradient features in the image, which are easy to be detected and recognized by computer vision algorithms. By designing a suitable black-and-white distance, the target can be stably recognized at different scales, and the robustness of the positioning method can be improved. The selection of the black-and-white distance needs to comprehensively consider factors such as the size of the target, the resolution of the camera, and the shooting distance. Generally speaking, the larger the black-and-white distance, the more obvious the visual features of the target and the farther the recognition distance; but too large a distance will also reduce the spatial resolution of the target. In practical applications, the black-and-white distance is usually selected in the order of centimeters to decimeters to balance the recognition effect and positioning accuracy. As a key parameter in the design of target patterns, the black-and-white distance plays an important role in the point cloud positioning method of mine tunnels. By optimizing the black-and-white distance, the recognition effect of the target in complex environments can be improved, and the occurrence of missed detection and false detection can be reduced. At the same time, the design of the target pattern also needs to be adapted to the characteristics of target detection algorithms such as YOLOv7 in order to give full play to the performance advantages of the deep learning model. In this application, the black and white distance of the target and its layout spacing along the axis of the tunnel together constitute the preset layout rules of the target in the mine tunnel lining map. This regularized layout method not only helps to improve the efficiency of target recognition and matching, but also provides reliable control information for subsequent point cloud coordinate transformation and registration. By reasonably designing the black and white distance and layout spacing of the target, high-precision positioning of the mine tunnel point cloud and lining map can be achieved, laying the coordinate foundation for tunnel defect detection and status assessment.

[0029] Furthermore, a tunnel detection model based on deep learning is constructed. The tunnel detection model includes a backbone network and a neck network, including: obtaining the obtained shield tunnel point cloud data after ring and block division or the obtained mine tunnel point cloud data after positioning, and converting the point cloud data into a tunnel lining image; cropping the tunnel lining image into sub-images of size MxM, and obtaining a subset of shield tunnel lining images and a subset of mine tunnel lining images; preprocessing the subset of shield tunnel lining images and the subset of mine tunnel lining images respectively; annotating the preprocessed subset of shield tunnel lining images and the subset of mine tunnel lining images, and annotating tunnel defects; tunnel defects include tunnel water leakage; using the preprocessed and annotated subset of shield tunnel lining images and the subset of mine tunnel lining images as training sets, using a convolutional neural network as the backbone network, and setting a ConvNeXt in the backbone network. Block network extracts tunnel defects; a neck network is set after the backbone network, and the neck network adopts the CBAM attention mechanism; in the training process of the tunnel detection model, GIOU-Loss is used as the positioning loss function; the trained tunnel detection model is used to detect the shield tunnel point cloud after ring and block division or the mine tunnel point cloud after positioning, and the detected tunnel defects are mapped to the point cloud position through the index relationship between the point cloud and the tunnel lining image, so as to obtain the three-dimensional spatial position of the tunnel defect as the tunnel detection result output.

[0030] Another aspect of the present application also provides a mobile tunnel detection system for executing a mobile tunnel detection method of the present application.

[0031] Compared with the prior art, the advantages of this application are:

[0032] Adopting targeted point cloud preprocessing strategies for tunnels with different construction methods can significantly improve the accuracy and efficiency of tunnel defect detection. For shield tunnels, bolt holes are extracted by CSF algorithm and DBSCAN algorithm, and the center point of bolt holes is obtained by mean-shift algorithm, so as to realize accurate ring and block division of point cloud. Bolt holes are important features in shield tunnel structure. By extracting this feature and dividing it into blocks, the structural characteristics of the tunnel can be fully utilized to make subsequent detection more targeted. At the same time, CSF algorithm can effectively separate ground points and non-ground points, DBSCAN algorithm can realize the clustering of bolt holes, and mean-shift algorithm can obtain the precise center of bolt holes. The combination of the three can realize high-precision extraction and positioning of bolt holes. For mine tunnels, point clouds are obtained by mobile laser equipment with target layout, and the index relationship between point clouds and lining diagrams is established to realize accurate positioning of point clouds. Due to the low construction accuracy of mine tunnels, the lining structure is not regular enough, and there is a lack of obvious features such as bolt holes in shield tunnels. Introducing external targets and establishing indexes can make up for the shortcomings of mine tunnels themselves and provide a reliable positioning basis for subsequent detection.

[0033] The use of advanced deep learning detection models, the introduction of attention mechanisms and innovative network structures can extract more refined and comprehensive tunnel defect features, thereby improving detection accuracy. The backbone network uses convolutional neural networks and sets ConvNeXt Block, where convolutional neural networks have been widely verified in image feature extraction. ConvNeXtBlock can model feature dependencies at a finer granularity through point-by-point convolution, Layer Normalization and other strategies, which helps to extract local and detail features. The neck network adopts the CBAM attention mechanism, which explicitly models the importance of features in both channel and space dimensions, so that the network focuses on key features related to tunnel defects, suppresses redundant and noisy information, and thus enhances detection accuracy. At the same time, the CBAM module has a small amount of calculation, is easy to embed, and does not significantly increase the complexity of the model. In addition, GIOU-Loss is used as the positioning loss function during training. Compared with the traditional IOU-Loss, it imposes a greater penalty on prediction boxes with smaller overlapping areas, thereby guiding the model to learn more accurate positioning.

[0034] By establishing an index relationship between the tunnel point cloud and the lining image, the detection results can be accurately matched to the three-dimensional spatial position, providing a reliable basis for the location and repair of tunnel defects. The lining image contains rich two-dimensional structural information, which is convenient for defect detection using mature image processing and analysis algorithms. However, tunnel engineering requires the spatial location of defects to be clear. Through the index relationship, it is easy to map and convert between two-dimensional detection and three-dimensional positioning, taking into account the needs of both detection and positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present application will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:

[0036] Figure 1 is a schematic diagram of an exemplary application scenario of a mobile tunnel detection method according to some embodiments of the present application;

[0037] Figure 2 is a schematic diagram of a cross-sectional coordinate system according to the present application;

[0038] Figure 3 is a schematic diagram of the tunnel wall deployment effect according to the present application;

[0039] Figure 4 is a schematic diagram of capping block bolt identification according to the present application;

[0040] Figure 5is a schematic diagram of the bolt hole center classification results according to the present application;

[0041] Figure 6 is a schematic diagram of straight line fitting of a capping block and an adjacent block according to the present application;

[0042] Figure 7 It is a rendering of the longitudinal joint positioning between the capping block and the adjacent block according to the present application;

[0043] Figure 8 is a schematic diagram of the circumferential seam fitting effect according to the present application;

[0044] Fig. 9 It is a rendering of the circumferential seam positioning according to the present application;

[0045] Fig.10 It is a diagram showing the effect of seam positioning in a plane state according to the present application;

[0046] Fig.11 It is a rendering of the shield tunnel joint positioning according to the present application;

[0047] Fig.12 Schematic diagram of the three-dimensional coordinate system of the tunnel according to the present application. DETAILED DESCRIPTION

[0048] The method and system provided in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0049] like Figure 1 As shown, this embodiment provides a hardware inspection vehicle for collecting point cloud data in a tunnel. The inspection vehicle adopts a segmented design and consists of four parts: a left wheel assembly module, a right wheel assembly module, a scanner module and a central control module. The weight of each module does not exceed 20kg, and the total weight does not exceed 40kg, which is convenient for manual handling and assembly. The left wheel assembly module and the right wheel assembly module are symmetrically arranged, respectively located on the left and right sides of the inspection vehicle. Each wheel assembly module includes a driving wheel and a driven wheel, and the distance between the two wheels is measured in real time by a gauge sensor to adapt to tunnel tracks with different gauges. A conductive rod is also provided on the wheel assembly, which can be lowered to contact the track or lifted up to be insulated from the track as needed. The left and right conductive rods are connected by wires to detect the conduction of the track. In addition, the wheel assembly module is also provided with a foldable carrying handle for easy manual handling.

[0050] The scanner module is located in the middle of the inspection vehicle, including the scanner body and the support base. The scanner uses a commercial 3D laser scanner, and the model can be selected according to the scanning accuracy and efficiency requirements. The support base is fixed on the top of the central control module, and a threaded hole is opened on it to be threadedly connected to the central control module. The scanner can be firmly installed on the inspection vehicle by tightening the bolts. The central control module is the core of the inspection vehicle, integrating multiple functions such as control, power supply, mileage measurement, and posture detection. Its external structure includes a removable battery, lighting lamps, industrial computer, control buttons, and carrying handles. The removable battery is easy to charge and replace, and is connected to other electrical components through an internal cable plug to ensure the reliability of power supply. The control button is used for on-site manual control of the start and stop and mode switching of the inspection vehicle. The on-board industrial computer can realize the automatic control of the inspection vehicle as well as data collection, storage, and upload. The built-in part includes a drive motor, a mileage encoder, an inclination sensor, and a motor controller. The drive motor is connected to the active wheel to drive the inspection vehicle forward or backward according to the control command. The mileage encoder records the driving distance in real time, synchronizes with the scanning data, and provides position information for 3D reconstruction. The inclination sensor is used to detect the pitch and roll angles of the vehicle body, which can be used to determine the flatness of the track. The left wheel module, the right wheel module and the central control module are connected by quick locks. When connecting, just align the quick lock connector with the connection seat and tighten the knob to quickly lock. When disassembling, rotate the knob in the opposite direction to separate. The operation is simple. There are carrying handles on the front and back of the central control module. The handles can be unfolded when in use and folded when not in use for easy transportation and storage.

[0051] According to the construction method of the tunnel, the tunnel is divided into shield tunnel and mine tunnel; there are obvious differences in the structural characteristics and surface morphology of these two types of tunnels: shield tunnels are excavated by shield machines, and the lining is usually assembled from prefabricated segments. Bolt connections are set between the segments, so there are a large number of bolt holes distributed on the wall of the shield tunnel. These bolt holes are of regular size and orderly arrangement, which is a typical feature of shield tunnels. Mine tunnels are mostly excavated by drilling and blasting, and the lining is generally made of cast-in-place concrete, with a rough surface and lack of obvious structural features. And due to construction process limitations, the lining shape and dimensional accuracy of mine tunnels are often not as good as those of shield tunnels. In view of the differences between the two types of tunnels, this application intends to design a point cloud preprocessing strategy in a targeted manner. For shield tunnels, the bolt hole features are extracted, and the point cloud is segmented accordingly; for mine tunnels, the corresponding relationship between the point cloud and the lining image is established by manually laying targets. This differentiated processing idea can make full use of the structural characteristics of different tunnels and improve the pertinence and effectiveness of preprocessing. In addition, due to the low construction accuracy of mine tunnels and irregular cross-sectional shapes, direct 3D reconstruction is difficult. However, the specifications of shield tunnel segments are uniform and the cross-sectional shapes are basically the same, so 3D reconstruction is relatively easy to achieve. Therefore, introducing the intermediate expression of lining unfolded images for mine tunnels can avoid shape errors in 3D reconstruction to a certain extent, which is convenient for subsequent defect analysis.

[0052] For shield tunnels, first use mobile laser scanning equipment to obtain the three-dimensional point cloud data of the tunnel. Since the overall structure of the shield tunnel is cylindrical, in order to facilitate subsequent processing, the point cloud needs to be projected and flattened. The coordinate system is established with the ellipse fitting center of the tunnel section as the origin. The vertical upward direction is the positive direction of the Z axis, the X axis is perpendicular to the Z axis and points to the right, and the Y axis is determined according to the right-hand rule, such as Figure 2 As shown. Taking each section as a unit, the angle θ between the section points on the left and right sides of the z-axis and the z-axis is calculated:

[0053] Combined with the tunnel segment design radius R, according to the arc length formula L = R × θ, where: i is the horizontal coordinate of the point on the cross section, z i is the vertical coordinate of the point on the cross section, x0 is the horizontal coordinate of the geometric center of the tunnel, and z0 is the vertical coordinate of the geometric center of the tunnel. Calculate the arc length L of each point in the positive direction of the z axis, and record it as the Y coordinate value of each point after projection; according to the design radius R of the shield segment, use the arc length formula to calculate P i The coordinates in the expanded image (u i ,v i ):u i =R×θ i , v i =y i , that is, point P i Rotation angle θ around the Z axis i And translate y along the Y axis i The position of the point in the expanded image can be obtained. To facilitate the recovery of the 3D information of the point cloud, the original 3D coordinates are saved while saving the expanded image coordinates. Therefore, each point contains 7 fields: (x i ,y i ,z i ,u i ,v i ,θ i ,R), Figure 3 The effect of tunnel unfolding is shown. Through the above flattening process, the shield tunnel point cloud is converted into a two-dimensional image, which is similar to the relationship between the ground point cloud and the non-ground point cloud in the airborne laser radar scene. This creates conditions for further extracting tunnel features.

[0054] After the flattening of the tunnel point cloud is completed, the bolt hole features need to be further extracted. Since the flattening process changes the coordinate system of the point cloud, in order to facilitate subsequent processing and visualization, the unfolded point cloud needs to be restored to the coordinate system of the initial acquisition. i The coordinates of (u i ,v i ), the corresponding rotation angle is θ i, the radius is R, then its three-dimensional coordinates (x i ',y i ',z i ') is: x i '=R×sin(θ i );y i '=v i ; z i '=R×cos(θ i ); after coordinate transformation, the direction of the expanded tunnel wall point cloud is consistent with that of the original point cloud.

[0055] The CSF (Cloth Simulation Filter) algorithm extracts ground point clouds by simulating a virtual cloth drooping and gradually fitting the ground. min ,x max ,y min ,y max ) and the average point spacing d avg , initialize a rectangular grid cloth model. The resolution of the cloth grid is r c Set to D times the average point spacing: r c =D×d avg , where D is the cloth resolution factor, usually ranging from 2 to 5. The larger the D value, the sparser the cloth mesh; the smaller the D value, the denser the cloth mesh. The value of D needs to balance the computational efficiency and ground fitting accuracy. In the xoy plane, the node coordinates (x c ,y c )for:

[0056]

[0057] Initially, the z coordinate of the cloth is c Set to a preset distance h0:z above the highest point of the point cloud c =max(z i )+h0; where z i is the point cloud P i The z-coordinate component of h0 is the initial height of the cloth, which should be greater than the maximum possible depth of the bolt hole. This setting is to ensure that the cloth can fully cover and droop to all bolt hole areas.

[0058] The constructed CSF model is used to preliminarily classify the tunnel wall point cloud to obtain tunnel wall points (ground points) and potential bolt hole points (non-ground points). First, iteratively calculate each point P i Height difference Δh from the cloth particles i . Let P i The coordinates of (x i ,yi ,z i ), the coordinates of the cloth particle closest to its xoy projection are (x c ,y c ,z c ), then the height difference is: Δh i =z i -z c , then set a height threshold h t If Δh i >h t , then P i Mark as potential bolt hole points, otherwise mark as tunnel wall points. t There are two common methods for determining the value of h: statistical estimation based on training data. In the labeled training data, the height difference between the bolt hole point and the surrounding tunnel wall is counted, and the mean or median is taken as h. t This method requires manual annotation of a certain amount of data, but it can better adapt to the characteristics of different tunnels. Estimation based on prior knowledge. Since the design depth of the bolt hole is usually known, it can be directly used as h t Taking into account construction errors and wear and tear, h can be appropriately relaxed. t The value range of is, for example, 1.5 times the design depth. This method is simple and intuitive, but the generalization may be poor. The CSF model achieves preliminary classification of the tunnel wall point cloud by constructing a grid-like cloth and solving the height difference between each point and the cloth. The classification threshold h t The selection of can be done through data-driven statistical estimation or by utilizing prior knowledge of bolt hole design parameters. This process can effectively distinguish the tunnel lining surface and bolt hole area, laying the foundation for further bolt hole clustering and extraction.

[0059] After completing the CSF model classification, the potential bolt hole point cloud is obtained. In order to further extract the precise location of each bolt hole, these point clouds need to be clustered. DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density-based clustering algorithm that can classify potential bolt hole point clouds into different bolt holes. DBSCAN has two key parameters: Neighborhood radius ε (eps): determines the maximum connection distance between points. If the distance between two points is less than or equal to ε, they are considered density-reachable. Minimum number of included points minPts: determines the minimum number of points in a cluster. If the number of points in the ε-neighborhood of a point is greater than or equal to minPts, the point is a core point. The value of ε needs to consider the diameter of the bolt hole and the sampling spacing of the point cloud. Generally, ε can be set to 1 / 4 to 1 / 2 of the bolt hole diameter. This ensures that the points in the same bolt hole can be connected, while the points between different bolt holes are not connected. The value of minPts needs to consider the point cloud density of the bolt hole area. minPts can be estimated based on the average number of points in each bolt hole. For example, if each bolt hole has an average of 100 points, minPts = 50, that is, a cluster must contain at least 50% of the average number of points. Using the determined ε and minPts, DBSCAN clustering is performed on the potential bolt hole point cloud to obtain a series of point cloud clusters {C1, C2, ..., C n}. Each cluster C i Corresponds to a bolt hole. For each point cloud cluster C i , if the number of points it contains is C i ≥minPts, the cluster is retained, otherwise it is considered as noise and removed. This can filter out some pseudo bolt holes formed by scattered points or small areas. For each retained bolt hole point cloud cluster C i , use the mean-shift algorithm to extract its centroid as the precise location of the bolt hole. Use the mean-shift algorithm to cluster each bolt hole point cloud, and calculate the centroid after clustering as the center point of the bolt hole. The bandwidth parameter Bandwidth of Mean-shift is set to 0.1m, that is, the points in the spherical neighborhood with each point as the center and a radius of 0.1m will gather to the highest density and eventually converge to the local maximum point, which is the center position of the bolt hole.

[0060] Capping block bolt hole identification: Figure 4As shown in the figure, there are 6 bolt holes in the capping block, and their center points are approximately distributed on a circle with a radius of R, and the center of the circle is the geometric center C(x0,y0) of the capping block. Based on this prior knowledge, a recognition template for the bolt holes of the capping block can be defined: draw a circle with C(x0,y0) as the center and R as the radius, and select the center points of all bolt holes that fall within the circular area, which is recorded as the set S = {P1,P2,......,P n}, where n≥6. For each point P in the set S i (x i ,y i ) to determine whether it is located in the given 5 rectangular areas. The positions and sizes of these 5 rectangles are determined according to the design drawings of the capping block, and correspond to the theoretical positions of the 5 bolt holes except the center of the circle. If there are exactly 5 points in the set S that are located in the specified rectangular area, the 6 bolt holes of the capping block are identified, including the center of the circle C and these 5 points. Through the above template matching method, the bolt hole positions of the capping block can be accurately located and marked as specific categories, such as red.

[0061] Identification of bolt holes in adjacent blocks and standard blocks: After identifying the bolt holes in the capping block, the positional relationship of the adjacent blocks and standard blocks relative to the capping block can be used to infer the ownership of the remaining bolt holes. According to the design parameters, calculate the theoretical position of the bolt holes in the adjacent blocks and standard blocks relative to the bolt holes in the capping block. Usually, the center points of the bolt holes in the left and right adjacent blocks are located on both sides of the line connecting the center points of the bolt holes in the capping block, and the distance is the width of the adjacent block; the center point of the bolt hole in the standard block is located below the center point of the bolt hole in the adjacent block, and the distance is the block height. Traverse all the remaining bolt hole center points, and for each point P (x, y): calculate the horizontal distance d from P to the center points of the 6 bolt holes in the capping block. h and vertical distance d v If d h Within the error range of the adjacent block width, and d v If it is close to 0, P is classified as a bolt hole of an adjacent block and is given a blue color.

[0062] Furthermore, it can be subdivided into left adjacent blocks and right adjacent blocks according to whether P is on the left or right side of the center line of the capping block. v Within the error range of the standard block height, and d h If the horizontal distance is close to the bolt hole of the adjacent block, P is classified as a bolt hole of the standard block and given a green color. For the center points of the bolt holes that failed to match successfully, they can be marked as unknown categories, or interpolated and estimated based on the categories of their neighboring points. Through the above methods, the bolt holes belonging to the capping block, adjacent block and standard block can be identified in turn and distinguished by different colors. Figure 5The final classification results are shown. This process makes full use of the relative position relationship between different blocks, reduces the search range during identification, and improves the efficiency and accuracy of the algorithm. At the same time, a special recognition template is designed for the bolt holes of the capping block, which can effectively deal with the special structure of the tunnel vault and enhance the robustness of identification.

[0063] The longitudinal joint is a joint along the longitudinal direction of the tunnel (the direction of tunnel advancement) used to connect adjacent ring blocks. In each ring block, the longitudinal joint separates the capping block from the adjacent block, and the adjacent block from the standard block. Figure 6 , select two bolt hole center points on the left and right sides of the capping block, namely A1, A2 and B1, B2, which are located at the junction of the capping block and the left (right) adjacent block. Interpolate A1, A2 and B1, B2 respectively to obtain points A and B, then the straight line where ray AB is located is the left longitudinal seam (straight line 1). Similarly, select the bolt hole center points C1, C2 and D1, D2 on the right side of the capping block, interpolate to obtain points C and D, then the straight line where ray CD is located is the right longitudinal seam (straight line 2). Straight lines 1 and 2 are symmetrical about the center line of the tunnel, and they separate the capping block from the left and right adjacent blocks. In order to determine the longitudinal dividing line between the adjacent block and the standard block, take the two bolt hole center points E and F near the edge of the capping block, and fit the straight line EF (straight line 3), which is approximately parallel to the center line of the tunnel. According to the design drawings, the width of the adjacent block is 3.7m. Therefore, the straight line 3 is translated 3.7m to the left and right respectively to obtain two new straight lines, which are the boundary lines between the adjacent blocks and the standard blocks. Through the above steps, the longitudinal block division inside the tunnel ring block is completed. Straight lines 1 and 2 are the boundary lines of the capping block, and the translated straight line 3 is the boundary line of the adjacent block. A ring block is divided into 5 longitudinal sub-blocks. The longitudinal joint positioning effect is shown as follows: Figure 7 .

[0064] The circumferential joint is a joint along the tunnel circumference, used to connect the adjacent ring blocks. Since the standard blocks of adjacent ring blocks are distributed on the same circumference, the circumferential joint can be obtained by fitting the center points of the bolt holes on the standard blocks. Figure 8 As shown in the figure, on the standard blocks of the two adjacent ring blocks, multiple pairs of bolt hole center points are selected along the circumferential direction, such as a1-a2, b1-b2, c1-c2, d1-d2, e1-e2, f1-f2, g1-g2, h1-h2, etc. The midpoints of each pair of points (a1, a2), (b1, b2), (c1, c2), (d1, d2), (e1, e2), (f1, f2), (g1, g2), (h1, h2) are respectively taken to obtain

[0065] points, they fall roughly on a circle.

[0066] Use the least squares method to fit points a, b, c, d, e, f, g, h to get a plane ring, the plane where it is located is the annular joint surface. Specifically, given point a(x a ,y a ,z a )、b(x b ,y b ,z b )、......、h(x h ,y h ,z h ) are the coordinates of a total of 8 points. We want to fit a plane ring to make it as close to these points as possible. The plane ring can be described by the coordinates of the center (x0, y0, z0), the normal vector (A, B, C) and the radius R. Among them, (A, B, C) is the unit normal vector of the ring plane, satisfying A 2 +B 2 +C 2 = 1. The goal of least squares fitting is to solve a set of parameters (x0, y0, z0, A, B, C, R) so that the sum of the squares of the distances from these 8 points to the fitting circle is minimized. We define point P i (x i ,y i ,z i ) to the ring i for:

[0067]

[0068] ; The first term represents point P i The square of the difference between the distance to the center of the circle and the radius R. The second term represents point P i The square of the distance to the annulus plane. Construct the least squares objective function: Constraints: A 2 +B 2 +C 2 =1; This is a nonlinear optimization problem with equality constraints, which can be solved by Lagrange multiplier method or sequential convex optimization. To simplify the calculation, we can solve it in two steps: Step 1: Initialize the plane normal vector (A, B, C); calculate the center coordinates of the 8 points (x c ,y c ,z c ), as the initial estimate of the center of the circle. Calculate the radial unit vectors from the 8 points to the center of the circle Calculate v i The mean vector of And normalize it as the initial value of the normal vector (A, B, C). This direction is approximately perpendicular to the plane of the ring. Step 2: Optimize the center coordinates, radius and normal vector; fix the normal vector (A, B, C), optimize the center coordinates (x0, y0, z0) and radius R: This is an unconstrained nonlinear least squares problem that can be solved using the Gauss-Newton method or the Levenberg-Marquardt algorithm. At each iteration, substitute the current solution (x0, y0, z0, R) into f1, calculate its first-order derivative (Jacobian matrix) and second-order derivative (Hessian matrix), and update the solution using Newton descent until convergence. Fix the center (x0, y0, z0) and radius R, and optimize the normal vector Constraints:

[0069] A 2 +B 2 +C 2 =1. This is an optimization problem of a quadratic function on the unit sphere, which can be solved using the Lagrange multiplier method. Let L(A, B, C, lambda) be the Lagrange function:

[0070] L(A,B,C,λ)=f2+λ(A 2 +B 2 +C 2 -1); find the partial derivative of (A, B, C) and set it equal to 0, and then combine the constraints to get the optimal solution (A*, B*, C*). Alternately execute the two sub-steps of step 2 until the change of (x0, y0, z0, A, B, C, R) is less than the given threshold. The plane ring equation obtained at this time is:

[0071] A(x-x0)+B(y-y0)+C(z-z0)=0;(x-x0) 2 +(y-y0) 2 +(z-z0) 2 =R 2 ; Substitute the plane equation into the coordinates (x, y, z) of the point cloud to obtain the annular joint surface. By cutting the point cloud along the plane, the two adjacent rings can be separated. Use the obtained plane equation to cut the tunnel point cloud, separate the two adjacent rings, and complete the annular segmentation. Use the located longitudinal seams (straight lines 1, 2, and translation line 3) and the annular joint surface to cut the entire tunnel point cloud to obtain a series of annular blocks. Each annular block is divided into 5 sub-blocks (1 capping block, 2 adjacent blocks, and several standard blocks) by the longitudinal seams. Render different types of sub-block point clouds with different colors to obtain the following: Fig.10 , 11This ring-segmentation method makes full use of the structural characteristics of the tunnel assembly blocks and the distribution law of bolt holes, and fits the longitudinal and annular joints through the sparse center points of the bolt holes, and then cuts the entire tunnel point cloud.

[0072] The mobile tunnel inspection system is usually a mobile vehicle equipped with data acquisition instruments. It adopts an integrated design, powered by the internal battery of the vehicle, and controlled by the remote control or the body button. The bracket base carrying the instrument (such as laser scanner, camera, etc.) is installed in the middle of the vehicle and fixed by bolts. If necessary, lighting can be installed in front and behind the vehicle to improve the lighting conditions inside the tunnel. For target layout, select a flat target with appropriate size and good reflective performance, such as a black and white checkerboard target or a dot target. A target is laid out at a preset distance (such as 5m) along the axis of the tunnel, and the center of the target should be aligned with the center line of the tunnel as much as possible. Record the number and mileage position information of each target as the basis for subsequent point cloud splicing. Mobile scanning uses a three-dimensional laser scanner (such as Leica P50, FAROFocus, etc.) mounted on the inspection vehicle to continuously collect point cloud data along the axis of the tunnel. The scanning frequency of the scanner should be set according to the travel speed of the inspection vehicle and the required point cloud density, and generally more than 1 million points / second can be selected. The ranging range of the scanner should cover the entire tunnel section, usually 30-200 meters. The scanning field of view of the scanner should be as large as possible, with a horizontal field of view of not less than 360° and a vertical field of view of not less than 150°.

[0073] Point cloud stitching, based on the correspondence between the targets at different scanning sites, calculate the coordinate transformation matrix between the point clouds of adjacent sites. Commonly used target registration algorithms include the least squares method and the ICP algorithm, which solve the transformation matrix by minimizing the coordinate residual of the target center at two sites. Using the calculated transformation matrix, the point cloud data of all sites are unified into a global coordinate system to obtain a complete tunnel point cloud model. The spliced ​​point cloud is pre-processed by denoising and downsampling to improve data quality and processing efficiency.

[0074] In this embodiment, the ICP algorithm is used for point cloud registration. ICP (Iterative Closest Point) is a commonly used point cloud registration algorithm. Through iterative optimization, the coordinate transformation matrix between two point clouds is estimated, thereby realizing point cloud splicing. The point cloud data of two adjacent stations are recorded as P and Q, respectively, and the corresponding target center coordinates in the two stations are recorded as {p i} and {q i}, i = 1, 2, ..., n. Initialization: Use P point cloud as reference point cloud, Q point cloud as source point cloud, and set the upper limit of the number of iterations max iter , such as max iter= 100, set the convergence threshold ε, such as ε = 1e-6, and initialize the coordinate transformation matrix T as the unit matrix. Iterative optimization: For the kth iteration, k = 1, 2, ..., max iter :For each target center q in Q i , find the nearest neighbor target center p in P i , forming corresponding point pairs (p i ,q i ), use SVD (Singular Value Decomposition) or quaternion method to estimate the optimal coordinate transformation matrix T k , so that Minimize, use T k Transform the Q point cloud to the coordinate system of the P point cloud to obtain the transformed Q k , calculate Q k The root mean square error between If RMSE k <εork>max iter , then stop the iteration and output the current transformation matrix T k ; Otherwise, let Q = Q k , continue to the next iteration. Point cloud stitching: Use the estimated optimal transformation matrix T k , transform the Q point cloud to the coordinate system of the P point cloud, splice the transformed Q point cloud with the P point cloud to obtain the locally spliced ​​point cloud, repeat the splicing of the point clouds of all adjacent stations in sequence, and finally obtain a complete tunnel point cloud model.

[0075] The lining diagram of a mine tunnel is generated by taking the design axis of the tunnel as the positive direction of the y-axis and any point on the y-axis (such as the starting point) as the origin to establish a tunnel coordinate system; the spliced ​​tunnel point cloud is converted from the global coordinate system to the tunnel coordinate system, keeping the relative position relationship unchanged; along the tunnel axis direction (y-axis), a number of sections are divided according to the equal spacing Δy (such as 0.1m), and each section is perpendicular to the y-axis; for the i-th section (i=1, 2, ..., n), the y coordinate is extracted in [y i -Δy / 2,y i+Δy / 2] to form the i-th section point cloud. The point cloud in each section is expanded according to the annular angle θ, and the three-dimensional point cloud (x, y, z) is converted into a two-dimensional plane image (θ, y); the expanded image takes the y axis as the vertical direction and the annular angle θ as the horizontal direction to form a rectangular lining expansion diagram; the range of the annular angle θ is [0, 2π), and it can be sampled at equal angle intervals Δθ (such as 0.1°). For each expanded pixel point (θ, y), according to its corresponding point cloud distance d or reflection intensity value I, it is mapped to a grayscale value g; when mapping the distance, the distance d from the point cloud to the scanner can be normalized to the range of [0, 255]. The closer the distance, the larger the grayscale value, such as: g = 255*(1-d / d_max); when mapping the reflection intensity, the reflection intensity value I of the point cloud (usually 0-255) can be directly used as the grayscale value, such as: g = I. After mapping, the lining grayscale image G of the i-th section is obtained. i For the grayscale image G i Perform histogram equalization and contrast enhancement to improve the display effect of lining defects. Histogram equalization can stretch the dynamic range of grayscale values ​​and enhance the contrast of the image. Contrast enhancement can highlight the difference between lining defects and background and improve the recognition of defects. i (i=1,2,...,n) are spliced ​​in sequence along the y-axis direction to form a complete tunnel lining development diagram The spliced ​​lining map G has the y-axis (tunnel axial direction) as the vertical direction and the circumferential angle θ as the horizontal direction. The resolution of the lining map G depends on the axial sampling interval Δy and the circumferential sampling interval Δθ, which can usually reach the millimeter level.

[0076] Establishing the index relationship between the point cloud and the lining diagram is to achieve mutual positioning and mapping between the two, and define the tunnel coordinate system: the design axis direction of the mine tunnel is the positive direction of the y-axis, indicating the mileage direction of the tunnel, and the normal direction of the section where the center point of the scanner is located is the positive direction of the z-axis, indicating the vertical direction of the tunnel. The x-axis direction is determined according to the right-hand coordinate system and is orthogonal to the y-axis and z-axis. The origin O can be selected at the starting point of the tunnel, the initial position of the scanner, etc., to ensure the uniqueness and consistency of the coordinate system. Generate a section sequence: along the axis direction of the tunnel (y-axis), divide several sections at equal intervals Δy (such as 0.1m), each section is perpendicular to the y-axis, and these sections are numbered according to the acquisition order (such as from the starting point to the end point of the tunnel), and the number i increases from 0. The y coordinate y of the i-th section is y. i It can be expressed as: i=i×Δy, section number i is used as the column index of the lining development diagram, indicating the position of the image in the y-axis direction. Establish the circumferential angle index: In each section, the point cloud is expanded into a two-dimensional image according to the circumferential angle θ. The value range of the circumferential angle θ is [0, 2π). It can be sampled at equal angle intervals Δθ (such as 0.1°). The sampled circumferential angle θ j (j=0, 1, ..., m-1), the number j increases from 0, and the jth circumferential angle θ j It can be expressed as: j =j×Δθ, the circumferential angle number j is used as the row index of the lining development diagram, indicating the position of the image in the circumferential direction.

[0077] Construct point cloud index: For each point P in the point cloud k (k=0,1,...,n-1), calculate its coordinates (x k ,y k ,z k ), according to point P k The y coordinate of k , determine the section number i k : According to point P k The circumferential angle θ in the section k , determine the corresponding circumferential angle number Point P k The index is recorded as (i k ,j k ), indicating its position in the lining development diagram. Generate lining development diagram: According to the number of sections n and the number of circumferential angle sampling m, create an empty lining development diagram G with a size of m rows and n columns. For each point cloud point P k , according to its index (i k ,j k ), and its corresponding gray value g k Assign the corresponding pixel G[j k ,i k ], gray value g k According to point P k The distance d k Or reflection intensity I k After all the point cloud points are mapped, the complete lining development diagram G is obtained.

[0078] For each section, select the point P that intersects the negative direction of the z-axis as the starting point, and the horizontal coordinate x of P is p and the ordinate y pIt can be obtained by interpolation. Taking P as the starting point, the point cloud in the section is sampled at equal angles (such as 0.1°) in the clockwise direction to obtain a series of sampling points. These sampling points are numbered in clockwise order, and the numbers start from 0 and increase. The sampling point number is used as the row index of the lining development map, indicating the position of the image in the circumferential angle. According to the row and column index (i, j) of the point cloud, its pixel coordinates (x, y) in the lining development map are calculated: x = j, y = (i * angle resolution) / (2 * pi) * image height; the three-dimensional coordinates (X, Y, Z) and attribute information (such as reflection intensity I) of the point cloud are written into the corresponding position (x, y) of the lining map pixel matrix. For distance mapping, the distance d from the point cloud to the scanner can be normalized to the range of 0-255 as the pixel value: pixel value = (d-d_min) / (d_max-d_min) * 255; for reflection intensity mapping, the reflection intensity value I of the point cloud can be directly used as the pixel value.

[0079] The mine tunnel point cloud is located according to the index relationship, and the pixel coordinates (x, y) on the lining map are mapped to the three-dimensional coordinates (X, Y, Z) of the point cloud using the established index relationship between the point cloud and the lining map. For each pixel point P (x, y) on the lining map, the corresponding point cloud coordinates Q (X, Y, Z) are found through the index relationship: QX = f (Px, Py); QY = g (Px, Py); QZ = h (Px, Py); where f, g, h are index mapping functions, which can be constructed by interpolation, fitting and other methods. Repeat the above process for all pixel points to obtain the ordered point cloud data corresponding to the lining map, that is, the located mine tunnel point cloud.

[0080] Construct a target recognition training set, and generate the theoretical position and size of the target on the lining map according to the preset target layout rules. The targets are arranged at equal intervals along the tunnel axis, with a spacing of d (such as 5m). The ordinate of the i-th target on the lining map is: y_i = round (i*d / image resolution); the target pattern uses a black and white chessboard, and the size of the black and white grid is sxs, so the theoretical width and height of the target on the lining map are: w = h = round (s / image resolution); with the target theoretical position as the center, intercept the image block of sizexsize (such as 64x64) as the positive sample, and mark the pixel coordinates of the target. Randomly intercept several sizexsize image blocks in other areas of the lining map as negative samples, and do not mark the target coordinates. The positive and negative sample images and their annotation information are organized into a data set for training the target recognition model.

[0081] YOLOv7 is selected as the target recognition model. It is an end-to-end real-time target detection algorithm with high accuracy and high speed. The constructed training set is randomly divided into training set, validation set and test set, and the ratio can be set to 8:1:1. Set the hyperparameters of YOLOv7, such as learning rate, batch size, number of iterations, etc., to train the model. Evaluate the model performance on the validation set, and tune the hyperparameters according to indicators such as precision and recall until the optimal model is obtained. Evaluate the optimal model on the test set to ensure that its generalization performance meets the requirements. Input the complete mine tunnel lining map into the trained YOLOv7 model to obtain the target detection results. The detection results include information such as the pixel coordinates (x, y) of the target, the confidence score and the bounding box size (w, h). According to the confidence threshold (such as 0.5), filter out reliable target detection results and obtain their pixel coordinates (x, y). Using the index relationship between the point cloud and the lining map, the pixel coordinates (x, y) of the target are mapped to the three-dimensional coordinates (X, Y, Z) in the point cloud. Repeat the above process for each detected target to obtain the three-dimensional coordinate set of the target in the point cloud {(X i ,Y i ,Z i )}.

[0082] According to the three-dimensional coordinate set of the target in the point cloud {(X i ,Y i ,Z i )}, calculate the average coordinates of the target center (X c ,Y c ,Z c ). Take the target center coordinates as the control point and calculate the transformation matrix from the point cloud coordinate system to the lining map coordinate system Where (X c ',Y c ',Z c ') is the coordinate of the target center in the lining map coordinate system. The transformation matrix T can be solved by the least squares method, random sampling consensus algorithm (RANSAC) and other methods. The transformation matrix T is used to transform the coordinates of the entire mine tunnel point cloud to obtain the registered point cloud data: The registered point cloud and lining map completely overlap in spatial position, achieving seamless integration of point cloud and image.

[0083] Dataset preparation: The tunnel lining images were cropped into sub-images of 1000×1000 pixels, and 322 shield tunnel datasets and 318 mining tunnel datasets were obtained. The training set and test set were randomly divided according to the ratio of 4:1, that is, 258 shield tunnel training sets and 64 test sets, and 254 mining tunnel training sets and 64 test sets. In order to expand the training data, the original images were enhanced. The specific method is: based on the original image, the images were rotated clockwise and counterclockwise by 5° and 10° respectively to generate 4 new images. After data enhancement, the shield tunnel training set was expanded to 1290 (258x5), and the validation set was expanded to 320 (64x5); the mining tunnel training set was expanded to 1275 (254x5), and the validation set was expanded to 315 (64x5).

[0084] Data annotation: LabelImg software is used to annotate the tunnel lining image dataset. LabelImg is an image annotation tool based on Python and Qt. It supports multiple annotation methods such as rectangular boxes and polygons. It is suitable for data preparation for tasks such as target detection and semantic segmentation. Load the lining images in turn on the LabelImg interface and annotate the water leakage areas in the images. The specific operation is: use the mouse to draw a rectangular box on the image so that it is close to the maximum boundary of the water leakage area, that is, the rectangular box should frame the entire water leakage area as much as possible, but it should not be too large to include too much background area. After drawing the rectangular box, select "water leakage" as the label name corresponding to the current rectangular box in the label list. If the "water leakage" label does not exist in the label list, you need to create this label. Repeat the above process of drawing rectangular boxes and specifying labels until all water leakage areas in the current image are annotated. Then save the annotation results and load the next image until the annotation task of the entire data set is completed. After the annotation is completed, LabelImg will generate an XML format annotation file with the same name as the image file in the directory where each image file is located. The annotation file records information such as the image name, the label name of the annotation object (ie, water leakage), the corresponding rectangular frame coordinates and size, etc.

[0085] Detection model construction, the tunnel water leakage detection model uses the current mainstream convolutional neural network (CNN) as the backbone network (Backbone) to extract multi-scale and multi-level features in tunnel lining images. Commonly used CNN backbone networks include VGG, ResNet, MobileNet, etc. They are pre-trained on large datasets such as Image Net and have strong feature extraction and semantic representation capabilities. Select a suitable backbone network, such as ResNet-50, as the feature extractor of the detection model. The backbone network consists of multiple convolutional layers (Conv), pooling layers (Pool) and fully connected layers (FC); the convolutional layer uses convolution kernels of different sizes such as 3x3 and 5x5 to extract local features of the image; the pooling layer uses the maximum pooling (Max Pooling) or average pooling (Average Pooling) to downsample the feature map to reduce the feature size; the fully connected layer classifies or regresses the extracted features and outputs the detection results. In order to further enhance the backbone network's feature extraction performance for tunnel lining images, the ConvNeXt Block subnetwork module is embedded in the backbone network. Conv NeXt Block is a novel CNN building block proposed by Facebook AI Research in 2022. It significantly improves the network's feature representation capability and computational efficiency through depthwise separable convolution and inverted residual structure. Depthwise separable convolution splits the standard convolution operation into two steps: channel-wise convolution and pointwise convolution, reducing the number of parameters and computation, and enhancing the network's nonlinear expression ability. Depthwise separable convolution splits the standard convolution operation into two steps: channel-wise convolution and pointwise convolution. Channel-wise convolution performs convolution on each input feature channel separately, using a convolution kernel size of (k, k, 1), where k is the convolution kernel size. Pointwise convolution uses a 1x1 convolution kernel to combine the output features of channel-wise convolution and adjust the number of feature channels. Depthwise separable convolution reduces the number of parameters and computation, and enhances the network's nonlinear expression ability.

[0086] The inverse residual structure adopts a bottleneck design of "dilation-convolution-compression". It first expands the number of feature channels through 1x1 convolution, then extracts features through 3x3 depth-separable convolution, and finally compresses the number of feature channels through 1x1 convolution, which reduces the computational complexity while maintaining the richness and diversity of features. Embedding ConvNeXt Block in several key layers of the backbone network (such as downsampling layers or feature fusion layers) can significantly enhance the network's feature extraction capabilities, especially the modeling capabilities of small targets such as leaking water. Add a detection head on top of the backbone network to generate the final detection results; the detection head can use a single-stage detector (such as YOLO, SSD) or a two-stage detector (such as Faster R-CNN, MaskR-CNN).

[0087] After the backbone network, a neck network (Neck) is connected in series to fuse the multi-scale feature maps output by the backbone network. The neck network can highlight the significant features of the water leakage area and improve the detection accuracy. Common neck networks include FPN, PAN, etc., which integrate multi-scale information through feature pyramid or feature fusion. The neck network adopts the CBAM (Convolutional Block Attention Module) attention mechanism module. CBAM adaptively adjusts the weight distribution of the feature map through channel attention (ChannelAttention) and spatial attention (SpatialAttention). CBAM can effectively highlight the feature response of the water leakage area, suppress the interference of the background area, and improve the detection accuracy. Channel attention learns the importance weights of different channels through global average pooling and fully connected layers. Global average pooling compresses the feature map of each channel into a scalar value to represent the global information of the channel. The fully connected layer converts the global information into a channel weight vector through learning to represent the importance of different channels. The channel weight vector is multiplied with the original feature map to adaptively enhance or suppress the feature response of different channels. Spatial attention learns the saliency weights of different spatial positions through convolutional layers and Sigmoid functions. The convolutional layer performs convolution operations on the original feature map to extract local features of different spatial positions. The Sigmoid function maps the convolution results to a spatial weight matrix in the range of (0, 1) to represent the saliency of different positions. The spatial weight matrix is ​​multiplied by the original feature map to adaptively enhance or suppress the feature responses of different spatial positions. The CBAM module is placed after the backbone network and before the neck network, playing a connecting role.

[0088] The CBAM module can make full use of the high-level semantic features learned by the backbone network and perform adaptive channel and spatial modulation on them. The modulated feature map can provide a more refined and focused feature representation for the subsequent detection head, which is conducive to improving the detection performance. The CBAM module consists of a channel attention submodule and a spatial attention submodule, which can be used in series or in parallel. The channel attention submodule includes a global average pooling layer, a fully connected layer, and a Sigmoid activation function. The spatial attention submodule includes a convolutional layer, a BatchNorm normalization layer, and a Sigmoid activation function. The hyperparameters of the CBAM module, such as the number of hidden units in the fully connected layer and the size of the convolution kernel, can be adjusted according to task requirements. Suppose the feature map output by the backbone network is F, and its size is (C, H, W), that is, the number of channels is C, the height is H, and the width is W. The channel attention submodule compresses F to (C, 1, 1) through global average pooling, and then calculates the channel weight vector W through the fully connected layer and the Sigmoid function. c , the size is (C, 1, 1), the spatial attention submodule calculates the spatial weight matrix W through the convolution layer and the Sigmoid function of F s , with a size of (1, H, W), the channel weight vector W c and the spatial weight matrix W s Multiply it with the original feature map F to get the modulated feature map F', the size is still (C, H, W).

[0089] The loss function design of the detection model uses GIOU (Generalized Intersection over Union) Loss as the positioning loss function of the detection model to optimize the model's spatial positioning ability for the leakage area. Compared with the traditional IOU Loss, GIOULoss not only considers the overlap between the predicted box and the true box, but also considers geometric factors such as the center distance between the two boxes. Therefore, it has a stronger constraint and guidance on the regression optimization of the detection box. IOU Loss only considers the intersection over union ratio of the predicted box and the true box, and has two defects: when the two boxes do not overlap, IOU is always 0, and the loss function cannot optimize the detection box; when the two boxes overlap but the center distance is large, IOU Loss cannot well reflect the actual distance between them, resulting in inaccurate optimization direction. GIOULoss introduces the concept of the smallest enclosing box of the two boxes on the basis of IOU. By calculating the IOU between the predicted box and the true box and the smallest enclosing area, a more comprehensive and detailed loss metric is obtained. When the two boxes do not overlap, GIOULoss uses the minimum closure area as a reference to obtain a negative value, which can guide the detection box to move toward the true box; when the two boxes overlap but the center distance is large, GIOULoss uses the minimum closure area as a reference to obtain a smaller positive value, which can guide the detection box to move closer to the true box. GIOULoss is used as the positioning loss and is weighted and summed with the classification loss (such as FocalLoss) to obtain the total loss function of the detection model. During the model training process, the optimization iteration of the model parameters is achieved by minimizing the total loss function, thereby improving the performance of the detection model in the task of water leakage positioning and identification.

[0090] Specifically, in this embodiment, in a certain training sample, the coordinates of the tunnel defect (such as water leakage) bounding box predicted by the tunnel detection model and the ground truth bounding box are: Tunnel detection model predicted bounding box: upper left corner coordinate (x 1p ,y 1p )=(50,80), the coordinate of the lower right corner (x 2p ,y 2p )=(200,280); predicted bounding box area Ap=(200-50)*(280-80)=30000 square pixels. Ground truth bounding box: upper left corner coordinate (x 1g ,y 1g )=(80,100), the lower right corner coordinate (x 2g ,y 2g)=(240,300); the area of ​​the ground truth bounding box Ag=(240-80)*(300-100)=32000 square pixels. The calculation process is as follows: Calculate the intersection area U of the predicted bounding box and the ground truth bounding box: the upper left corner coordinate (x 1i ,y 1i )=(max(x 1p ,x 1g ),max(y 1p ,y 1g ))=(max(50,80),max(80,100))=(80,100); The lower right corner coordinate of the intersection (x 2i ,y 2i )=(min(x 2p ,x 2g ),min(y 2p ,y 2g ))=(min(200,240),min(280,300))=(200,280); U=(200-80)*(280-100)=21600 square pixels. Calculate the union area A of the predicted bounding box and the ground truth bounding box: A=Ap+Ag-U=30000+32000-21600=40400 square pixels. Calculate the minimum rectangular box area C that contains the predicted bounding box and the ground truth bounding box: the coordinates of the upper left corner of the minimum rectangular box are (min(50, 80), min(80, 100)) = (50, 80); the coordinates of the lower right corner of the minimum rectangular box are (max(200, 240), max(280, 300)) = (240, 300); C = (240-50)*(300-80) = 41800 square pixels; Calculate the intersection over union (IOU): IOU = U / A = 21600 / 40400 = 0.5347; Substitute the above calculation results into the GIOU-Loss formula: GIOU-Loss = 1-IOU + (CA) / C; = 1-0.5347 + (41800-40400) / 41800; = 0.4987. In this specific sample, the positioning loss GIOU-Loss of the tunnel detection model is 0.4987. By minimizing GIOU-Loss, the defect bounding box predicted by the detection model can be closer to the ground truth.

[0091] S5, using the tunnel detection model, detect the shield tunnel point cloud after ring segmentation or the mine tunnel point cloud after positioning, and obtain the detection results of the shield tunnel or mine tunnel, which include tunnel leakage, cable loosening and equipment changes. For shield tunnels, the obtained tunnel point cloud data after ring segmentation and for mine tunnels, the obtained tunnel point cloud data after positioning are used to convert the point cloud into a two-dimensional tunnel lining image according to the scanner's mileage data and tunnel coordinate system. The tunnel lining image is cropped into sub-images of a fixed size (such as 256x256); data enhancement is performed on the sub-images, such as random flipping and rotation, to enrich data diversity; the sub-images are standardized, the mean is subtracted and divided by the standard deviation, to accelerate model convergence. The preprocessed sub-images are input into the tunnel detection model. The model detects various tunnel defects in the lining image through feature extraction and attention mechanism, including water leakage, cable loosening and equipment changes, and the model outputs the detected defect location coordinates and category confidence.

[0092] The model output results are subjected to non-maximum suppression (NMS) to filter out overlapping detection frames. After the model infers each lining sub-image, it usually outputs multiple overlapping detection frames, which need to be screened. First, the detection frames are sorted in descending order according to their category confidences. Then, the detection frame A with the highest confidence is selected, and the IoU (Intersection over Union) between A and other detection frames is calculated. The detection frames whose IoU with A is greater than a preset threshold (such as 0.5) are filtered out, and A with the highest confidence is retained. The above process is recursively performed on the remaining detection frames until all detection frames are processed, and finally a set of non-overlapping tunnel defect detection frames and their position coordinates and categories are obtained. According to the position coordinates of the detection frames, the index relationship established in claim 7 is used to map the detection results from the lining image back to the three-dimensional point cloud space.

[0093] After establishing a one-to-one correspondence between the tunnel point cloud and the lining image, for each pixel point P(u, v) on the lining image, the corresponding point C(x, y, z) in the point cloud can be found through the row and column indexes (r, c). Specifically, the pixel coordinates of the upper left corner and lower right corner of the detection box B are (u1, v1) and (u2, v2), respectively. According to the index relationship, the corresponding coordinates C1(x1, y1, z1) and C2(x2, y2, z2) in the point cloud of (u1, v1) and (u2, v2) are calculated respectively. The location of the tunnel defect in the three-dimensional point cloud space is determined by C1 and C2. The cubic bounding box visually marks the detected tunnel defect location and category in the point cloud, and defines a set of unique colors for different categories of tunnel defects, such as blue for water leakage, yellow for loose cables, and red for equipment changes. For each detection box B, at the corresponding bounding box positions C1 and C2 of the 3D point cloud, all points in the box are rendered as the color corresponding to the defect category. At the same time, a text label is generated on the top of the bounding box to indicate the category name of the defect. Use point cloud visualization engines such as Open3D, PCL, etc. to render the annotated tunnel point cloud to intuitively display the distribution of tunnel defects.

[0094] The category, location, quantity and other information of water leakage, loose cables and equipment changes are organized into an inspection report output. All tunnel defect bounding boxes are traversed to extract information such as their category, center point coordinates, length, width and height. The defects are classified and summarized according to category, and the number of each type of defects is calculated. The extracted information is filled into the preset inspection report template to generate a structured tunnel defect inspection report. The report intuitively presents the distribution characteristics of tunnel defects in the form of tables and charts. The inspection report is sent to tunnel maintenance personnel through web pages, PDFs, etc. to guide actual maintenance work.

Claims

1. A mobile tunnel detection method, characterized in that: include: Based on the tunnel construction method, tunnels are divided into shield tunnels and mine tunnels; For shield tunnels, the 3D point cloud of the shield tunnel is obtained, and the CSF algorithm and DBSCAN algorithm are used to extract the bolt holes; The mean-shift algorithm is used to obtain the center point of the bolt hole, and the three-dimensional point cloud of the shield tunnel is divided into rings and blocks according to the center point of the bolt hole to obtain the shield tunnel point cloud after ring and block division; For mine tunnels, a mobile tunnel laser device with target placement is used to place a target at a preset distance along the tunnel axis to obtain a mine tunnel point cloud with target position information. Generate mine tunnel lining diagram based on mine tunnel point cloud with target location information; Establish the index relationship between the mine tunnel point cloud and the mine tunnel lining diagram; Positioning the mine tunnel point cloud according to the index relationship to obtain the positioned mine tunnel point cloud; Build a tunnel detection model based on deep learning. The tunnel detection model includes a backbone network and a neck network. The backbone network adopts convolutional neural network, and ConvNeXt Block network is set in the backbone network to extract the characteristics of tunnel leakage; The neck network adopts the CBAM attention mechanism to enhance the weight of tunnel water leakage features in the channel dimension and spatial dimension; The tunnel detection model is used to detect the shield tunnel point cloud after ring and block division or the mine tunnel point cloud after positioning to obtain the detection results of the shield tunnel or mine tunnel. The detection results include tunnel leakage, cable looseness and equipment changes.

2. The mobile tunnel detection method according to claim 1, characterized in that: For shield tunnels, obtain the 3D point cloud of the shield tunnel, including: Obtain 3D point cloud data of shield tunnel; Establish a tunnel section coordinate system with the geometric center of the tunnel as the origin, the vertical upward Z axis as the positive direction, and the X axis perpendicular to the Z axis and pointing to the right; Calculate the angle θ between the tunnel wall point and the cross-section coordinate system; According to the tunnel design parameters and the angle θ, the tunnel wall points are converted from the cross-section coordinate system to the coordinates of the plane rectangular coordinate system to obtain the unfolded tunnel wall point cloud.

3. The mobile tunnel detection method according to claim 2, characterized in that: Calculate the angle θ between the tunnel wall point and the cross-section coordinate system using the following formula: Where: x i is the horizontal coordinate of the point on the cross section, z i is the vertical coordinate of the point on the cross section, x0 is the horizontal coordinate of the geometric center of the tunnel, and z0 is the vertical coordinate of the geometric center of the tunnel.

4. The mobile tunnel detection method according to claim 2, characterized in that: The CSF algorithm and DBSCAN algorithm are used to extract bolt holes, including: Perform coordinate transformation on the unfolded tunnel wall point cloud to obtain a tunnel wall point cloud that is consistent with the direction of the initially collected three-dimensional point cloud; Construct the CSF model, initialize the cloth mesh, and set the mesh resolution to D times the average point spacing of the expanded tunnel wall point cloud; the initial cloth position is within the preset range above the highest point of the point cloud; The CSF model is used to obtain the classification threshold hcc of the tunnel wall point cloud. The classification threshold hcc is used to distinguish between tunnel wall points and bolt hole points. Iteratively calculate the height difference between each point in the tunnel wall point cloud and the CSF model cloth particles; The points with height difference greater than the classification threshold hcc are marked as potential bolt hole point clouds, otherwise they are marked as tunnel wall point clouds; The DBSCAN algorithm is used to cluster the point clouds marked as potential bolt holes to obtain the final bolt hole point clouds.

5. The mobile tunnel detection method according to claim 4, characterized in that: The three-dimensional point cloud of the shield tunnel is divided into rings and blocks according to the center points of the bolt holes, including: According to the collected shield tunnel 3D point cloud, using the final bolt hole point cloud, the mean-shift clustering algorithm is used to cluster each bolt hole and obtain the center point of the corresponding bolt hole; According to the preset splicing block design data, an identification template of the center of the bolt hole of the splicing block is defined, and the center point of the obtained bolt hole is matched with the identification template to obtain the center point of the bolt hole of different splicing blocks; Determine the center point of the bolt hole of each splicing block according to the relative position relationship of the center points of the bolt holes of different splicing blocks; Using the center points of the bolt holes of adjacent splicing blocks, the longitudinal joints between adjacent splicing blocks are located by interpolation and linear fitting; Using the center points of the bolt holes in the same splicing block, the longitudinal joints between adjacent splicing blocks are located by linear fitting and translation; Using the center points of the bolt holes of two adjacent rings, the annular joint is located by interpolation and linear fitting; The collected three-dimensional point cloud of the shield tunnel is divided into rings and blocks by using the located seams to obtain the shield tunnel point cloud after ring and block division.

6. The mobile tunnel detection method according to claim 5, characterized in that: The splicing blocks include: capping blocks, adjacent blocks and standard blocks; The capping block is located at the tunnel vault position and is composed of multiple splicing blocks. Multiple bolt holes are arranged in the capping block. The center points of the bolt holes in the capping block are distributed in a ring shape. The spatial position of the tunnel vault is determined by identifying the center points of the bolt holes in the capping block. The adjacent blocks are located on both sides of the capping block, adjacent to and connected to the capping block, and a plurality of bolt holes are arranged in the adjacent blocks, and the center points of the bolt holes in the adjacent blocks are distributed in the same horizontal plane; by identifying the center points of the bolt holes in the adjacent blocks, the spatial positions of the two sides of the tunnel are determined; The standard blocks are located below the capping blocks and adjacent blocks, and are arranged longitudinally and circumferentially along the tunnel to form the main body of the tunnel lining structure. A plurality of bolt holes are provided in the standard blocks, and the center points of the bolt holes in the standard blocks are distributed in the same vertical plane. By identifying the center points of the bolt holes in the standard blocks, the longitudinal and circumferential joints of the tunnel lining are determined.

7. The mobile tunnel detection method according to claim 5, characterized in that: Establish the index relationship between the mine tunnel point cloud and the mine tunnel lining diagram, including: Use a scanner to collect two-dimensional cross-section point clouds along the design axis of the mine tunnel; According to the mileage data of the scanner, the collected two-dimensional cross-section point cloud is converted into a three-dimensional point cloud; The design axis direction of the mine tunnel is taken as the positive direction of the y-axis, the normal direction of the section where the center point of the scanner is located is taken as the positive direction of the z-axis, and the three-dimensional coordinate system of the tunnel is established with the center point as the origin o. Along the positive direction of the y-axis, the sections perpendicular to the y-axis are numbered according to the acquisition order and a column index is established; Taking the intersection point P of the negative direction of the z-axis and the cross section as the starting point, number the point cloud in the cross section in a clockwise direction to establish a row index; According to the row index and column index, the three-dimensional coordinates and point cloud information of the corresponding point cloud are written into the pixel matrix, and the index relationship between the mine tunnel point cloud and the mine tunnel lining map is established.

8. The mobile tunnel detection method according to claim 7, characterized in that: The mine tunnel point cloud is located according to the index relationship to obtain the located mine tunnel point cloud, including: A training set is constructed based on the preset layout rules of the targets in the mine tunnel lining diagram; the preset layout rules include the equal spacing of the targets along the tunnel axis and the black and white spacing of the target images; Use the training set to train the YOLOv7 model to obtain the target recognition model; Using the target recognition model, the pixel coordinates of each target in the mine tunnel lining map are obtained; According to the pixel coordinates of the targets, the three-dimensional coordinates of each target are obtained in the mine tunnel point cloud by using the index relationship between the mine tunnel point cloud and the mine tunnel lining map; According to the three-dimensional coordinates of the target, the mine tunnel point cloud is transformed to align the mine tunnel point cloud with the mine tunnel lining map to obtain the positioned mine tunnel point cloud.

9. The mobile tunnel detection method according to any one of claims 2 to 8, characterized in that: Build a tunnel detection model based on deep learning, including: Obtain the shield tunnel point cloud data after ring division or the mine tunnel point cloud data after positioning, and convert the point cloud data into a tunnel lining image; The tunnel lining image is cropped into sub-images of size MxM, and a shield method tunnel lining image subset and a mining method tunnel lining image subset are obtained; Preprocess the shield tunnel lining image subset and the mining tunnel lining image subset respectively; The pre-processed shield tunnel lining image subset and mining tunnel lining image subset are annotated to mark tunnel defects; tunnel defects include tunnel water leakage; The preprocessed and annotated subsets of shield tunnel lining images and mining tunnel lining images are used as training sets, and a convolutional neural network is used as the backbone network. A ConvNeXt Block network is set in the backbone network to extract tunnel defects. A neck network is set after the backbone network, and the neck network adopts the CBAM attention mechanism; In the training process of the tunnel detection model, GIOU-Loss is used as the positioning loss function; The trained tunnel detection model is used to detect the shield tunnel point cloud after ring and block division or the mine tunnel point cloud after positioning. Through the index relationship between the point cloud and the tunnel lining image, the detected tunnel defects are mapped to the point cloud position, and the three-dimensional spatial position of the tunnel defect is obtained as the tunnel detection result output.

10. A mobile tunnel detection system, characterized in that: include: At least one processing unit; used to execute instructions to implement the mobile tunnel detection method according to any one of claims 1 to 9.

Images