Tunnel face support parameter determination method based on RGB-D fusion

Abstract

The invention discloses a tunnel face support parameter determination method based on RGB-D fusion, and the method comprises the steps: inputting a tunnel face RGB image and a depth map into an RGB-D double-flow feature fusion network, and outputting a structural plane planar region semantic mask and a confidence map; a pixel set corresponding to the structural plane semantic mask is projected back to the point cloud space to obtain a structural plane three-dimensional point set, plane robust fitting is conducted on the structural plane three-dimensional point set, a structural plane fitting plane is obtained, and structural plane occurrence parameter data are obtained; arranging a virtual measuring line set in the three-dimensional structural plane model, and calculating an RQD value of each virtual measuring line and an RQD statistical summary value of a plurality of virtual measuring lines; based on the RQD statistical summary value and the structural plane occurrence parameter data, the stability of the surrounding rock is evaluated; and on the basis of a pre-constructed support parameter recommendation table, obtaining a support parameter recommendation interval matched with the surrounding rock stability evaluation result. According to the invention, integration of surrounding rock grading and support parameter output can be realized.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering surrounding rock control and digital surveying of tunnel face, specifically to a method for determining tunnel face support parameters based on RGB-D fusion. Background Technology

[0002] The classification of surrounding rock and the determination of support parameters in tunnels usually rely on manual recording, manual surveying and experience-based interpretation at the tunnel face. This method is significantly affected by factors such as light reflection, water stains and dust, as well as differences among personnel, resulting in low efficiency and insufficient consistency.

[0003] Existing technologies can identify crack traces in tunnel face images or extract cracks from point clouds, but they mostly remain at the level of "recognition / display," making it difficult to form a three-dimensional model of the structural surface and parameters such as attitude and spacing that can be directly used for engineering evaluation.

[0004] On the other hand, rock mass quality indicators such as RQD / RMR are universal in engineering, but traditional calculations rely on manual survey lines or a small number of boreholes, which are difficult to link with digital data of the working face. Moreover, the evaluation results fluctuate greatly and are difficult to directly convert into actionable recommendations for support parameters such as shotcrete, anchor bolts and steel arches. Summary of the Invention

[0005] This invention provides a method for determining tunnel face support parameters based on RGB-D fusion, which solves the problems of strong subjectivity in the evaluation of surrounding rock at the tunnel face, non-repeatable RQD calculation, and difficulty in directly guiding support parameters by classification in the prior art, and realizes the integration of surrounding rock classification and support parameter output.

[0006] According to the first aspect, one embodiment provides a method for determining tunnel face support parameters based on RGB-D fusion, the method comprising: Simultaneously acquire RGB images and 3D point cloud data of the tunnel face, and generate a depth map D that is pixel-aligned with the RGB image, establishing the correspondence between the RGB image, depth map D, and 3D point cloud in the same coordinate system; The RGB image and depth map are input into a pre-trained RGB-D dual-stream feature fusion network, which outputs a semantic mask and confidence map of the structural planar region. The pixel set corresponding to the semantic mask of the structural surface is projected back into the point cloud space to obtain the three-dimensional point set of the structural surface. The three-dimensional point set of the structural surface is subjected to planar robust fitting processing to obtain the structural surface fitting plane. The structural surface attitude parameter data is obtained based on the fitting plane. In the reconstructed three-dimensional structural surface model, a set of virtual survey lines is laid out, the RQD value of each virtual survey line is calculated, and the RQD statistical summary value of multiple virtual survey lines is obtained. Based on the obtained RQD statistical summary value and structural surface attitude parameter data, the stability of the surrounding rock is evaluated. Based on a pre-constructed support parameter recommendation table, a support parameter recommendation interval matching the surrounding rock stability evaluation results is obtained. Using the support parameter recommendation interval as candidates, the optimal support parameter combination is obtained through optimization.

[0007] Furthermore, an RGB-D dual-stream feature fusion network is constructed, specifically including: RGB encoding branch Used for input RGB images Extracting multi-scale texture semantic features ; Deep coding branches , used for the input depth map Extracting multi-scale geometric features ; Cross-modal interaction module Used to apply at least one scale layer and Alignment and fusion are performed to obtain fused features. ; Decoding branch It is used to upsample the fused features and fuse them with the encoded features in a skip connection to obtain the decoded features. ; Output head , for use based Generate semantic masks for surface-shaped regions. ; , for use based Generate confidence plot ,in It characterizes the degree of confidence that a pixel belongs to a structural plane region.

[0008] Furthermore, the RGB-D two-stream feature fusion network is trained, including: Multiple sets of RGB images and pixel-level aligned depth maps were acquired at the tunnel face. The true value mask for the planar region of the structure is annotated manually or semi-automatically. to form training samples ; After data augmentation, the samples are input into an RGB-D dual-stream feature fusion network for training. The network parameters are iteratively updated through backpropagation until the validation set index converges, resulting in an RGB-D dual-stream feature fusion network for structural surface recognition. During the online inference phase, semantic masks of the structural planar regions output by the RGB-D dual-stream feature fusion network are applied. Post-processing is performed, including thresholding, connected component filtering, hole filling, and small region removal, to obtain a set of structural surface regions.

[0009] Furthermore, the pixel set corresponding to the semantic mask of the structural surface region is projected back into the point cloud space to obtain the three-dimensional point set of the structural surface. Planar robust fitting processing is then performed on the three-dimensional point set of the structural surface to obtain the structural surface fitting plane. Based on the fitting plane, structural surface attitude parameter data is obtained, specifically including: The obtained semantic mask of the structural surface region Back to point cloud Obtain the set of structural surface points and to Divide into several subsets ; For each The structural plane is obtained by performing a robust planar fitting using RANSAC or weighted RANSAC. and its normal direction And extract the structural surface boundary as a finite patch based on the point set projection; thus, The components are combined to form a three-dimensional structural surface model, which is used for subsequent intersection determination of virtual survey lines and structural surfaces; The orientation, dip, and dip angle of the structural surfaces are calculated based on the fitted plane normal vectors, and the number of structural surface groups, the spacing between structural surfaces, and the extension scale parameters are statistically analyzed.

[0010] Furthermore, a set of virtual survey lines is laid out in the reconstructed 3D structural surface model, and the RQD value of each virtual survey line is calculated, specifically including: Virtual survey lines are laid out at preset intervals within the projection area of ​​the tunnel face. The tunnel should be evenly distributed and include at least three directions: along the tunnel axis, radial, and circumferential. Each measuring line Expressed using the three-dimensional linear parametric equations as follows: , As the starting point of the survey line, Unit direction vector, survey line parameters , This represents the total length of the survey line; For any survey line Calculate its relationship with each structural surface in the three-dimensional structural surface model. The set of intersection points, and along the survey line parameters Sort the intersection points to obtain the intersection point sequence. The distance between adjacent intersection points is defined as the equivalent core segment length of the survey line, i.e.:

[0011] In the formula, Indicates the first The equivalent core segment length of the kth segment on the test line; and They represent survey lines respectively. The kth and (k+1)th intersection points obtained after intersecting with the three-dimensional structural surface model are sorted in ascending order according to the survey line parameter s; Represents the Euclidean norm; ,in For surveying lines Number of intersection points; Will As a survey line The corresponding set of equivalent core segment lengths, the physical meaning of which is: the equivalent core segment length obtained by dividing the virtual survey line under the cutting action of the structural surface, used to simulate the statistical process of the length of the complete core segment in the drill core; Will satisfy The segment lengths are summed and divided by the total length of the survey line. , obtain the survey line of

[0012]

[0013] In the formula, Indicates the first strip survey line Core recovery rate index; For surveying lines The total length; The effective segment threshold; where This is an indicator function that takes the value 1 if the condition is true and 0 otherwise. Indicates the survey line Summing the lengths of segments formed by all adjacent intersection points, where only the segments satisfying the condition are summed. The segment length is included in the statistics; For multiple survey lines The calculation results are used to evaluate stability, and the stability evaluation indicators include variance. or confidence interval width Variance or confidence interval width With the corresponding threshold , The comparison is used as a convergence criterion. If the convergence condition is not met, the number of survey lines is increased or the spacing is decreased according to the adaptive densification strategy. In particular, the RQD results are made more convergent by prioritizing the densification of the RQD array in areas with high fracture density or large RQD fluctuations.

[0014] Furthermore, the RQD statistical summary values ​​of multiple virtual survey lines were obtained, specifically including: If a total of n survey lines are set up, the overall RQD is obtained by taking the statistical summary value of the results of multiple survey lines and then calculating the weighted average or arithmetic average. The formula for calculating the weighted average is as follows:

[0015] When the total length of each survey line is the same or no weighting is required, the arithmetic mean calculation formula is used:

[0016] In the formula, RQD represents the statistical summary value of the results from multiple survey lines; n is the number of survey lines; For the first The total length of the survey lines is used as a weighting factor; For the first Core recovery rate index obtained from the measurement line.

[0017] Furthermore, based on the obtained RQD statistical summary values ​​and structural plane attitude parameter data, the stability of the surrounding rock is evaluated, specifically including: Determine the surrounding rock grade and RMR score, with the RMR score as an alternative. Among them, a preset rule function is used. Determining the grade of surrounding rock ,in The evaluation index vector includes the RQD statistical summary value, structural surface attitude parameter data, and related indicators; To utilize threshold segmentation rules, rule trees, or trained mapping models; The RMR score is obtained by summing the scores of each item according to the preset RMR scoring table / scoring rules, as shown in the following formula:

[0018] in For the first Individual section scores, The number of items; The method of obtaining it is: to take the first Input indicators for each item The sub-item scores are obtained by interval mapping according to a preset scoring table, or by calculating the sub-item scores according to a preset scoring rule function, that is:

[0019] in, For the first The scoring mapping function for each sub-indicator is used to map the sub-indicators to the sub-indicators. Mapped to corresponding sub-item scores , Implemented using a pre-set scoring table or segmented mapping rules: The range of values ​​is divided into several intervals, each corresponding to a specific score. Output the corresponding score when the value falls into the corresponding interval; For the first The input indicators include the RQD statistical summary value, structural surface attitude parameter data, and related indicators.

[0020] Furthermore, based on the pre-constructed support parameter recommendation table, a recommended range of support parameters matching the surrounding rock stability evaluation results is obtained, specifically including: Based on the determined surrounding rock grade The system outputs recommended support parameter ranges based on the pre-built support parameter recommendation table, along with the range of alternative RMR scores. The support parameters include: shotcrete thickness. Anchor bolt length Anchor spacing Spacing of steel arches Or steel arch frame model; the recommended support parameter table is used to establish the surrounding rock grade. It can also map the RMR score range to the recommended range of support parameters and provide two sets of recommended ranges for support parameters: one for economical use and one for conservative use.

[0021] Furthermore, using the recommended range of support parameters as candidates, the optimal combination of support parameters is obtained through optimization, specifically including: The feasible region is formed by the recommended range of support parameters given in the support parameter recommendation table. An objective function is established to minimize cost / construction period, and deformation and deformation rate constraints are established. The optimal combination of support parameters that satisfies the deformation and deformation rate constraints is obtained by optimization within the feasible region.

[0022] According to a second aspect, one embodiment provides a tunnel face support parameter determination system based on RGB-D fusion, the system comprising: The data acquisition and alignment module is used to simultaneously acquire RGB images and 3D point cloud data of the tunnel face, and generate a depth map D that is pixel-level aligned with the RGB image, establishing the correspondence between the RGB image, the depth map D, and the 3D point cloud in the same coordinate system. The structural surface recognition module is used to input RGB images and depth maps into a pre-trained RGB-D dual-stream feature fusion network and output a semantic mask and confidence map of the structural surface region. The structural plane fitting module is used to back-project the pixel set corresponding to the semantic mask of the structural plane region back into the point cloud space to obtain the three-dimensional point set of the structural plane, perform planar robust fitting processing on the three-dimensional point set of the structural plane to obtain the structural plane fitting plane, and obtain the structural plane attitude parameter data based on the fitting plane. The RQD value statistics module is used to lay out a set of virtual survey lines in the reconstructed 3D structural surface model, calculate the RQD value of each virtual survey line, and obtain the RQD statistical summary value of multiple virtual survey lines. The stability evaluation module is used to evaluate the stability of the surrounding rock based on the obtained RQD statistical summary value and structural surface attitude parameter data. The support parameter recommendation module is used to obtain the recommended range of support parameters that matches the surrounding rock stability evaluation results based on the pre-built support parameter recommendation table, and to obtain the optimal combination of support parameters by using the recommended range of support parameters as candidates through optimization solution.

[0023] According to three aspects, one embodiment provides an electronic device, the device comprising: a processor and a memory; The memory is used to store one or more program instructions; The processor is configured to run one or more program instructions to perform the steps of a method for determining tunnel face support parameters based on RGB-D fusion as described in any of the preceding claims.

[0024] According to a fourth aspect, one embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a method for determining tunnel face support parameters based on RGB-D fusion as described in any of the preceding claims.

[0025] This invention provides a method for determining tunnel face support parameters based on RGB-D fusion, which has the following advantages: 1) The structural surface “surface” is automatically reconstructed from the RGB-D data of the working face and the attitude parameters such as strike, dip and dip angle are inverted, reducing the workload of manual recording and measurement and improving the consistency of evaluation.

[0026] 2) An adaptive virtual survey line array RQD calculation mechanism is proposed. Through multi-survey line statistical stability evaluation and survey line densification strategy, the repeatability and stability of RQD calculation are improved.

[0027] 3) The RQD and surrounding rock evaluation grade (RMR optional) are directly mapped to the recommended range of shotcrete thickness, anchor length / spacing and steel arch spacing, and can output two sets of parameter combinations, namely the economic scheme and the conservative scheme, to facilitate rapid on-site decision-making. Attached Figure Description

[0028] Figure 1 A flowchart illustrating a method for determining tunnel face support parameters based on RGB-D fusion, as provided in one embodiment of the present invention; Figure 2A flowchart illustrating the specific implementation of a method for determining tunnel face support parameters based on RGB-D fusion, as provided in one embodiment of the present invention; Figure 3 A schematic diagram of RGB-D alignment and structural surface semantic mask in a method for determining tunnel face support parameters based on RGB-D fusion, provided in an embodiment of the present invention; Figure 4 A schematic diagram of semantic mask projection point cloud, robust fitting and attitude inversion in a method for determining tunnel face support parameters based on RGB-D fusion provided in an embodiment of the present invention; Figure 5 This diagram illustrates the adaptive virtual survey line layout, RQD statistics, and support parameter output in a tunnel face support parameter determination method based on RGB-D fusion, as provided in one embodiment of the present invention. Detailed Implementation

[0029] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0030] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.

[0031] The first embodiment of this invention provides a method for determining tunnel face support parameters based on RGB-D fusion. The following is in conjunction with... Figure 1 and Figure 2 Please provide a detailed explanation.

[0032] like Figure 1As shown, in step S100, RGB images and three-dimensional point cloud data of the tunnel face are acquired simultaneously, and a depth map D is generated that is pixel-level aligned with the RGB image, establishing the correspondence between the RGB image, the depth map D, and the three-dimensional point cloud in the same coordinate system.

[0033] In this embodiment, a depth camera, a structured light camera, a lidar, or a combination thereof are used to simultaneously acquire RGB color images and 3D point clouds of the tunnel face. The camera intrinsic and extrinsic parameters are obtained through calibration, and the 3D point cloud is then processed. Projecting the image onto the image plane according to the camera calibration parameters generates a depth map aligned with the RGB image at the I-pixel level. (Depth Map), such as Figure 3 As shown. Among them, the depth map It is a two-dimensional matrix of the same size as image I. Represents pixel coordinates The corresponding depth value at that location. Its projection relationship can be expressed as: (1) To enable the subsequent back projection of the semantic mask onto the 3D point cloud, the back projection relationship between pixels and 3D points is established: (2) In equation (1), Scale factor; These are pixel coordinates; This is the camera intrinsic parameter matrix; and These are the camera's extrinsic rotation matrix and translation vector, respectively. For point clouds Coordinates of a three-dimensional point in a coordinate system.

[0034] In equation (2), For depth map In pixels Depth value. The RGB image I, depth map, and point cloud are obtained through equations (1)-(2). Correspondence in the same coordinate system.

[0035] like Figure 1 As shown, in step S200, the RGB image and depth map are input into a pre-trained RGB-D dual-stream feature fusion network, which outputs a semantic mask and confidence map of the structural surface region.

[0036] In this embodiment, a structural surface recognition model is constructed, which aligns the RGB image I with the depth map. Input model, output semantic mask of structural planar regions and confidence plot ,like Figure 3As shown, this model employs an RGB-D dual-stream feature fusion network and sets up cross-modal interactions in multi-scale feature layers to utilize geometric information to suppress lighting artifacts and texture interference.

[0037] An RGB-D two-stream feature fusion network should include at least: RGB encoding branches. Deep coding branches Cross-modal interaction module (Used for aligning and fusing geometric and texture information at multi-scale feature layers), decoding branch (Upsampling the fused features step by step and fusing them with the encoder features in skip connections to recover pixel-level details), and the output head. and Generate semantic masks respectively With confidence plot .

[0038] RGB encoding branch Used for input RGB images Extracting multi-scale texture semantic features ; Deep coding branches , used for the input depth map Extracting multi-scale geometric features ; Cross-modal interaction module Used to apply at least one scale layer and Alignment and fusion are performed to obtain fused features. ; Decoding branch It is used to upsample the fused features and fuse them with the encoded features in a skip connection to obtain the decoded features. ; Output head , for use based Generate semantic masks for surface-shaped regions. ; , for use based Generate confidence plot ,in It characterizes the degree of confidence that a pixel belongs to a structural plane region.

[0039] The input-output relationship of each module can be represented as follows: , , , , , .in Used to characterize pixels The reliability of the structural surface region can be used to assign weights to point cloud points during subsequent 2D-3D retrospective projection and to perform weighted sampling or weighted residual calculations during structural surface fitting, thereby improving robustness.

[0040] The structural surface recognition model is preferably implemented using an "offline training + online inference" approach: multiple sets of RGB images and pixel-level aligned depth maps are acquired at the tunnel face. The true value mask for the planar region of the structure is annotated manually or semi-automatically. To form training sample pairs After data augmentation such as rotation, cropping, brightness perturbation, and depth noise perturbation, the samples are input into a two-stream network for training. The training process can employ a segmentation loss function. (For example, a weighted combination of cross-entropy / binary cross-entropy and Dice loss), and consistency constraint loss can be introduced to the confidence branch. The network parameters are iteratively updated through backpropagation until the validation set index converges, thus obtaining the structural surface recognition model parameters for online inference.

[0041] For output mask Post-processing is performed, including thresholding, connected component filtering, hole filling, and small region removal, to obtain a set of structural surface regions.

[0042] Unlike schemes that only output fracture traces, this step outputs a "planar region" mask of the structural surface, providing a complete point set support for subsequent 3D fitting and attitude inversion.

[0043] like Figure 1 As shown, in step S300, the pixel set corresponding to the semantic mask of the structural surface area is projected back into the point cloud space to obtain the three-dimensional point set of the structural surface. The three-dimensional point set of the structural surface is subjected to planar robust fitting processing to obtain the structural surface fitting plane. The structural surface attitude parameter data is obtained based on the fitting plane.

[0044] In this embodiment, as Figure 4 As shown, the pixel set corresponding to the planar mask M of the structure surface is back-projected into the point cloud space according to Equation (2) to extract the three-dimensional point set of the structure surface. The three-dimensional point set of the structural surface is processed using either Random Sample Consensus Algorithm (RANSAC) or Weighted RANSAC. For robust planar fitting, the fitting residuals can be characterized by the distance from the point to the fitting plane: (3)

[0045] (5) In equations (3), (4), and (5), For point The distance residual to the fitted plane, For point set The Middle Three-dimensional points. Fitting plane parameters. Satisfying the plane equation Based on the residual threshold Determine interior points ( It can extract multiple structural surfaces using a "fit-reject-refit" method. In weighted RANSAC, pixel confidence can be... The data is then projected back to the corresponding 3D point and used as the sampling probability or residual weight to improve robustness in areas with lighting artifacts / low texture.

[0046] The strike, dip, and dip angle of the structural planes are calculated based on the fitted plane normal vectors, and parameters such as the number of structural plane groups, spacing, and extension scale are statistically analyzed. The structural plane parameters obtained by robust fitting are... Construct the normal vector of the structural surface And normalize it to obtain the unit normal vector. ( (The unit normal vector).

[0047] Taking the tunnel's local coordinate system X (along the tunnel axis), Y (horizontal), and Z (vertical upward) as an example: The inclination angle of a structural surface can be determined by the angle between the unit normal vector and the vertical direction, for example, it can be determined by... calculate; The dip of the structural plane can be taken as the azimuth angle of the projection of the unit normal vector onto the horizontal plane, for example, it can be calculated as follows: Calculate and convert to The direction is the azimuth angle orthogonal to the dip, which can be calculated according to... and normalized to This yields the orientation, dip, and dip angle parameters of each structural plane. Furthermore, all structural surfaces can be clustered according to the similarity of their normal vector directions to obtain the number of structural surface groups. Within the same structural surface group, the spacing between structural surfaces can be obtained statistically from the normal distance between structural surface planes that are approximately parallel to each other. The extension scale of a structural surface can be characterized by the length of the principal axis of the set of interior points of the structural surface in the plane. For example, principal component analysis can be performed on the two-dimensional coordinates of the set of interior points projected onto the fitted plane to obtain the length of the major axis and the length of the minor axis, and the length of the major axis can be used as an index of the extension scale of the structural surface.

[0048] like Figure 1 As shown, in step S400, a set of virtual survey lines is laid out in the reconstructed three-dimensional structural surface model, the RQD value of each virtual survey line is calculated, and the RQD statistical summary value of multiple virtual survey lines is obtained.

[0049] In this embodiment, a set of virtual survey lines is laid out in the reconstructed three-dimensional structural surface model. Preferably, the virtual survey lines are spaced at preset intervals within the projection area of ​​the tunnel face. The survey lines are evenly distributed and include at least three directions: along the tunnel axis, radial, and circumferential; each survey line... It can be expressed by the three-dimensional linear parametric equation as follows: , As the starting point of the survey line, It is a unit direction vector. .

[0050] When the confidence level of structural surface identification is low or the statistical results of the survey lines do not meet the convergence criteria, the number of survey lines should be increased or the spacing reduced according to the adaptive densification strategy. Furthermore, priority can be given to densely deploying in areas with dense or uncertain structural surfaces, and the deployment strategy is subject to approval. Figure 5 .

[0051] For any survey line Calculate its relationship with each structural surface in the structural surface model. The set of intersection points, and along the survey line parameters Sort the intersection points to obtain the intersection point sequence. The distance between adjacent intersection points is defined as the equivalent core segment length of the survey line, i.e. (6) In equation (6), Indicates the first The equivalent core segment length of the kth segment on the test line; and They represent survey lines respectively. The k-th and (k+1)-th intersection points obtained after intersecting with the structural surface model (sorted in ascending order by the survey line parameter s); Represents the Euclidean norm (distance between two points); ,in For surveying lines Number of intersection points.

[0052] And As a survey line The corresponding "equivalent core segment length set". The physical meaning of the equivalent core segment length is: the equivalent core segment length obtained by dividing the virtual survey line under the cutting action of the structural surface, used to simulate the statistical process of the length of the complete core segment in the drill core. Furthermore, it will satisfy... The segment lengths are summed and divided by the total length of the survey line. , obtain the survey line of .

[0053]

[0054] In equation (7), Indicates the first strip survey line Core recovery rate index; For surveying lines The total length; The effective segment threshold (e.g., 0.1 m); where This is an indicator function that takes the value 1 if the condition is true and 0 otherwise. Indicates the survey line Summing the lengths of segments formed by all adjacent intersection points, where only the segments satisfying the condition are summed. The segment length is included in the statistics.

[0055] Furthermore, assuming a total of n survey lines are laid out, the overall RQD can be taken as the statistical summary value of the results from multiple survey lines:

[0056] In equation (8), RQD represents the statistical summary value of the results of multiple survey lines; n is the number of survey lines; For the first The total length of the survey lines is used as a weighting factor; For the first The core recovery rate index is calculated from each survey line. Equation (8) characterizes the core recovery rate index for each survey line according to its length. For the overall RQD of weighted average, when the total length of each survey line is the same or no weighting is required, equation (8) can be replaced with an arithmetic mean. .

[0057] For multiple survey lines The stability index was calculated as follows: Let the first... The core recovery rate corresponding to each survey line is Its mean ,variance Or calculate the confidence interval width. ,by or With threshold , Comparison is used as a convergence criterion.

[0058] When the overall RQD is summarized using the arithmetic mean, we have When the overall RQD is summed using length-weighted aggregation, the RQD is calculated according to equation (8), and , For measurement purposes only Stability.

[0059]

[0060] (10) When stability is insufficient, the number / density of survey lines is adaptively increased in areas with high fracture density or large RQD fluctuations to bring the RQD results closer together.

[0061] like Figure 1 As shown, in step S500, the stability of the surrounding rock is evaluated based on the obtained RQD statistical summary value and structural surface attitude parameter data.

[0062] RQD and structural surface attitude statistical parameters (e.g., number of structural surface groups) obtained from multi-line statistics. Average spacing (Including dominant dip angle / dip distribution, etc.), outputting the surrounding rock stability evaluation results.

[0063] Preferably, the surrounding rock evaluation result can be expressed as the surrounding rock grade. Its grade set can be set according to the current tunnel surrounding rock classification system. (or equivalent set of levels).

[0064] Surrounding rock grade The determination can be made using a preset rule function:

[0065] The evaluation index vector It may include at least RQD is composed of multiple survey lines. The overall RQD obtained by statistical summarization (e.g., weighted by survey line length or arithmetic mean). The number of structural plane groups. The average spacing of the structural planes in the same group is used to determine the dominant attitude index, which can be obtained by statistically analyzing the normal vectors of each structural plane (e.g., dominant dip / angle and its dispersion). It can be a threshold segmentation rule, a rule tree, or a trained mapping model, and is pre-installed in the system in the form of "surrounding rock grade determination rule (or determination table)" for table lookup calculation.

[0066] Surrounding rock grade Its function is to serve as an entry evaluation quantity for subsequent support parameter recommendations, and to establish a traceable correspondence between the results of face identification and three-dimensional structural surface inversion and the output of support design parameters.

[0067] When RMR components are needed and available, an RMR score can be further calculated as an optional output to achieve compatibility and comparison with commonly used rock mass evaluation systems. Specifically, the RMR score can be obtained by summing the scores of each component according to a preset "RMR Scoring Table / Scoring Rules," denoted as […].

[0068] in For the first Individual section scores, This represents the number of items. The method of obtaining it is: to take the first Input indicators for each item The sub-item scores are obtained by interval mapping according to a preset scoring table, or by calculating the sub-item scores according to a preset scoring rule function. (12) in, For the first The scoring mapping function for each sub-indicator is used to map the sub-indicators to the sub-indicators. Mapped to corresponding sub-item scores , Implemented using a pre-set scoring table or segmented mapping rules: The range of values ​​is divided into several intervals, each corresponding to a specific score. Output the corresponding score when the value falls into the corresponding interval; For the first The input indicators for each item include the RQD statistical summary value, structural plane attitude parameters, and related indicators. Among the items, RQD, number / spacing / attitude of structural plane groups, etc., can be automatically obtained by this method; other items (such as groundwater conditions, structural plane condition, or rock strength, etc.) can be entered on-site or set as default values. The role of RMR scoring is to serve as the surrounding rock grade. The optional control quantity is used to improve the consistency and portability of the evaluation when there is sub-item data, and can be used to match or verify the correspondence with the RMR interval in the "Recommended Support Parameter Table".

[0069] Based on surrounding rock grade (and optional RMR range), the recommended range of support parameters is output through a preset "Recommended Support Parameter Table / Mapping Table". Support parameters must include at least: shotcrete thickness. Anchor bolt length Anchor spacing Spacing of steel arches (or discrete parameters such as steel arch frame model); among which the recommended table is used to establish { The mapping relationship between the RMR interval and the support parameter interval can be established, and two sets of recommended ranges, one for economical and one for conservative purposes, can be given respectively; for example, given... After (or RMR range), look up the table to get [ , ]、[ , ]、[ , ]、[ , The range of parameters such as ] is used as the candidate range for support parameters.

[0070] like Figure 1 As shown, in step S600, based on the pre-constructed support parameter recommendation table, the support parameter recommendation interval that matches the surrounding rock stability evaluation result is obtained, and the optimal support parameter combination is obtained by optimization solution using the support parameter recommendation interval as a candidate.

[0071] In this embodiment, to balance safety and economy, after obtaining the above-mentioned recommended range, the selection of support parameters can also be expressed as a constrained optimization problem: let the combination of support parameters... Under the constraint of the recommended interval (in Given the parameter ranges provided in the recommendation table, minimize the cost / schedule objective function. And it satisfies deformation and deformation rate constraints. Its optimization model can be expressed as: (13) in and It can be calculated by empirical prediction models, numerical simulation models, or historical data regression models. Therefore, the recommended interval is used to limit the feasible region and give the initial solution / candidate set, while constrained optimization is used to further give the optimal or near-optimal combination of support parameters and its output within the feasible region.

[0072] In equation (10), The support parameter combination vector is preferably represented as follows: ,in For the thickness of the sprayed concrete, The length of the anchor bolt. The anchor spacing is... The spacing of the steel arch frames (or the steel arch frame model is represented as a discrete variable and included) ). The feasible region, formed by the recommended intervals of support parameters given in the recommendation table, is used to limit the candidate range and provide an initial solution / candidate set; for example... It can be obtained by [ , ]、[ , ]、[ , ]、[ , The interval constraints are defined together. The objective function is cost / time. In support parameters The predicted deformation index is as follows. For predicting deformation rate indicators; and This corresponds to the allowed threshold. and It can be calculated by empirical prediction models, numerical simulation models, or historical data regression models. Therefore, the recommended interval is used to limit the feasible region and give the initial solution / candidate set, while constraint optimization is used to further obtain the optimal or near-optimal combination of support parameters that satisfy deformation constraints within the feasible region.

[0073] According to stability index ( or and identification confidence level It outputs two sets of support parameter combinations: an economical solution and a conservative solution. The preferred generation method is "two-step solution": Economic solution: When stability satisfies convergence (e.g.) / )and At that time, the feasible region given by the recommendation table With allowable threshold , To constrain the solution, we solve equation (10) to obtain the economic solution. (Under the condition of satisfying safety constraints) Minimum).

[0074] Conservative approach: When stability is insufficient or At that time, a conservative constraint set is obtained by "tightening the feasible region + increasing the safety margin". With stricter thresholds Then, solving equation (10) yields the conservative solution. The preferred tightening rule is to increase the lower / upper limit of the load-bearing capacity parameter and decrease the upper limit of the spacing parameter, for example, taking... Shift towards the upper boundary of the interval Shift towards the upper boundary of the interval and Offset towards the lower boundary of the interval; simultaneously set To increase safety margins.

[0075] Example explanation: If the recommended range is [ , ]、 [ , ]、 [ , ]、 [ , If so, then it is acceptable. For the above intervals, a conservative subset (e.g., close to) , , , (on one side), and and Solve equation (10) separately to obtain two sets of output parameter combinations. and .

[0076] Figure 5 The results of processing data from a certain working face using this method are shown: the attitude parameters of the structural surface area are obtained by back-projection fitting, and the RQD results tend to be stable after adaptive virtual survey line densification.

[0077] Traditional manual surveying: Due to the influence of the survey line's location and direction, RQD results fluctuate greatly, and it is difficult to quickly generate support parameter recommendations.

[0078] The method of this invention is to trigger the densification of the survey line by assessing the stability of the survey line, so that the RQD fluctuations converge, and output the recommended ranges for shotcrete, anchor bolts and steel arches based on the surrounding rock evaluation results (such as RMR range or surrounding rock grade), so as to provide a basis for rapid on-site decision-making.

[0079] Corresponding to the aforementioned method for determining tunnel face support parameters based on RGB-D fusion, this invention also discloses a system for determining tunnel face support parameters based on RGB-D fusion, which specifically includes: The data acquisition and alignment module is used to simultaneously acquire RGB images and 3D point cloud data of the tunnel face, and generate a depth map D that is pixel-level aligned with the RGB image, establishing the correspondence between the RGB image, the depth map D, and the 3D point cloud in the same coordinate system. The structural surface recognition module is used to input RGB images and depth maps into a pre-trained RGB-D dual-stream feature fusion network and output a semantic mask and confidence map of the structural surface region. The structural plane fitting module is used to back-project the pixel set corresponding to the semantic mask of the structural plane region back into the point cloud space to obtain the three-dimensional point set of the structural plane, perform planar robust fitting processing on the three-dimensional point set of the structural plane to obtain the structural plane fitting plane, and obtain the structural plane attitude parameter data based on the fitting plane. The RQD value statistics module is used to lay out a set of virtual survey lines in the reconstructed 3D structural surface model, calculate the RQD value of each virtual survey line, and obtain the RQD statistical summary value of multiple virtual survey lines. The stability evaluation module is used to evaluate the stability of the surrounding rock based on the obtained RQD statistical summary value and structural surface attitude parameter data. The support parameter recommendation module is used to obtain the recommended range of support parameters that matches the surrounding rock stability evaluation results based on the pre-built support parameter recommendation table, and to obtain the optimal combination of support parameters by using the recommended range of support parameters as candidates through optimization solution.

[0080] It should be noted that for a detailed description of the tunnel face support parameter determination system based on RGB-D fusion provided in the embodiments of the present invention, please refer to the relevant description of the tunnel face support parameter determination method based on RGB-D fusion provided in the embodiments of the present invention, which will not be repeated here.

[0081] In addition, embodiments of the present invention also provide an electronic device, the device comprising: a processor and a memory; the memory being used to store one or more program instructions; the processor being used to execute one or more program instructions to perform the steps of a method for determining tunnel face support parameters based on RGB-D fusion as described in any of the preceding embodiments.

[0082] It should be noted that for a detailed description of the electronic device provided in the embodiments of the present invention, please refer to the relevant description of the method for determining tunnel face support parameters based on RGB-D fusion provided in the embodiments of this application, which will not be repeated here.

[0083] In addition, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for determining tunnel face support parameters based on RGB-D fusion as described in any of the preceding claims.

[0084] It should be noted that for a detailed description of the computer-readable storage medium provided in the embodiments of the present invention, please refer to the relevant description of the method for determining tunnel face support parameters based on RGB-D fusion provided in the embodiments of this application, which will not be repeated here.

[0085] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the principles of this invention.

Claims

1. A method for determining tunnel face support parameters based on RGB-D fusion, characterized in that, The method includes: Simultaneously acquire RGB images and 3D point cloud data of the tunnel face, and generate a depth map D that is pixel-aligned with the RGB image, establishing the correspondence between the RGB image, depth map D, and 3D point cloud in the same coordinate system; The RGB image and depth map are input into a pre-trained RGB-D dual-stream feature fusion network, which outputs a semantic mask and confidence map of the structural planar region. The pixel set corresponding to the semantic mask of the structural surface is projected back into the point cloud space to obtain the three-dimensional point set of the structural surface. The three-dimensional point set of the structural surface is subjected to planar robust fitting processing to obtain the structural surface fitting plane. The structural surface attitude parameter data is obtained based on the fitting plane. In the reconstructed three-dimensional structural surface model, a set of virtual survey lines is laid out, the RQD value of each virtual survey line is calculated, and the RQD statistical summary value of multiple virtual survey lines is obtained. Based on the obtained RQD statistical summary value and structural surface attitude parameter data, the stability of the surrounding rock is evaluated. Based on a pre-constructed support parameter recommendation table, a support parameter recommendation interval matching the surrounding rock stability evaluation results is obtained. Using the support parameter recommendation interval as candidates, the optimal support parameter combination is obtained through optimization.

2. The method for determining tunnel face support parameters based on RGB-D fusion as described in claim 1, characterized in that, Constructing an RGB-D dual-stream feature fusion network specifically includes: RGB encoding branch Used for input RGB images Extracting multi-scale texture semantic features ; Deep coding branches , used for the input depth map Extracting multi-scale geometric features ; Cross-modal interaction module Used to apply at least one scale layer and Alignment and fusion are performed to obtain fused features. ; Decoding branch It is used to upsample the fused features and fuse them with the encoded features in a skip connection to obtain the decoded features. ; Output head , for use based Generate semantic masks for surface-shaped regions. ; , for use based Generate confidence plot ,in It characterizes the degree of confidence that a pixel belongs to a structural plane region.

3. The method for determining tunnel face support parameters based on RGB-D fusion as described in claim 1, characterized in that, Training an RGB-D two-stream feature fusion network includes: Multiple sets of RGB images and pixel-level aligned depth maps were acquired at the tunnel face. The true value mask for the planar region of the structure is annotated manually or semi-automatically. to form training samples ; After data augmentation, the samples are input into an RGB-D dual-stream feature fusion network for training. The network parameters are iteratively updated through backpropagation until the validation set index converges, resulting in an RGB-D dual-stream feature fusion network for structural surface recognition. During the online inference phase, semantic masks of the structural planar regions output by the RGB-D dual-stream feature fusion network are applied. Post-processing is performed, including thresholding, connected component filtering, hole filling, and small region removal, to obtain a set of structural surface regions.

4. The method for determining tunnel face support parameters based on RGB-D fusion as described in claim 1, characterized in that, The pixel set corresponding to the semantic mask of the structural surface region is projected back into the point cloud space to obtain the three-dimensional point set of the structural surface. The three-dimensional point set of the structural surface is then subjected to robust planar fitting to obtain the structural surface fitting plane. Based on the fitting plane, the attitude parameter data of the structural surface are obtained, specifically including: The obtained semantic mask of the structural surface region Back to point cloud Obtain the set of structural surface points and to Divide into several subsets ; For each The structural plane is obtained by performing a robust planar fitting using RANSAC or weighted RANSAC. and its normal direction And extract the structural surface boundary as a finite patch based on the point set projection; thus, The components are combined to form a three-dimensional structural surface model, which is used for subsequent intersection determination of virtual survey lines and structural surfaces; The orientation, dip, and dip angle of the structural surfaces are calculated based on the fitted plane normal vectors, and the number of structural surface groups, the spacing between structural surfaces, and the extension scale parameters are statistically analyzed.

5. The method for determining tunnel face support parameters based on RGB-D fusion as described in claim 1, characterized in that, In the reconstructed 3D structural surface model, a set of virtual survey lines is laid out, and the RQD value of each virtual survey line is calculated, specifically including: Virtual survey lines are laid out at preset intervals within the projection area of ​​the tunnel face. The tunnel should be evenly distributed and include at least three directions: along the tunnel axis, radial, and circumferential. Each measuring line Expressed using the three-dimensional linear parametric equations as follows: , As the starting point of the survey line, Unit direction vector, survey line parameters , This represents the total length of the survey line; For any survey line Calculate its relationship with each structural surface in the three-dimensional structural surface model. The set of intersection points, and based on the survey line parameters Sort the intersection points to obtain the intersection point sequence. The distance between adjacent intersection points is defined as the equivalent core segment length of the survey line, i.e.: In the formula, Indicates the first The equivalent core segment length of the kth segment on the test line; and They represent survey lines respectively. The kth and (k+1)th intersection points obtained after intersecting with the three-dimensional structural surface model are sorted in ascending order according to the survey line parameter s; Represents the Euclidean norm; ,in For surveying lines Number of intersection points; Will As a survey line The corresponding set of equivalent core segment lengths, the physical meaning of which is: the equivalent core segment length obtained by dividing the virtual survey line under the cutting action of the structural surface, used to simulate the statistical process of the length of the complete core segment in the drill core; Will satisfy The segment lengths are summed and divided by the total length of the survey line. , obtain the survey line of In the formula, Indicates the first strip survey line Core recovery rate index; For surveying lines The total length; The effective segment threshold; where This is an indicator function that takes the value 1 if the condition is true and 0 otherwise. Indicates the survey line Summing the lengths of segments formed by all adjacent intersection points, where only the segments satisfying the condition are summed. The segment length is included in the statistics; For multiple survey lines The calculation results are used to evaluate stability, and the stability evaluation indicators include variance. or confidence interval width Variance or confidence interval width With the corresponding threshold , The comparison is used as a convergence criterion. If the convergence condition is not met, the number of survey lines is increased or the spacing is decreased according to the adaptive densification strategy. In particular, the RQD results are made more convergent by prioritizing the densification of the RQD array in areas with high fracture density or large RQD fluctuations.

6. The method for determining tunnel face support parameters based on RGB-D fusion as described in claim 5, characterized in that, The statistical summary values ​​of RQD for multiple virtual survey lines were obtained, specifically including: If a total of n survey lines are set up, the overall RQD is obtained by taking the statistical summary value of the results of multiple survey lines and then calculating the weighted average or arithmetic average. The formula for calculating the weighted average is as follows: When the total length of each survey line is the same or no weighting is required, the arithmetic mean calculation formula is used: In the formula, RQD represents the statistical summary value of the results from multiple survey lines; n is the number of survey lines; For the first The total length of the survey lines is used as a weighting factor; For the first Core recovery rate index obtained from the measurement line.

7. The method for determining tunnel face support parameters based on RGB-D fusion as described in claim 1, characterized in that, Based on the obtained RQD statistical summary values ​​and structural plane attitude parameter data, the stability evaluation of the surrounding rock is carried out, specifically including: Determine the surrounding rock grade and RMR score, with the RMR score as an alternative. Among them, a preset rule function is used. Determining the grade of surrounding rock ,in The evaluation index vector includes the RQD statistical summary value, structural surface attitude parameter data, and related indicators; To utilize threshold segmentation rules, rule trees, or trained mapping models; The RMR score is obtained by summing the scores of each item according to the preset RMR scoring table / scoring rules, as shown in the following formula: in For the first Individual section scores, The number of items; The method of obtaining it is: to take the first Input indicators for each item The sub-item scores are obtained by interval mapping according to a preset scoring table, or by calculating the sub-item scores according to a preset scoring rule function, that is: in, For the first The scoring mapping function for each sub-indicator is used to map the sub-indicators to the sub-indicators. Mapped to corresponding sub-item scores , Implemented using a pre-set scoring table or segmented mapping rules: The range of values ​​is divided into several intervals, each corresponding to a specific score. Output the corresponding score when the value falls into the corresponding interval; For the first The input indicators include the RQD statistical summary value, structural surface attitude parameter data, and related indicators.

8. The method for determining tunnel face support parameters based on RGB-D fusion as described in claim 7, characterized in that, Based on a pre-constructed support parameter recommendation table, a recommended range of support parameters matching the surrounding rock stability evaluation results is obtained, specifically including: Based on the determined surrounding rock grade The system outputs recommended support parameter ranges based on the pre-built support parameter recommendation table, along with the range of alternative RMR scores. The support parameters include: shotcrete thickness. Anchor bolt length Anchor spacing Spacing of steel arches Or steel arch frame model; the recommended support parameter table is used to establish the surrounding rock grade. It can also map the RMR score range to the recommended range of support parameters and provide two sets of recommended ranges for support parameters: one for economical use and one for conservative use.

9. The method for determining tunnel face support parameters based on RGB-D fusion as described in claim 8, characterized in that, Using recommended ranges of support parameters as candidates, the optimal combination of support parameters is obtained through optimization, specifically including: The feasible region is formed by the recommended range of support parameters given in the support parameter recommendation table. An objective function is established to minimize cost / construction period, and deformation and deformation rate constraints are established. The optimal combination of support parameters that satisfies the deformation and deformation rate constraints is obtained by optimization within the feasible region.

10. A system for determining tunnel face support parameters based on RGB-D fusion, characterized in that, The system includes: The data acquisition and alignment module is used to simultaneously acquire RGB images and 3D point cloud data of the tunnel face, and generate a depth map D that is pixel-level aligned with the RGB image, establishing the correspondence between the RGB image, the depth map D, and the 3D point cloud in the same coordinate system. The structural surface recognition module is used to input RGB images and depth maps into a pre-trained RGB-D dual-stream feature fusion network and output a semantic mask and confidence map of the structural surface region. The structural plane fitting module is used to back-project the pixel set corresponding to the semantic mask of the structural plane region back into the point cloud space to obtain the three-dimensional point set of the structural plane, perform planar robust fitting processing on the three-dimensional point set of the structural plane to obtain the structural plane fitting plane, and obtain the structural plane attitude parameter data based on the fitting plane. The RQD value statistics module is used to lay out a set of virtual survey lines in the reconstructed 3D structural surface model, calculate the RQD value of each virtual survey line, and obtain the RQD statistical summary value of multiple virtual survey lines. The stability evaluation module is used to evaluate the stability of the surrounding rock based on the obtained RQD statistical summary value and structural surface attitude parameter data. The support parameter recommendation module is used to obtain the recommended range of support parameters that matches the surrounding rock stability evaluation results based on the pre-built support parameter recommendation table, and to obtain the optimal combination of support parameters by using the recommended range of support parameters as candidates through optimization solution.

Images