Methods, equipment, media, and products for 3D modeling of tunnels based on geological features

Abstract

This invention discloses a method, equipment, medium, and product for 3D tunnel modeling incorporating geological features, relating to the field of geographic information modeling technology. It includes: acquiring point cloud data of a tunnel area using vehicle-mounted radar and fixed-mount radar; determining the terrain complexity index of each sub-region within the tunnel area based on the point cloud data; determining the number of fitting layers and the type of fitting function for each sub-region based on the terrain complexity index; fitting the point cloud data of each sub-region according to the number of fitting layers and the type of fitting function to obtain a terrain model of the tunnel area; constructing a geological constraint field based on the point cloud data and geological attribute information of the tunnel area; and performing differentiated modeling for different regions and at different times based on the terrain model, the geological constraint field, and corresponding time information to obtain the 3D tunnel modeling result. This invention achieves a dual improvement in modeling accuracy and efficiency through differentiated fitting based on terrain complexity and differentiated modeling for different regions and at different times.

Description

Technical Field

[0001] This invention relates to the field of underground spatial geographic information modeling technology, and in particular to a method, equipment, medium and product for three-dimensional modeling of tunnels that incorporates geological features. Background Technology

[0002] Existing 3D modeling technologies for tunnels mainly generate 3D models through data acquisition, processing, and modeling processes, and have been initially applied in scenarios such as tunnel engineering and mining. For example, existing technologies propose a 3D modeling method for oblique tunnel portals, which achieves efficient modeling of the tunnel portal area by creating a model of the portal section, obtaining modeling parameters and contour lines, etc.

[0003] However, existing technologies mostly focus on modeling local areas and do not fully consider the spatial relationship between the overall structure of the tunnel and the surrounding environment in complex terrain. In scenarios with dramatic terrain undulations and complex geology, it is difficult to achieve high-precision fitting, and the model is prone to distortion. Furthermore, existing rapid modeling methods lack optimization measures for abrupt terrain changes and geological anomalies, resulting in insufficient accuracy. High-precision modeling methods suffer from redundant calculations, low efficiency, and difficulty in prioritizing the reconstruction of key tunnel structures. Summary of the Invention

[0004] This invention provides a method, equipment, medium, and product for three-dimensional tunnel modeling that incorporates geological features, in order to solve the problems of low accuracy and efficiency in tunnel modeling under complex terrain.

[0005] According to one aspect of the present invention, a method for three-dimensional modeling of tunnels incorporating geological features is provided, comprising:

[0006] Point cloud data of the tunnel area is acquired using vehicle-mounted radar and fixed-frame radar.

[0007] The terrain complexity index of each sub-region in the alleyway region is determined based on the point cloud data. The number of fitting layers and the type of fitting function for each sub-region are determined based on the terrain complexity index. The point cloud data of each sub-region is then fitted based on the number of fitting layers and the type of fitting function to obtain the terrain model of the alleyway region.

[0008] Obtain the geological attribute information of the tunnel area, and construct a geological constraint field based on the point cloud data of the tunnel area and the geological attribute information;

[0009] Based on the terrain model, the geological constraint field, and the corresponding time information, differentiated modeling is performed for different regions and at different times to obtain the three-dimensional modeling results of the tunnel.

[0010] According to another aspect of the present invention, a three-dimensional tunnel modeling apparatus incorporating geological features is provided, comprising:

[0011] The point cloud data acquisition module is used to acquire point cloud data of the alleyway area through vehicle-mounted radar and fixed-frame radar;

[0012] The terrain model construction module is used to determine the terrain complexity index of each sub-region in the alleyway region based on the point cloud data, determine the number of fitting layers and the type of fitting function for each sub-region based on the terrain complexity index, and fit the point cloud data of each sub-region based on the number of fitting layers and the type of fitting function to obtain the terrain model of the alleyway region.

[0013] The geological constraint field construction module is used to acquire geological attribute information of the tunnel area and construct a geological constraint field based on the point cloud data of the tunnel area and the geological attribute information.

[0014] The tunnel model generation module is used to perform differentiated modeling for different regions and different times based on the terrain model, the geological constraint field and the corresponding time information, so as to obtain the three-dimensional modeling results of the tunnel.

[0015] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0016] At least one processor; and

[0017] A memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the tunnel three-dimensional modeling method incorporating geological features as described in any embodiment of the present invention.

[0019] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the tunnel three-dimensional modeling method incorporating geological features as described in any embodiment of the present invention.

[0020] According to another aspect of this application, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the tunnel 3D modeling method incorporating geological features as described in any embodiment of this application.

[0021] The technical solutions of this invention improve the richness and accuracy of point cloud data by collecting point cloud data in multiple ways; achieve high-precision fitting of tunnels and complex terrain by performing regional differentiation fitting through terrain complexity; enhance the geological information expression of the model by integrating point cloud data and geological exploration data through multi-source data fusion; and achieve a dual improvement in modeling accuracy and efficiency through differentiated modeling of different regions and different times.

[0022] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a three-dimensional tunnel modeling method that incorporates geological features, according to an embodiment of the present invention.

[0025] Figure 2 This is a flowchart of another method for three-dimensional tunnel modeling that incorporates geological features, provided by an embodiment of the present invention.

[0026] Figure 3 This is a flowchart of another method for three-dimensional tunnel modeling that incorporates geological features, provided by an embodiment of the present invention.

[0027] Figure 4 This is a schematic diagram of a tunnel three-dimensional modeling device that incorporates geological features, according to an embodiment of the present invention.

[0028] Figure 5 This is a schematic diagram of the structure of an electronic device that implements the three-dimensional tunnel modeling method combining geological features according to embodiments of the present invention. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "candidate," "target," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0031] Figure 1 This invention provides a flowchart of a method for 3D tunnel modeling incorporating geological features. This embodiment is applicable to 3D modeling and optimization of tunnels in complex terrain. The method can be executed by a 3D tunnel modeling device incorporating geological features, which can be implemented in hardware and / or software and can be configured on a server. Figure 1 As shown, the method includes:

[0032] S110: Obtain point cloud data of the tunnel area through vehicle-mounted radar and fixed-frame radar.

[0033] Among them, vehicle-mounted radar refers to mobile devices that can deploy radar devices, while fixed-mount radar refers to fixed equipment that deploys radar devices. Specifically, vehicle-mounted radar collects point cloud data of the internal space of the alleyway, while fixed-mount radar collects point cloud data of the surrounding terrain of the alleyway area.

[0034] For example, the vehicle-mounted radar starts from the entrance of the alley and uses an S-shaped round-trip path to collect point cloud data inside the alley. At the same time, during the collection process, key structural areas such as intersections, emergency exits, and fault outcrops are marked, and a key collection scheme is used in key structural areas, such as 360-degree rotation collection. Fixed frame points are arranged in a triangle on the ground in the alley area, and the spacing between the fixed frame radars is determined according to the coverage range of the radar.

[0035] For example, the first step is to prepare for data acquisition. As a fundamental step in system implementation, data acquisition preparation mainly includes two aspects: on-site survey and route planning, and equipment debugging and calibration. On-site survey and route planning require a physical survey of the length, width, and height of the tunnels, as well as the slope of the surrounding terrain and the distribution of obstacles. Based on this, the S-shaped round-trip route for the LiDAR vehicle-mounted acquisition (2m lane spacing) and the triangular deployment of fixed stations (3-5 points, each covering a radius of 100m) are determined. Simultaneously, key structural areas such as intersections, emergency exits, and fault outcrops are marked, and targeted key acquisition plans are developed (e.g., 360-degree rotating acquisition at intersections). In equipment debugging and calibration, the ranging accuracy (±2cm at 10m) and point cloud output rate (1.536 million points / second) of the lidar vehicle need to be tested, and positioning calibration (planar accuracy ≤1cm) needs to be completed through static RTK. A customized data acquisition APP needs to be installed on the geological data terminal, and the communication stability between Bluetooth and lidar needs to be tested (time synchronization error ≤1ms). At the same time, the rebound hammer (intensity measurement error ±2MPa) and geological compass (strike / dip angle error ±1 degree) need to be calibrated. Core modules such as preprocessing, fitting, and fusion need to be deployed on the backend server, and the GPU parallel computing performance needs to be tested. In the data acquisition phase, the main tasks are to collect point clouds inside the tunnel and point clouds of the surrounding terrain. For vehicle-mounted data collection inside the tunnels, the LiDAR equipment is mounted on the roof of the engineering vehicle (1.8m high). After starting the onboard computer and LiDAR, the vehicle travels at a speed of 3-5km / h along the planned route. When reaching key structural areas such as intersections and emergency exits, the vehicle stops and performs a 360-degree rotation for data collection (5 degrees / s rotation speed, 72s collection time). Point cloud data is cached in real time to the onboard SSD in LAS 1.4 format, and data is backed up to the cloud every 5 minutes to ensure data security. For fixed-station data collection in the surrounding terrain, LiDAR is set up at preset points using tripods (2.5m high). After starting static RTK observation for 10 minutes to obtain the reference coordinates, data is collected for 30 minutes at each point, focusing on covering tunnel entrances and exits, mountain slopes, and fault outcrops to ensure a point cloud density of no less than 50 points / m².

[0036] This embodiment addresses the differences in data acquisition environments between the narrow spaces inside tunnels and the open terrain surrounding them by designing a dual-mode acquisition scheme of "vehicle-mounted mobile acquisition + fixed-station acquisition." It fully utilizes the reflectivity characteristics of LiDAR to ensure the completeness and reliability of the acquired data coverage. Specifically, in vehicle-mounted mobile mode, the point cloud density inside the tunnel can reach over 200 points / m², while in fixed-station mode, the point cloud density of the surrounding terrain is no less than 50 points / m². Dedicated synchronization technology ensures that the time synchronization error between devices is controlled within 1ms, and the coordinate consistency deviation does not exceed 2cm. The core value of this technology lies in replacing the redundant acquisition scheme of traditional "UAV + LiDAR," reducing equipment costs while avoiding the technical limitations of UAVs operating underground or in narrow spaces, thus meeting the acquisition needs of complex tunnel scenarios.

[0037] Optionally, after acquiring the point cloud data, the method further includes:

[0038] Noise points are removed from the point cloud data based on the point density and distance threshold of the local neighborhood of the point cloud to obtain denoised point cloud data. A hybrid smoothing algorithm combining Gaussian filtering and bilateral filtering is used to smooth the denoised point cloud data to obtain filtered point cloud data. A spatial segmentation box is constructed based on the tunnel design axis and a preset spatial range to remove points outside the tunnel from the filtered point cloud data to obtain the final point cloud data.

[0039] For example, based on the layered characteristics of reflection intensity in LiDAR point clouds (the reflection intensity of noisy points is typically below 20, while the reflection intensity of effective structures is between 80 and 120), this embodiment proposes a dual-criteria denoising algorithm of "improved Gaussian-bilateral hybrid filtering + reflection intensity threshold" to complete point cloud smoothing, efficiently removing noise while preserving structural details to the greatest extent. This technology effectively solves the problem that traditional single filtering algorithms are prone to accidentally deleting fine structures (such as cracks with a width of 0.1 mm) or leaving residual noise, achieving a three-dimensional balance of "denoising effect - detail preservation - processing efficiency".

[0040] S120. Determine the terrain complexity index of each sub-region in the tunnel area based on the point cloud data. Determine the number of fitting layers and the type of fitting function for each sub-region based on the terrain complexity index. Fit the point cloud data of each sub-region based on the number of fitting layers and the type of fitting function to obtain the terrain model of the tunnel area.

[0041] Specifically, the terrain features of each sub-region of the tunnel area are determined based on point cloud data. The terrain complexity index of a sub-region with complex terrain features is greater than that of a sub-region with simple terrain features. The terrain features are determined based on parameters such as elevation difference and slope of each sub-region. For example, a deep learning model is used to divide the tunnel area into sub-regions based on point cloud data and to determine the terrain complexity index of each sub-region. Therefore, this embodiment does not limit the scope or total number of sub-regions.

[0042] To address different terrain complexity indices, varying numbers of fitting layers and fitting function types are determined. For each sub-region, the point cloud data is fitted using the corresponding number of fitting layers and fitting function type to determine the terrain model for that sub-region. The terrain models from all sub-regions are then merged to obtain the overall terrain model. When determining the terrain model, an improved adaptive terrain fitting algorithm is employed based on terrain features, fitting the point cloud data region by region and dynamically adjusting parameters to adapt to complex terrain characteristics, generating an accurate terrain model. The number of fitting layers and fitting function type corresponding to different terrain complexity indices can be determined according to actual needs and are not restricted here; the number of fitting layers and fitting function type used for sub-regions corresponding to different terrain complexity indices are different.

[0043] In one feasible embodiment, the terrain complexity index of each sub-region in the alleyway region is determined based on point cloud data, including:

[0044] The elevation difference, slope, and curvature of each sub-region in the tunnel area are determined based on point cloud data; the terrain complexity index of each sub-region is determined based on the fusion results of elevation difference, slope, and curvature.

[0045] By processing the spatial coordinates of each point in the point cloud data, the elevation difference (i.e., the change in height in the vertical direction), slope (i.e., the degree of inclination of the ground surface, usually expressed as an angle or percentage), and curvature (i.e., the degree of drastic change in the direction of the centerline or floor of the tunnel in the horizontal or vertical plane, i.e., a quantitative indicator of the degree of curvature of its path) between different locations in the tunnel area are calculated as terrain features.

[0046] The terrain complexity index is determined based on the weighted fusion result of elevation difference, slope and curvature for each sub-region. For example, the elevation difference, slope and curvature are normalized, and the weighted sum of the normalized elevation difference, slope and curvature is used as the terrain complexity index. The weights of elevation difference, slope and curvature can be adjusted according to the actual needs of the scenario, and are not restricted here.

[0047] For example, a correspondence is pre-established between candidate terrain complexity indices, the number of fitting layers, and the type of fitting function. This embodiment constructs a terrain complexity index ranging from 3 to 9 levels by fusing three-dimensional data of elevation difference, slope, and curvature from lidar point clouds. Based on the terrain complexity index level, the corresponding number of fitting layers (3-8 layers) and the type of fitting function (including but not limited to linear functions, quadratic polynomial functions, and cubic spline functions) are automatically determined through matching relationships, achieving adaptive and accurate fitting of complex terrain. In terms of fitting accuracy, the fitting error for simple terrain (low terrain complexity index level) can be controlled within 0.03m, and the fitting error for high-complexity terrain (high terrain complexity index level) does not exceed 0.05m, with the generated terrain model achieving an integrity of over 95%. This embodiment solves the problem of large errors in existing fixed-layer fitting techniques for complex terrain areas such as steep slopes and fault zones, achieving dynamic adaptation between terrain complexity and fitting strategies, and improving the accuracy and reliability of the terrain model.

[0048] S130. Obtain geological attribute information of the tunnel area, and construct a geological constraint field based on the point cloud data and geological attribute information of the tunnel area.

[0049] In collecting geological attribute information, personnel carry geological data terminals to input data such as lithology, rock compressive strength, fault strike, and dip angle at the site and key structural areas. The data is automatically linked to the coordinates and timestamps of the lidar, and transmitted hourly to the backend server via a 5G network. For example, geological attribute information includes, but is not limited to, lithology, rock compressive strength, fault strike, and dip angle. The geological attribute information includes geological exploration data from various locations within the tunnel area. Based on spatial coordinate correlation, the geological attribute data from each location is fused with the point cloud data of the terrain model, achieving a binding of geological attributes and terrain model geometric information to generate a geological constraint field.

[0050] For example, the tunnel area is divided into grids, and the geological attribute information (lithology, faults, stability) of the tunnel area is converted into a gridded constraint field. Each grid node contains multi-dimensional geological attributes of the area corresponding to that grid node, realizing the spatial quantitative expression of geological information and solving the problem of fragmented geological information in existing technologies. Then, through spatial coordinate association, the geological attributes of the constraint field are bound to the geometric information of the terrain model to generate a geological constraint field with two layers of information, supporting the visualization query and analysis of geological attributes, and providing comprehensive support for emergency rescue decision-making.

[0051] By constructing a geological constraint field and integrating geological exploration data with point cloud data, a deep fusion of geometric and geological information can be achieved.

[0052] S140. Based on the terrain model, geological constraint field and corresponding time information, perform differentiated modeling for different regions and different times to obtain the three-dimensional modeling results of the tunnel.

[0053] Feature extraction is performed on the terrain model to determine the key structures in the tunnel area. For example, a convolutional neural network is used to extract features from the fused terrain model that integrates geological attributes and geometric information, effectively extracting low-level (texture) and high-level (structure) features of the tunnel. Based on the extracted features, the location information of key structures in the tunnel area is identified. Key structures refer to intersections and supports in the tunnel area. Based on the terrain model, geological constraint field, the time information of each data acquisition, and the location of key structure identification, differentiated modeling is performed for different key structure areas and different times in different areas to generate a 3D tunnel model. The aforementioned convolutional neural network can be any existing CNN network, or a 3D-CNN + dual attention model, which will be described in detail later and will not be elaborated here.

[0054] For example, the modeling priority of different locations in the tunnel area is determined based on the location information of the key structures. Different modeling priorities correspond to different modeling parameters, so as to adaptively reconstruct different locations in the tunnel area according to the corresponding modeling parameters to obtain the three-dimensional model of the tunnel. Modeling is performed separately based on the data collected at different time points corresponding to the same location to obtain the three-dimensional model of the tunnel at different times. The time dimension collection sequence information is retained. The final three-dimensional modeling result of the tunnel is obtained by combining the differentiated modeling results of different locations with the differentiated modeling results of different times.

[0055] The technical solution of this embodiment improves the richness and accuracy of point cloud data by collecting point cloud data in multiple ways; it achieves high-precision fitting of tunnels and complex terrain by performing regional differentiation fitting through terrain complexity; it enhances the geological information expression of the model by integrating point cloud data and geological exploration data through multi-source data fusion; and it achieves a dual improvement in modeling accuracy and efficiency through differentiated modeling of different regions and different times.

[0056] Figure 2 This is a flowchart illustrating another method for three-dimensional tunnel modeling incorporating geological features, provided by an embodiment of the present invention. This embodiment further refines the process of constructing the geological constraint field in the above embodiments. Figure 2 As shown, the method includes:

[0057] S210: Point cloud data of the tunnel area is acquired through vehicle-mounted radar and fixed-frame radar.

[0058] S220. Determine the terrain complexity index of each sub-region in the tunnel area based on the point cloud data. Determine the number of fitting layers and the type of fitting function for each sub-region based on the terrain complexity index. Fit the point cloud data of each sub-region based on the number of fitting layers and the type of fitting function to obtain the terrain model of the tunnel area.

[0059] S230. Obtain geological attribute measurement information of multiple discrete sample points in the tunnel area.

[0060] Because geological attribute information is collected by personnel carrying geological data terminals, who input data such as lithology, rock compressive strength, fault strike, and dip angle at the sampling sites and key structural areas, accurate geological attribute measurement information can be obtained at each sampling point. These sampling points are discrete sample points. However, since the number of manually collected discrete sample points is limited and cannot reflect the overall continuous geological attribute information of the tunnel area, it is necessary to estimate the geological attribute information corresponding to other uncollected points in the tunnel area.

[0061] S240. Determine multiple discrete sample points corresponding to the point to be interpolated, and determine the weight of each discrete sample point relative to the point to be interpolated based on the distance between each discrete sample point and the point to be interpolated.

[0062] In this system, the interpolation point refers to a point that has not been actually measured. Discrete sample points whose distance to the interpolation point is less than a threshold are assigned to the interpolation point. The weight of the corresponding discrete sample point relative to the interpolation point is determined based on the distance between the corresponding discrete sample point and the interpolation point. Distance and weight are inversely correlated; that is, the greater the distance between the discrete sample point and the interpolation point, the smaller the corresponding weight. For example, if the distance between the first and second discrete sample points and the interpolation point is less than the distance threshold, then the first and second discrete sample points are assigned to the interpolation point. The distance between the first and second discrete sample points is designated as the first distance, and the distance between the second and second discrete sample points is designated as the second distance. The weight of the first discrete sample point relative to the interpolation point is the first weight, and the weight of the second discrete sample point relative to the interpolation point is the second weight. If the first distance is greater than the second distance, then the first weight is less than the second weight. The specific weight values ​​can be adjusted according to actual circumstances and are not limited here.

[0063] In one feasible embodiment, S240 includes:

[0064] Determine the distance between the target discrete sample point and the interpolation point, and determine the distance-matched sample point pair from multiple discrete sample points based on the distance;

[0065] The weight of the target discrete sample point relative to the interpolation point is determined by the ratio of the sum of squared differences in the geological attribute measurement information corresponding to the discrete sample points included in all sample point pairs to the number of sample point pairs.

[0066] Here, the target discrete sample point is any sample point among the discrete sample points. First, the distance between two discrete sample points is determined to match the distance between the target discrete sample point and the interpolation point. These two discrete sample points are then identified as a sample point pair corresponding to the target discrete sample point. For example, this match can be two distances that are the same, or distances whose difference is within a preset difference range. For instance, the distance between the target discrete sample point and the interpolation point is a first distance. Multiple sample point pairs exist among all discrete sample points that correspond to the target discrete sample point, and the difference between the distance between the two discrete sample points in each sample point pair and the first distance is within a preset difference range.

[0067] Determine the squared difference of the geological attribute measurement information of the discrete sample points in each sample pair corresponding to the target discrete sample point. The ratio of the sum of the squared differences of all sample pairs to twice the number of sample point pairs is used as the weight of the target discrete sample point relative to the interpolation point.

[0068] For example, the weights of the target discrete sample points are determined according to the following formula:

[0069] ;

[0070] in, The weight of the target discrete sample point is represented when the distance to the point to be interpolated is h (reflecting the attribute correlation between sample points); h represents the spatial distance between the two sample points. This represents the number of sample point pairs that match the distance h. and This represents the geological attribute measurement information of two discrete sample points in a sample point pair that matches the distance h. This geological attribute measurement information can be determined based on the geological attribute measurement information of the currently sought interpolation point.

[0071] In this embodiment, discrete geological attribute sample points are interpolated into a continuous initial geological constraint field, ensuring that each small spatial unit corresponds to a unique geological attribute value (such as lithology code and intensity value), laying the foundation for subsequent incremental calculations.

[0072] S250. Determine the geological attribute estimation information of the points to be interpolated based on the geological attribute measurement information and corresponding weights of each discrete sample point.

[0073] The geological attribute estimation information of the point to be interpolated is determined by weighting the geological attribute measurement information and corresponding weights of each discrete sample point. For example, for geometric data (such as elevation difference and slope) and geological attribute data (such as lithology and rock strength) from lidar point clouds, the Kriging interpolation algorithm is used to construct the geological attribute estimation information for the entire area of ​​the tunnel region. For example, the formula for determining the geological attribute estimation information is as follows:

[0074] ;

[0075] in: Indicates the point to be interpolated The estimated geological properties (such as rock strength); n represents the interpolation point. The corresponding number of discrete sample points; Represents the weight coefficient of the i-th discrete sample point (satisfying) The weights are determined in S240. To determine, for example, directly As or to After appropriate processing, it is used as a weight. (e.g., multiplying by the corresponding preset value, etc.) This represents the geological property measurement information (such as rock strength obtained through geological boreholes) of the i-th discrete sample point.

[0076] S260. Based on the geological attribute measurement information of multiple discrete sample points in the tunnel area, the geological attribute estimation information of multiple interpolation points, and point cloud data, determine the geological constraint field.

[0077] Based on the geological attribute measurement information of multiple discrete sample points in the tunnel area and the geological attribute estimation information of multiple interpolation points, the geological attribute information of each location point in the tunnel area is determined. The tunnel area is then divided into grids. The geological constraint value of each grid node is determined based on the geological attribute information of all location points corresponding to each grid node. The point cloud constraint value of the grid node is also determined based on the point cloud data corresponding to the grid node. Finally, the final constraint value of the grid node is determined based on the geological constraint value and the point cloud constraint value. The final constraint values ​​of all grid nodes constitute the geological constraint field.

[0078] In one feasible embodiment, S260 includes:

[0079] An initial geological constraint field is constructed based on the geological attribute measurement information of multiple discrete sample points in the tunnel area and the geological attribute estimation information of multiple interpolation points. The initial geological constraint field includes the grid division results of the tunnel area, and the constraint value of each grid node is determined based on the geological attribute information of the included discrete sample points and interpolation points.

[0080] Spatial matching is performed between the point cloud data and the initial geological constraint field to determine the grid point cloud data corresponding to each grid node;

[0081] Features of grid point cloud data are extracted based on the PCA algorithm, and the constraint field increment of grid nodes is determined based on the features.

[0082] The gradient descent method is used to iteratively optimize the initial geological constraint field based on the constraint field increment to obtain the final geological constraint field.

[0083] Based on the geological attribute measurement information of multiple discrete sample points in the tunnel area and the geological attribute estimation information of multiple interpolation points, the geological attribute information of each location point in the tunnel area is determined. The tunnel area is then divided into grids. Based on the geological attribute information of all location points corresponding to each grid node, the initial constraint value of that grid node is determined, thus obtaining the initial geological constraint field.

[0084] For example, the tunnel area is divided into grids to obtain multiple grid nodes. The size of the grid nodes can be determined according to the actual needs of the scenario and is not limited here. Optionally, the size of the grid nodes corresponding to different terrain features is different. The constraint value of each grid node is determined based on the geological attribute measurement information of multiple discrete sample points and the geological attribute estimation information of multiple interpolation points. The geological exploration data (lithology, faults, stability) is converted into a gridded constraint field. Each grid node contains multi-dimensional geological attributes, realizing the spatial quantitative expression of geological information and solving the problem of fragmented geological information in existing technologies.

[0085] For each point cloud data point, the nearest neighbor interpolation method is used to match and determine the corresponding grid node in the initial geological constraint field, obtaining the grid point cloud data corresponding to each grid node. Features of the grid point cloud data are extracted based on the PCA algorithm, including microscopic features such as cracks and protrusions in the point cloud. The constraint field increment of the grid node is determined based on the difference between the features of the grid point cloud data and the constraint value corresponding to the grid node. For example, the actual terrain features (such as surface roughness and slope) reflected by the grid point cloud data are compared with... Whether the geological properties (such as stability level) of the node match, for example, if it is determined from the point cloud data that there are surface cracks in the area (identified by abrupt changes in local curvature of the point cloud), and the initial geological constraint field... If the constraint value is "stable", then the stability level of the node is adjusted to "relatively stable".

[0086] The gradient descent method is used to iteratively optimize the constraint field. The iterative formula is as follows:

[0087] ;

[0088] Where k is the number of iterations (preset to 5~10 iterations). The adjustment coefficient (dynamically adjusted based on error) balances the influence of initial geological data and actual point cloud data, avoiding bias caused by a single data source; if the iteration error decreases rapidly, Take the larger value; if the error fluctuates, (Take the smaller value). After each iteration, calculate the matching error between the constraint field and the point cloud data and geological exploration data (such as hardness value error, stability level matching rate). When the error is less than the error threshold and the error change is less than the error threshold for multiple consecutive iterations, stop the iteration and generate the final geological constraint field. The specific value of the error threshold can be adjusted according to the actual needs of the scenario, and is not restricted here.

[0089] For example, the gradient descent method is used to iteratively optimize the constraint field, with the objective function error E being less than the error threshold as the termination condition. The expression for the objective function error is: Where: E represents the iteration error of the global constraint field (reflecting the degree of deviation between the optimized constraint field and the "initial field + incremental field"); M represents the total number of grid nodes to be evaluated in the scene (e.g., in a 1km alleyway area scene). Covering all node); This represents the spatial domain of the entire modeled tunnel area scene. Indicates the points to be evaluated after iterative optimization. Geological attribute values; Indicates the points to be evaluated The theoretical geological property values ​​(superposition of initial field and incremental field).

[0090] Gradient descent iterative update formula: ;in, This represents the point to be evaluated after the (t+1)th iteration. Geological attribute values; This represents the point to be evaluated after the t-th iteration. Geological attribute values; Indicates the learning rate (in this scheme) To control the iteration step size and avoid error bounce caused by too rapid convergence, the specific value can be adjusted according to the actual needs of the scenario, and is not limited here. The objective function error E represents the pair of The partial derivatives (reflecting the trend of error as attribute values ​​change).

[0091] By calculating the error E for each iteration, if E is greater than the error threshold, the attribute values ​​are updated and the iteration continues until E is less than or equal to the error threshold, ensuring that the final output constraint field accuracy meets engineering requirements.

[0092] In one feasible embodiment, features of the grid point cloud data are extracted based on the PCA algorithm, and the constraint field increments of the grid nodes are determined based on the features, including:

[0093] Based on the PCA algorithm, features of grid point cloud data are extracted to obtain the covariance matrix corresponding to the grid point cloud data;

[0094] Determine the largest eigenvalue of the covariance matrix, and then determine the constraint field increment based on the largest eigenvalue and the constraint values ​​of the grid nodes.

[0095] The formula for calculating the eigenvalues ​​of principal components in PCA is expressed as follows:

[0096] ;

[0097] in, The covariance matrix of the gridded point cloud data (m point cloud samples, each sample contains...) Three-dimensional coordinates, therefore for (matrix); m represents the number of samples within the point cloud feature extraction window, i.e., the number of point clouds included in the grid node; Represents the three-dimensional coordinate vector of the j-th point cloud sample; This represents the mean vector of coordinates of m point cloud samples; Represents the covariance matrix The eigenvalues ​​(reflecting the variance contribution of the principal components); Represents eigenvalues The corresponding eigenvectors (reflecting the direction of the principal components, used to identify the fracture orientation).

[0098] The adjustment result through the constraint field increment is determined according to the following formula:

[0099] ;

[0100] in, Indicates the point to be adjusted The final geological property values ​​(such as adjusted rock strength); Indicates the point to be adjusted The initial geological attribute values; k represents the characteristic influence coefficient (in this scheme). Through engineering pilot calibration, it was ensured that the intensity reduction in the fractured zone conformed to actual geological patterns. Covariance matrix The maximum eigenvalue (reflects the dispersion of the point cloud sample; the larger the value, the more developed the cracks, and the greater the intensity needs to be reduced).

[0101] This embodiment extracts microscopic features such as cracks and protrusions from the point cloud based on the PCA algorithm, and dynamically adjusts geological properties according to feature differences to achieve the initial constraint field. To the incremental field Optimization improves the accuracy of the final geological constraint field.

[0102] S270. Based on the terrain model, geological constraint field and corresponding time information, differentiated modeling is performed for different regions and different times to obtain the three-dimensional modeling results of the tunnel.

[0103] The technical solution of this embodiment introduces a closed-loop logic of data input, initial constraint field construction, incremental calculation, and iterative optimization through three-dimensional geological constraint field iterative fusion technology. This enables deep binding of geometric data and geological attributes, improves the accuracy of geological constraint field construction, and thus improves the accuracy of tunnel three-dimensional modeling.

[0104] Figure 3 This is a flowchart illustrating another method for three-dimensional tunnel modeling incorporating geological features, provided by an embodiment of the present invention. This embodiment further refines the process of constructing the geological constraint field in the above embodiments. Figure 3 As shown, the method includes:

[0105] S310: Point cloud data of the tunnel area is acquired through vehicle-mounted radar and fixed-frame radar.

[0106] S320. Determine the terrain complexity index of each sub-region in the tunnel area based on the point cloud data. Determine the number of fitting layers and the type of fitting function for each sub-region based on the terrain complexity index. Fit the point cloud data of each sub-region based on the number of fitting layers and the type of fitting function to obtain the terrain model of the tunnel area.

[0107] S330. Obtain geological attribute information of the tunnel area, and construct a geological constraint field based on the point cloud data and geological attribute information of the tunnel area.

[0108] S340. Determine the priority of key structures in different sub-regions of the tunnel area based on the terrain model, and allocate Gaussian atom densities differently according to the priority of key structures.

[0109] Feature extraction from a terrain model is performed using a convolutional neural network (CNN), effectively extracting both low-level (texture) and high-level (structural) features of the alleyways. Based on the extracted features, the location and category information of key structures within the alleyway region are identified. Key structures refer to intersections and supports within the alleyway region. For example, a VNet network structure is used to extract features from the terrain model, outputting the location and type information of intersections and supports within the alleyway region. The priority of key structures in different sub-regions is determined based on the location and category information of the key structures. For example, priority is determined by the number of key structures included in the sub-region; the more key structures included, the higher the priority of the corresponding sub-region. Alternatively, a mapping relationship between the type of key structure and its corresponding priority is pre-determined, and then the priority of the corresponding key structure is determined based on this mapping relationship and the type information of the key structures included in each sub-region. The Gaussian atom density of different sub-regions is determined based on the differences in key structure priorities. The CNN can be any existing CNN network, or a 3D-CNN + dual attention model, which will be described in detail later and will not be elaborated here.

[0110] For example, based on the priority of critical structures (Level 1: intersections or emergency exits; Level 2: linings or pipelines; Level 3: general sections), Gaussian atomic densities are differentiated, and the formula for calculating Gaussian atomic densities is as follows: ;in, Gaussian atom density of the target sub-region (unit: atoms) ); Basic density value (in this scheme) (This is the default density for normal segments; its specific value can be adjusted according to the actual needs of the scenario, and is not limited here.) Priority coefficient (in this scheme) (obtained by reverse calculation through modeling accuracy). : Structural priority level (e.g., level 1) Level 2 Level 3 The specific values ​​can be adjusted according to the actual needs of the scenario, and are not limited here. For example, substituting the parameters, we can calculate: Level 1 structure (rounded to the nearest integer) ), 2-level structure 3-level structure (rounded to the nearest integer) This aligns with the differentiated density requirements in the plan, ensuring the accuracy of key structural details.

[0111] S350: Construct 3D projection spatial parameters from the terrain model, geological constraint field, and corresponding time information.

[0112] The terrain model provides the geometry of the surface or subsurface structure (such as elevation, slope, and fault strike) as a reference framework for spatial coordinates; the geological constraint field reflects the spatial distribution of different lithologies, physical properties, or tectonic attributes (such as porosity field, stress field, or lithofacies zoning) and is used to define the physical properties and boundary conditions of the modeling units; the corresponding temporal information records the timestamps of data acquisition or geological evolution to characterize dynamic changes (such as crack propagation and settlement deformation). Together, these three elements construct a 3D projection spatial parameter system that integrates the dimensions of "space-attribute-time". For example, in shale gas reservoir modeling, the initial topographic model describes the undulations of the formation around the horizontal well trajectory, the geological constraint field identifies the organic matter enrichment zone and the distribution of brittle minerals, and the temporal information corresponds to the fracture development status at different fracturing stages (such as t=0 min, t=30 min) in microseismic monitoring. By mapping these three types of information to a unified 3D projection space and combining multi-view acquisition (such as well logging or seismic data from different orientations) with time series correction, a high-precision three-dimensional geological model that reflects both static geological structure and dynamic evolution process can be generated, providing a reliable basis for subsequent numerical simulation and engineering decision-making.

[0113] Based on the terrain model representing spatial angles, the geological constraint field representing visual angles, and the corresponding time information as the time angle, we can define "space (x,y,z) + time t + visual angle". The seven-dimensional (7D) Gaussian parametric projection of the model represents the spatial parameters required for 3D modeling. The projection formula is as follows:

[0114] ;

[0115] in: : Projected 3D spatial coordinates (used for final modeling); : Original spatial coordinates in 7D Gaussian; t: Time dimension parameter (unit: s, reflecting the acquisition timing, such as...) This is the initial data acquisition time. (1 hour later) Rate of change of spatial coordinates over time (unit: (used to correct minor equipment displacements during the data acquisition process) In this solution ; Viewpoint correction function ( It is the azimuth angle. Angle of elevation, The roll angle is used to fuse multi-view acquisition biases in this solution. (This ensures that the perspective correction is within the allowable error range). The above formula projects the 7D parameters into 3D spatial coordinates, which not only preserves the temporal data acquisition information (such as the slight changes in the crack over time) but also integrates multi-view deviation correction to improve modeling accuracy.

[0116] Ultimately, this projection formula effectively reduces the 7D parameters to a 3D modeling space while preserving the evolutionary information implicit in the time series (such as the slight expansion of cracks over time). This ensures geometric consistency and improves the modeling accuracy and robustness under the fusion of multi-source heterogeneous data.

[0117] S360, based on the Gaussian splashing method, performs differentiated modeling of different sub-regions according to the Gaussian atomic density and 3D projection spatial parameters corresponding to each sub-region, and obtains the three-dimensional modeling results of the tunnel.

[0118] For critical structures such as emergency exits and intersections, assess the deviation between the modeled dimensions and the actual dimensions. The formula for calculating the dimensional error of critical structures is as follows:

[0119] ;

[0120] in, Dimensional error rate (required by this solution) The error corresponds to the 4m design dimension of the emergency exit. ); : Dimensional measurements of key structures after modeling (e.g., width of emergency exits, obtained through 3D model annotation tools); : Actual dimensional measurements of key structures (e.g., emergency exit width obtained through on-site measurements using a total station). The accuracy of the modeling is quantitatively evaluated. This is then fed back to the Gaussian atom density allocation stage (adjustment). Coefficients), re-optimize modeling parameters, and ensure that the errors in key structural dimensions meet engineering requirements ( ).

[0121] This embodiment introduces 7D Gaussian splashing technology. This differentiated modeling technology is based on 7-dimensional parameters (3D space + 1D time + 3D perspective), combined with the priority allocation of Gaussian atom density for key structures, achieving triple optimization of accuracy, efficiency, and real-time performance. Simultaneously, by incorporating the priority of key structures identified by a deep learning module (Level 1: intersections, emergency exits; Level 2: linings, pipelines; Level 3: ordinary tunnel sections), this technology can control the modeling time for a 1km-long tunnel to within 3.5 hours, achieving a real-time rendering frame rate of over 380FPS, with a modeling error not exceeding 0.03m. This technology effectively resolves the contradiction between "high accuracy (over 8 hours of processing time)" and "high real-time performance (frame rate below 30FPS)" inherent in traditional modeling techniques, providing technical support for interactive decision-making in engineering projects.

[0122] To address the shortcomings of existing technologies in adapting to complex terrain, the core objective of this application is to improve adaptation accuracy and clarify quantitative standards. By designing a "terrain feature-driven dynamic hierarchical fitting algorithm," based on terrain elevation difference thresholds and slope thresholds, it automatically divides the terrain into 3-8 fitting layers, such as slope. The terrain is divided into 3 layers for slopes of 25 degrees or more, and 8 layers for slopes of 25 degrees or more, achieving precise layered fitting. Simultaneously, a "Terrain Complexity Index (TCI)" is constructed, integrating three dimensions: slope, curvature, and elevation difference. For example, a TCI of 0.4 is used. Slope +0.3 Curvature +0.3 The formula for calculating elevation difference is used to dynamically adjust the fitting weight coefficients, ensuring that the model deviates from the actual terrain. (error In addition, a "tunnel-terrain spatial association model" was established, which uses spatial topology algorithms to associate the tunnel structure with surrounding mountains, water bodies, and other terrain features, ensuring the integrity of the model in complex terrain areas. .

[0123] To enhance geological information integration capabilities, this application sets clear information dimension objectives. A multi-source data fusion mechanism is constructed, establishing a correlation mapping model between point cloud data and geological exploration data. Spatial coordinate matching is performed using the WGS84 coordinate system to achieve lithological distribution (accuracy). ), geological boundary (error) Fault location (range error) Deep integration with the geometric model. A "3D geological constraint field iterative optimization algorithm" is designed. Based on initial geological data and combined with terrain details from point cloud feedback, the constraint field parameters are iteratively optimized using the gradient descent method to improve the matching rate between geological properties (hardness, stability) and the geometric model. A "Geological Information Visualization Module" was developed, supporting color coding (green for stable areas and red for risk areas) and attribute querying (clicking on a region of the model allows viewing lithology and hardness), reducing engineering risk assessment time from 4 hours to 30 minutes.

[0124] To address the imbalance between efficiency, accuracy, and real-time performance, this application establishes a 3D optimization goal and provides a solution. It introduces 7D Gaussian splashing technology, combined with deep learning, to achieve automatic identification of key structures (accuracy greater than 92%), replacing manual annotation and saving over one hour. A priority-driven differentiated modeling strategy is established, allocating high-density computing resources to high-priority areas and low-density resources to low-priority areas based on the key structure priorities output by deep learning. This reduces computational load while ensuring the high fidelity of key structural details. Ultimately, a "triple performance breakthrough" is achieved: modeling a 1-kilometer complex tunnel takes less than 3.5 hours, model error is less than 0.03m, and real-time rendering frame rate is greater than 380FPS, completely resolving the pain points of the imbalance among these three aspects in existing technologies.

[0125] To optimize adaptability for underground space scenarios, this application clearly defines adaptation goals and proposes corresponding measures. A "multi-format automatic conversion module" has been developed, supporting conversions to mainstream formats such as Engineering GIS (SHP), Project Management System (KML), and BIM platform (IFC), with a conversion time of less than 30 minutes and a data loss rate of less than 2%. A new "engineering simulation and deduction function" has been added, enabling construction path planning (accuracy greater than 90%) and equipment deployment simulation (automatically avoiding fault-affected areas) based on geological stability data, without relying on third-party software. A "real-time data interaction interface" has been constructed to connect with on-site monitoring equipment in underground space (such as stress sensors and cameras), enabling dynamic updates of modeling results and real-time data such as construction progress and environmental parameters, with a latency of less than 1 minute.

[0126] This application addresses the high-precision, high-efficiency, and dynamic requirements of 3D tunnel modeling in digital battlefields. It constructs a full-process technical system of "data acquisition-processing-fusion-modeling". Through innovative algorithms and modular design, it achieves deep integration of geological features and tunnel modeling. The underground space tunnel 3D modeling system takes full-link collaboration as its core design. Based on the business logic of "data acquisition-processing-modeling-application", it builds a four-layer architecture system of "core acquisition layer, key processing layer, application adaptation layer, and control and monitoring layer". Each layer is highly collaborative in terms of equipment selection, functional positioning, and data interaction. The core acquisition layer uses vehicle-mounted LiDAR as its core equipment, coupled with industrial-grade geological data acquisition terminals, vehicle-mounted industrial computers, and static RTK base stations, to complete high-precision synchronous acquisition of spatial data and geological attribute data inside and outside the tunnel. The key processing layer includes five modules: point cloud preprocessing, adaptive terrain fitting, multi-source data fusion, deep learning feature extraction, and 7D Gaussian splash modeling. Each module uses high-performance servers and GPU resources to achieve refined data processing. The application adaptation layer transforms the modeling results into engineering-usable information through multi-format conversion, engineering simulation, and real-time data interaction interfaces. The control and monitoring layer relies on industrial-grade hosts and dedicated protocols to achieve automated control of the entire system process and real-time monitoring of the operating status, ensuring efficient linkage between all levels.

[0127] As the core source of system data input, the core acquisition layer uses a vehicle-mounted LiDAR as its core device, with 1-2 units configured according to actual acquisition needs. It is also equipped with an industrial-grade geological data acquisition terminal (one per acquisition team), a vehicle-mounted industrial computer (one per acquisition vehicle), and a static RTK base station (one unit) covering the acquisition area. The core function of this layer is to collaboratively acquire geometric data of "internal tunnel space + surrounding terrain" and geological attribute data of "lithology, rock strength, and fault distribution," providing high-quality foundational data for subsequent processing. In terms of device connectivity, the LiDAR transmits point cloud data to the vehicle-mounted industrial computer in real time via 1000Mbps Ethernet. The geological data terminal uses Bluetooth 5.2 technology to obtain the LiDAR's positioning information and timestamp. The static RTK base station synchronizes the coordinate calibration data between the LiDAR and the vehicle-mounted computer via 4G / 5G networks, ensuring spatial consistency of the acquired data.

[0128] The critical processing layer is the core of the system's data processing, consisting of five collaborative functional modules, each relying on dedicated hardware resources to achieve specific processing objectives. The point cloud preprocessing module, deployed on a backend industrial server, primarily performs operations such as improved statistical filtering for noise reduction, Gaussian-bilateral hybrid filtering for smoothing, dynamic bounding box spatial segmentation, and quality inspection closed-loop control. The adaptive terrain fitting module shares backend server hardware resources with the preprocessing module. Its core consists of a Terrain Complexity Index (TCI) calculation unit, a dynamic hierarchical fitting unit, and a fitting result optimization and verification unit, responsible for the accurate generation of the terrain model. The multi-source data fusion module, based on independent computing nodes, achieves deep integration of geometric and geological data through three units: initial geological constraint field construction, incremental constraint field calculation, and gradient descent iterative optimization. The deep learning feature extraction module relies on a server cluster, equipped with a 3D-CNN network training and inference framework, a spatial-channel dual attention mechanism processing unit, and a key structure priority determination unit, to automatically identify key structures. The 7D Gaussian splash modeling module, as the core computing node, achieves high-precision real-time modeling through four units: 7D Gaussian parameter initialization, conditional slicing processing, adaptive Gaussian refinement, and differential rendering.

[0129] The application adaptation layer serves as a crucial bridge connecting modeling results with engineering applications. Its core comprises a multi-format conversion unit, an engineering simulation and deduction unit, and a real-time data interaction interface. The multi-format conversion unit supports conversions to commonly used engineering formats such as SHP, KML, and IFC, meeting the data usage needs of different platforms. The engineering simulation and deduction unit features construction path planning and equipment deployment simulation functions, providing engineering decision support based on modeling results. The real-time data interaction interface can connect to devices such as on-site stress sensors and construction monitoring cameras, enabling the linkage between modeling results and real-time monitoring data. This layer connects to the backend server via a local area network and also provides a web-based visual operation interface, allowing users to remotely set modeling parameters, view processing progress, and export result reports, enhancing system usability.

[0130] The control and monitoring layer is responsible for the coordinated scheduling and status management of the entire system. It connects modules at each level via a Modbus TCP protocol control bus, enabling automated process triggering (e.g., automatically starting the point cloud preprocessing process after data acquisition). In terms of monitoring, this layer can display the data processing progress of each module, GPU / CPU resource utilization, and error alarm information in real time, allowing operators to promptly grasp the system's operating status. It also possesses fault location and automatic recovery capabilities, automatically recording error logs and attempting to restart faulty modules to ensure the continuity of the modeling process. The hardware carrier for this layer is an industrial-grade monitoring host, supporting both local operation and remote control via a web interface, balancing operational flexibility and system stability.

[0131] The end-to-end processing stage is the core step in transforming the collected data into a 3D model, sequentially completing five major steps: point cloud preprocessing, adaptive terrain fitting, multi-source data fusion, deep learning feature extraction, and 7D Gaussian splash modeling. Point cloud preprocessing removes dust and air scattering points using an improved statistical filtering + reflection intensity threshold algorithm (noise removal rate ≥99%). Gaussian filtering with a standard deviation of 0.03m is applied to straight sections, and bilateral filtering is used for edge regions (edge ​​sharpness ≥95%). Redundant point clouds are eliminated using dynamic bounding boxes with a cross-section of "tunnel size + 2m" (data volume reduced by 40%-60%). Finally, random sampling is used to check the noise residue rate (≤1%), detail retention rate (≥95%), and coordinate consistency (≤0.02m). If these standards are not met, the data is reprocessed. Adaptive terrain fitting (40-50 minutes): First, the TCI index is calculated based on the point cloud elevation difference, slope, and curvature, and the terrain complexity level is classified. Then, the number of fitting layers and functions are matched according to the level (e.g., 8 layers + cubic spline function is used for high-complexity terrain). Iterative optimization is performed using the least squares method (residual ≤ 0.05m). Finally, 100 validation points are randomly selected to compare the fitted model with the original point cloud (average error ≤ 0.04m), and the terrain model in PLY format is output. Multi-source data fusion: Based on geological data, an initial geological constraint field with a resolution of 0.1m³ is constructed through kriging interpolation. Extract point cloud cracks and protrusion features to calculate constraint field increments ( For example, if the intensity of the fracture area is reduced by 10%-20%, the error (E < 0.01) is minimized using gradient descent (learning rate 0.01, 50 iterations), outputting an integrated "geometry + geology" model in E57 format. Deep learning feature extraction converts the E57 model into a voxel mesh and normalizes it. A pre-trained 3D-CNN + dual attention model is loaded to identify key structures (accuracy ≥ 92%) and determine their priorities, outputting key structure coordinates in JSON format, priority levels in XML format, and weighted feature maps in HDF5 format. 7D Gaussian splash modeling assigns ≥ 500 Gaussian atoms / m² to level 1 regions and ≤ 100 Gaussian atoms / m² to level 3 regions based on the weighted feature maps. The 7D Gaussian is projected into a 3D Gaussian model according to engineering requirements. The Gaussian parameters are adjusted using a neural network to optimize details. Finally, a 3D model is generated, and the rendering frame rate is tested (≥ 380 FPS), outputting a model error report (≤ 0.03m).

[0132] In this embodiment, the 3D-CNN+dual attention model uses a 3D-CNN combined with a spatial-channel dual attention fusion network to extract and prioritize key geological structural features of the tunnel. The overall structure consists of four parts: input preprocessing, 3D-CNN backbone feature extraction, dual attention feature enhancement, and dual-branch prediction output. During the model input stage, preprocessing is performed on the integrated point cloud model that combines geometric and geological attributes. This model is uniformly converted into a 0.1m³ resolution 3D voxel grid and numerical normalized. The standardized single-sample voxel size is 128×128×64, providing regular input data for subsequent network training and inference. The backbone network of this model uses a four-layer 3D convolutional structure to extract deep spatial features step by step. Each layer uses a 3×3×3 standard convolutional kernel with a stride of 1, and is paired with a ReLU activation function to complete nonlinear feature mapping. The first three convolutional layers are followed by 2×2×2 3D max pooling operations to achieve feature dimensionality reduction and key information filtering. The number of output channels for the four convolutional layers are 16, 32, 64, and 128 respectively, which can refine the 3D details of tunnel terrain, cracks, structural boundaries, etc. layer by layer. After the main convolutional feature output, a parallel dual attention module is connected to achieve adaptive weighted enhancement of features. The channel attention branch uses global 3D average pooling combined with two fully connected layers and a sigmoid function to generate channel weights, adaptively strengthening effective feature channels and suppressing ineffective and redundant channels. The spatial attention branch concatenates and fuses the maximum and mean information of multi-channel features, and generates a 3D spatial mask through 1×1×1 3D convolution and a sigmoid function to accurately highlight the key structural features of the tunnel. The outputs of the two attention branches are multiplied and fused element-wise to finally optimize and enhance the overall features. At the end of the model, a dual-branch parallel prediction structure is set to complete the tasks of structural classification and spatial coordinate regression respectively. The classification branch completes the dimensional transformation through two fully connected layers, and finally uses the Softmax function to achieve the probability discrimination of three types of roadway structures, corresponding to the first-level key structures of intersections and emergency exits, the second-level structures of lining and pipelines, and the third-level structures of ordinary roadways. The regression branch relies on fully connected layers and linear activation to output the three-dimensional bounding box coordinate parameters of the key structures of the roadways, accurately calibrating the spatial range of the structures, and finally outputting a priority label JSON file and a weighted feature map HDF5 file.

[0133] The input carrier for training and inference of the 3D-CNN+dual attention model is an integrated point cloud model that combines the geometry and geological attributes of the tunnel. After voxel meshing, the input data is formed into fixed-dimensional data. The input channels contain four sets of feature information: three-dimensional x, y, and z coordinates, and intensity attributes, comprehensively covering the spatial and geological dimensional parameters required for tunnel modeling. The training datasets are all derived from real-world engineering scenarios, with broad sample coverage and strong scenario adaptability. During data preprocessing, all sample values ​​are uniformly normalized to the 0-1 range. Data augmentation is performed using ±15-degree random rotation and random voxel noise perturbation to effectively improve the model's generalization ability. The model output contains two core types of effective information: first, a structure classification probability vector, which determines the tunnel structure level using three sets of probability values, selecting the structure type corresponding to the highest probability value as the final priority label; second, three-dimensional bounding box coordinate parameters, which accurately define the spatial boundaries of key structures using six-dimensional coordinate data. In actual engineering implementation, a confidence threshold of 0.7 is set, retaining only recognition results with predicted probabilities higher than the threshold, filtering out low-confidence noise interference, and ensuring the accuracy of structure recognition.

[0134] The model is a multi-task learning network that combines classification and regression. The overall loss function is a weighted fusion of structural classification loss and coordinate regression loss, which can simultaneously constrain the model's classification accuracy and localization accuracy. The overall formula is as follows: ;in, Represents the structural classification loss. For the weight of the classification task, Indicates the regression task weights. This corresponds to its weight structure. Specifically, the structural classification loss uses the multi-class cross-entropy loss function to accurately measure the deviation between the predicted value and the true label of the roadway structural level, as shown in the following formula: In the formula, Representative sample The true structural label, This represents the network's predicted classification probability for the sample. This represents the number of samples in a single training batch. The coordinate regression loss uses a smoothed L1 loss function, which effectively suppresses outlier errors in 3D coordinate prediction and improves structural localization stability. The formula is as follows: In the formula, The coordinates of the bounding box predicted by the network. These are the actual coordinates measured in the engineering project. Based on the engineering measurement and calibration results, the task weights are categorized. The value is 1.0, representing the regression task weight. A value of 0.8 is used to achieve a balanced optimization between the two types of tasks.

[0135] This model is trained in parallel using an NVIDIA A100 GPU, employs the Adam optimizer for iterative parameter updates, and has a base learning rate set to... The weight decay coefficient is set to The optimizer uses a first-order momentum coefficient of 0.9 and a second-order momentum coefficient of 0.999. This parameter configuration is adapted to the training patterns of 3D point cloud feature extraction, effectively avoiding gradient divergence and overfitting. The model training batch size is set to 16, with a total of 120 iterations. A segmented learning rate decay strategy is employed during training, reducing the learning rate to one-tenth of its original value in the 60th and 90th iterations respectively, gradually refining the parameter optimization effect. An early stopping mechanism is also configured, automatically terminating training when the validation set classification accuracy shows no improvement for 12 consecutive iterations, saving training computational resources while ensuring model accuracy.

[0136] The training process was accompanied by a fixed data augmentation strategy, with the random rotation probability set to 0.4 and the random voxel noise perturbation probability set to 0.3. The model was trained using transfer learning, loading pre-trained 3D-CNN weights from the alleyway point cloud domain and fine-tuning all network layers. The dual attention module was trained from scratch with entirely new parameters, balancing model convergence speed and feature adaptability.

[0137] The flowchart of the 3D geological constraint field iterative fusion technology presents the process of this technology. Using preprocessed point cloud data and a geological attribute table as input, an initial geological constraint field is first constructed through Kriging interpolation (0.1 m³ resolution). Next, features such as fractures and protrusions in the point cloud are extracted, and the increment of the constraint field is dynamically adjusted based on geological attributes. Then, gradient descent is used for iterative optimization (50 iterations), and the accuracy is controlled using an objective function method. Once the accuracy requirements are met, the final 3D geological constraint field is generated and bound to the terrain fitting model, ensuring an attribute matching rate of ≥95%. The entire process forms a closed loop of "construction-extraction-optimization-feedback," achieving precise iterative fusion of the 3D geological constraint field.

[0138] The 7D Gaussian splash differential modeling process takes an E57 format integrated model as input, first performing voxel mesh conversion and normalization; then, using deep learning to identify key structures and prioritize them (Level 1, such as intersections / emergency exits, ≥500 per square meter; Level 3, ordinary sections, ≤100 per square meter), achieving differential Gaussian atom allocation; subsequently, 7D Gaussian conditional slicing (3D space + 1D time + 3D viewpoint → 3D Gaussian) is performed, followed by adaptive Gaussian refinement and parameter optimization; after rendering testing (frame rate ≥380FPS), the final output is a 3D model and error report (≤0.03m). The entire process, through differential modeling and multi-dimensional optimization, achieves high-precision and high-efficiency 7D Gaussian model construction.

[0139] The application output phase primarily involves model format conversion, engineering simulation, and result delivery. For multi-format conversion, the 7D Gaussian modeling results are converted to commonly used engineering platform formats such as SHP (GIS platform), KML (project management system), and IFC (BIM platform), with conversion time controlled within 30 minutes and a data loss rate not exceeding 2%. In engineering simulation, based on the geological attributes included in the model (such as fault distribution and rock strength), construction path planning (accuracy ≥90%) and equipment deployment simulation are completed, automatically avoiding unstable areas and providing support for engineering decision-making. For result delivery, a web interface provides users with 3D model files, a full-process quality report on data acquisition and processing, and engineering simulation results, supporting online viewing, downloading, and printing to ensure the successful application of modeling results in engineering practice.

[0140] To achieve the system goals of "high-precision acquisition, refined processing, and high real-time modeling", this embodiment innovatively developed five core technologies, forming differentiated technological advantages and breaking through the limitations of traditional modeling technologies. The dual-mode collaborative acquisition technology of lidar solves the problem of full-coverage acquisition in narrow underground spaces and open terrain by combining the acquisition modes of "vehicle-mounted mobile + fixed station" and combining lidar reflection intensity characteristics. The reflection intensity-driven point cloud fine preprocessing technology is based on the reflection intensity layering characteristics of point clouds. It adopts the dual criteria of "improved statistical filtering + reflection intensity threshold" for denoising and Gaussian-bilateral hybrid filtering to achieve a balance between denoising effect and detail preservation. The terrain complexity index (TCI)-driven dynamic layered fitting technology constructs the TCI index by fusing point cloud elevation difference, slope and curvature data, and automatically matches the fitting strategy to improve the fitting accuracy of complex terrain. The three-dimensional geological constraint field iterative fusion technology takes geometric features and geological attributes as dual inputs. Through constraint field construction, incremental calculation and iterative optimization, it achieves deep binding of geometric and geological information. The 7D Gaussian splash differential modeling technology combines key structure priority to achieve differential modeling, taking into account modeling accuracy, efficiency and real-time performance.

[0141] In terms of data acquisition technology, an innovative dual-mode collaborative acquisition scheme using lidar is adopted for complex scenarios involving underground tunnels and surface terrain: vehicle-mounted mobile acquisition is suitable for tunnel interiors, employing an S-shaped path at a speed of 3-5 km / h, achieving a point cloud density of ≥300 points / ㎡; fixed-station acquisition is suitable for surface terrain, using a triangular station layout with a coverage radius of 100m and an overlap rate of ≥60%. Simultaneously, a time synchronization of ≤1ms is achieved through a combination of WGS84 coordinate system, Bluetooth 5.2, and static RTK, resolving the issues of incomplete coverage and data asynchrony in single acquisition modes. Furthermore, enhanced acquisition strategies are designed for key areas, employing 360-degree rotation acquisition (5 degrees / s rotation speed, 72s duration) at tunnel intersections and focusing on acquisition in surface fault outcrop areas (point cloud density ≥50 points / ㎡), specifically improving data accuracy in complex structural areas and avoiding the distortion in key areas caused by the "one-size-fits-all" approach of traditional acquisition methods.

[0142] In terms of core technological logic, an innovative three-dimensional geological constraint field iterative fusion technology is proposed, constructing a closed-loop logic of "data input - initial constraint field construction - incremental calculation - iterative optimization": the initial constraint field is constructed through kriging interpolation with a resolution of 0.1 m³. The incremental feature is calculated by combining PCA algorithm to extract point cloud crack features with a width ≥0.1mm. Then, using a learning rate of 0.01 and 50 iterations (error) The gradient descent method (with time termination) was optimized to achieve deep binding of PLY format geometric data and JSON format geological attributes (matching rate ≥95%), breaking through the limitations of traditional modeling that only focuses on geometry and lacks geological attributes. Simultaneously, a differentiated Gaussian splash modeling logic was designed, allocating differentiated Gaussian atom densities based on the priority of key structures, transforming the 7D Gaussian model ("spatial 3D + temporal 1D + visual 3D") into a 3D modeling output. This achieved an efficiency of ≤3.2 hours for modeling a 1km tunnel while ensuring an emergency exit error of ≤0.02m.

[0143] In terms of implementation process and results, a standardized Gantt chart was designed for the entire process, breaking down the implementation into four stages: "Data Acquisition Preparation (1-2 days) - Data Acquisition (1km, 1-2 days) - End-to-End Processing (1km, 3-4 hours) - Application Output (30-60 minutes)". The time consumption of sub-tasks (e.g., point cloud preprocessing, 30-40 minutes) and dependencies (preprocessing → fitting → fusion) were clearly defined, and three levels of nodes were set: "Preparation Acceptance Point - Data Acceptance Point - Final Delivery Point". This addresses the problems of chaotic implementation processes and uncontrollable progress in traditional methods. From a modeling perspective, compared to traditional technologies, the clarity of key structures increased from 72% to 96%, the tomography error decreased from 5m to 0.8m, the rendering frame rate increased from 28FPS to 385FPS, and the data volume decreased from 15GB to 6GB, achieving quadruple optimization in "accuracy, efficiency, real-time performance, and lightweight design".

[0144] In terms of technological innovation, the first innovation is the dual-mode collaborative acquisition method of lidar. This method combines "vehicle-mounted S-shaped path acquisition (including 360-degree rotation strategy at intersections) + ground triangular station acquisition (including key marking of fault outcrops)" with static RTK (planar accuracy ≤1cm) and Bluetooth 5.2 time synchronization (≤1ms) to achieve collaborative acquisition of underground-ground data. The method aims to protect the acquisition path design, synchronization mechanism, and key area acquisition parameters. The second innovation is the iterative fusion algorithm for three-dimensional geological constraint fields, which uses "Kriging interpolation to construct..." PCA feature extraction calculation gradient descent iteration ( The core of the proposed protection algorithm is the constraint field construction parameters (0.1m³ resolution), incremental adjustment rules (such as reducing the intensity of the fracture zone by 10%-20%), iteration termination conditions, and the implementation method of "geometric + geological" data binding. Furthermore, there is the 7D Gaussian splash differential modeling method, which allocates Gaussian atom density based on the priority of key structures (levels 1-3), slices the 7D Gaussian (spatial 3D + temporal 1D + visual 3D) into a 3D model, proposes a protection priority division standard, Gaussian atom density threshold (≥500 atoms / m² vs ≤100 atoms / m²), and a balancing strategy between modeling accuracy (error ≤0.02m) and efficiency (3.2 hours / 1km).

[0145] Regarding system and process innovations, the first is a four-layer collaborative architecture system, which includes a core acquisition layer (LiDAR + terminal + computer) - a key processing layer (5 sub-module clusters) - an application adaptation layer (multi-format conversion + simulation) - a control and monitoring layer (Modbus TCP protocol + fault recovery). It aims to protect the connection logic between layers (blue forward flow + red reverse control), hardware configuration standards for each layer, and functional module division. The second is a standardized implementation process system, using a Gantt chart as its framework. This system covers "four-stage task decomposition (acquisition preparation - data acquisition - processing - output), three-level acceptance nodes (preparation - data - delivery), and sub-task time quantification (e.g., fusion processing 50-60 minutes)." It aims to protect the phase division standards, node acceptance conditions, and time quantification indicators of the process, ensuring the solution is replicable and implementable.

[0146] In terms of innovative application value, one aspect is a solution to improve the accuracy of complex scene modeling. This solution achieves a modeling error of ≤0.02m for key areas of underground tunnels (intersections, emergency exits, faults) through a combination of "dual-mode acquisition (coverage) + iterative fusion (attributes) + differentiated modeling (accuracy)". It is intended to protect the application method of this solution in complex scenarios such as mining and tunnel engineering, as well as the technical combination strategy for improving accuracy. The other aspect is a lightweight modeling and real-time rendering solution. This solution achieves 385FPS real-time rendering while ensuring accuracy through differentiated Gaussian atomic allocation and data compression. It is intended to protect the application of this solution in interactive decision-making at engineering sites (such as construction path planning) to solve the problems of large data volume and rendering lag in traditional modeling.

[0147] Compared to single-parameter Gaussian modeling, this approach achieves an optimization that balances "priority differentiation" and "accuracy and efficiency." Single-parameter Gaussian modeling is a commonly used and efficient method in existing 3D modeling, but it suffers from a "one-size-fits-all" approach to modeling strategies. On the one hand, these techniques often use a fixed density of Gaussian atoms (e.g., uniformly allocated at 200 atoms / ㎡), failing to consider the varying importance of different structures within the tunnel. For example, for critical structures such as intersections and emergency exits, a fixed density can easily lead to insufficient detail restoration (edge ​​sharpness often falls below 75%), while for ordinary tunnel sections, it results in wasted computational power. On the other hand, traditional Gaussian modeling is mostly based on 3D and lacks consideration of the time dimension (acquisition sequence) and the perspective dimension (multi-view fusion), resulting in a low real-time rendering frame rate (usually ≤30FPS), making it difficult to support interactive decision-making on-site (such as dynamic adjustment of construction paths), and the modeling time is relatively long (8-10 hours for 1km tunnel), failing to meet the needs of rapid exploration.

[0148] The 7D Gaussian splash differential modeling technology proposed in this application achieves dual optimization to address the aforementioned shortcomings, with particularly prominent technical advantages: First, the "key structure priority division" mechanism divides the tunnel structure into multiple levels according to the importance of the project and matches differentiated Gaussian atomic densities to reduce redundant data, thereby avoiding distortion of key structures and reducing computing power and storage costs; Second, it expands the Gaussian parameters to 7 dimensions (3D space + 1D time + 3D perspective). The time dimension can be correlated with time-series data, and the perspective dimension can be integrated with perspectives collected from multiple devices, enabling the model to not only restore static outlines but also implicitly contain dynamic change information. At the same time, the 7D parameter optimization algorithm significantly improves real-time rendering performance, which can directly support real-time interaction on the engineering site (such as construction personnel viewing the model and adjusting work plans via tablets); Third, it achieves a balance between modeling accuracy and efficiency. Through differentiated density allocation and 7D parameter optimization, the modeling time for 1km tunnel is reduced from 8.5 hours with traditional technology to 3.2 hours, finding the optimal solution between the engineering requirements of "rapid exploration" and "accurate modeling," and solving the dilemma of "either low accuracy or low efficiency" in traditional technology.

[0149] From an engineering implementation perspective, existing technologies, due to their "separate" and "single" nature, often require multiple systems to work together in complex underground tunnel scenarios. This not only results in high equipment costs but also cumbersome data interaction between systems, making it difficult to form a complete solution. This application proposal, however, deeply integrates 3D geological constraint field iterative fusion technology with 7D Gaussian splash differential modeling technology, forming an integrated technology chain of "data fusion - dynamic optimization - differential modeling - real-time application." This allows for the completion of the entire process from data acquisition to application without the need for multiple systems, reducing equipment costs by over 40% and improving data interaction efficiency by over 80%. Furthermore, all parameters in the technical solution (such as Kriging interpolation resolution of 0.1 m³, Gaussian atomic density threshold, and iteration termination conditions) are optimized. All of these technologies are calibrated based on actual engineering measurement data and can be directly applied to underground tunnel scenarios with different widths (3-8m) and different slopes (0-45 degrees). Their adaptability is far superior to existing technologies (traditional technologies require parameter readjustment for different scenarios, resulting in a long adaptation cycle), laying the foundation for subsequent large-scale promotion.

[0150] Figure 4 This is a schematic diagram of a three-dimensional tunnel modeling device incorporating geological features, provided as an embodiment of the present invention. Figure 4 As shown, the device includes:

[0151] The point cloud data acquisition module 410 is used to acquire point cloud data of the alleyway area through vehicle-mounted radar and fixed-frame radar.

[0152] The terrain model construction module 420 is used to determine the terrain complexity index of each sub-region in the alleyway region based on the point cloud data, determine the number of fitting layers and the type of fitting function for each sub-region based on the terrain complexity index, and fit the point cloud data of each sub-region based on the number of fitting layers and the type of fitting function to obtain the terrain model of the alleyway region.

[0153] The geological constraint field construction module 430 is used to acquire the geological attribute information of the tunnel area and construct a geological constraint field based on the point cloud data of the tunnel area and the geological attribute information.

[0154] The tunnel model generation module 440 is used to perform differentiated modeling of different regions and different times based on the terrain model, the geological constraint field and the corresponding time information, so as to obtain the three-dimensional modeling result of the tunnel.

[0155] The technical solution of this embodiment improves the richness and accuracy of point cloud data by collecting point cloud data in multiple ways; it achieves high-precision fitting of tunnels and complex terrain by performing regional differentiation fitting through terrain complexity; it enhances the geological information expression of the model by integrating point cloud data and geological exploration data through multi-source data fusion; and it achieves a dual improvement in modeling accuracy and efficiency through differentiated modeling of different regions and different times.

[0156] Optionally, the geological attribute information of the tunnel area includes geological attribute measurement information of multiple discrete sample points;

[0157] The geological constraint field construction module includes:

[0158] The weight determination submodule is used to determine multiple discrete sample points corresponding to the interpolation point, and to determine the weight of each discrete sample point relative to the interpolation point based on the distance between each discrete sample point and the interpolation point.

[0159] The geological attribute estimation submodule is used to determine the geological attribute estimation information of the point to be interpolated based on the geological attribute measurement information and corresponding weights of each discrete sample point.

[0160] The geological constraint field construction submodule is used to determine the geological constraint field based on the geological attribute measurement information of multiple discrete sample points in the tunnel area, the geological attribute estimation information of multiple interpolation points, and the point cloud data.

[0161] Optional, the weight determination submodule is specifically used for:

[0162] Determine the distance between the target discrete sample point and the interpolation point, and determine the distance-matching sample point pair from the plurality of discrete sample points based on the distance;

[0163] The weight of the target discrete sample point relative to the interpolation point is determined by the ratio of the sum of squared differences in the geological attribute measurement information corresponding to the discrete sample points included in all the sample point pairs to the number of the sample point pairs.

[0164] The geological constraint field construction submodule includes:

[0165] An initial geological constraint field construction unit is used to construct an initial geological constraint field based on the geological attribute measurement information of multiple discrete sample points in the tunnel area and the geological attribute estimation information of multiple interpolation points; wherein, the initial geological constraint field includes the grid division result of the tunnel area, and the constraint value of each grid node is determined based on the geological attribute information of the included discrete sample points and interpolation points;

[0166] A multi-source data matching unit is used to spatially match the point cloud data with the initial geological constraint field to determine the grid point cloud data corresponding to each grid node.

[0167] The constraint field increment determination unit is used to extract features of the grid point cloud data based on the PCA algorithm, and determine the constraint field increment of the grid node based on the features.

[0168] The constraint field iteration unit is used to iteratively optimize the initial geological constraint field based on the constraint field increment using the gradient descent method to obtain the final geological constraint field.

[0169] Optional, constraint field increment determination element, specifically used for:

[0170] The features of the grid point cloud data are extracted based on the PCA algorithm to obtain the covariance matrix corresponding to the grid point cloud data;

[0171] The largest eigenvalue of the covariance matrix is ​​determined, and the constraint field increment is determined based on the largest eigenvalue and the constraint value of the grid node.

[0172] Optionally, the terrain complexity index determination unit in the terrain model construction module is specifically used for:

[0173] The elevation difference, slope, and curvature of each sub-region in the tunnel area are determined based on the point cloud data.

[0174] The terrain complexity index of each sub-region is determined based on the fusion results of the elevation difference, slope, and curvature.

[0175] Optional, a tunnel model generation module, specifically used for:

[0176] The key structural priorities of different sub-regions in the tunnel area are determined based on the terrain model, and Gaussian atom densities are allocated differentially based on the key structural priorities.

[0177] The terrain model, the geological constraint field, and the corresponding time information are used to construct 3D projection spatial parameters.

[0178] Based on the Gaussian splashing method, differentiated modeling of different sub-regions is performed according to the Gaussian atomic density and the 3D projection space parameters corresponding to each sub-region, resulting in a three-dimensional modeling result of the tunnel.

[0179] The tunnel 3D modeling device combining geological features provided in the embodiments of the present invention can execute the tunnel 3D modeling method combining geological features provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method execution.

[0180] The acquisition, storage, use, and processing of data in this application comply with relevant national laws and regulations and do not violate public order and good morals.

[0181] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0182] Figure 5 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0183] like Figure 5 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0184] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0185] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods described above, such as the tunnel 3D modeling method incorporating geological features.

[0186] In some embodiments, the tunnel 3D modeling method incorporating geological features can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the tunnel 3D modeling method incorporating geological features described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the tunnel 3D modeling method incorporating geological features by any other suitable means (e.g., by means of firmware).

[0187] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific reference products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0188] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0189] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0190] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0191] The systems and technologies described herein can be implemented in computing systems that include back-end components (e.g., as data servers), or computing systems that include switching components (e.g., application servers), or computing systems that include front-end components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such back-end, switching, or front-end components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0192] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0193] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication unit 19, or installed from storage unit 18, or installed from ROM 12. When the computer program is executed by processor 11, it performs the functions defined in the methods of the embodiments of the present invention.

[0194] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the tunnel three-dimensional modeling method incorporating geological features as provided in any embodiment of this application.

[0195] In implementing the computer program product, computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0196] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0197] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for three-dimensional modeling of tunnels incorporating geological features, characterized in that, The method includes: Point cloud data of the tunnel area is acquired using vehicle-mounted radar and fixed-frame radar. The terrain complexity index of each sub-region in the alleyway region is determined based on the point cloud data. The number of fitting layers and the type of fitting function for each sub-region are determined based on the terrain complexity index. The point cloud data of each sub-region is then fitted based on the number of fitting layers and the type of fitting function to obtain the terrain model of the alleyway region. Obtain the geological attribute information of the tunnel area, and construct a geological constraint field based on the point cloud data of the tunnel area and the geological attribute information; Based on the terrain model, the geological constraint field, and the corresponding time information, differentiated modeling is performed for different regions and at different times to obtain the three-dimensional modeling results of the tunnel.

2. The method according to claim 1, characterized in that, in, The geological attribute information of the tunnel area includes geological attribute measurement information of multiple discrete sample points; Based on the point cloud data of the tunnel area and the geological attribute information, a geological constraint field is constructed, including: Multiple discrete sample points corresponding to the interpolation point are determined, and the weight of each discrete sample point relative to the interpolation point is determined based on the distance between each discrete sample point and the interpolation point. The geological attribute estimation information of the interpolation point is determined based on the geological attribute measurement information and corresponding weights of each discrete sample point. Based on the geological attribute measurement information of multiple discrete sample points in the tunnel area, the geological attribute estimation information of multiple interpolation points, and the point cloud data, the geological constraint field is determined.

3. The method according to claim 2, characterized in that, Determining the weight of each discrete sample point relative to the interpolation point based on the distance between each discrete sample point and the interpolation point includes: Determine the distance between the target discrete sample point and the interpolation point, and determine the distance-matching sample point pair from the plurality of discrete sample points based on the distance; The weight of the target discrete sample point relative to the interpolation point is determined by the ratio of the sum of squared differences in the geological attribute measurement information corresponding to the discrete sample points included in all the sample point pairs to the number of the sample point pairs.

4. The method according to claim 2, characterized in that, Based on the geological attribute measurement information of multiple discrete sample points in the tunnel area, the geological attribute estimation information of multiple points to be interpolated, and the point cloud data, the geological constraint field is determined, including: An initial geological constraint field is constructed based on the geological attribute measurement information of multiple discrete sample points in the tunnel area and the geological attribute estimation information of multiple interpolation points; wherein, the initial geological constraint field includes the grid division result of the tunnel area, and the constraint value of each grid node is determined based on the geological attribute information of the included discrete sample points and interpolation points; Spatially match the point cloud data with the initial geological constraint field to determine the grid point cloud data corresponding to each grid node; The features of the grid point cloud data are extracted based on the PCA algorithm, and the constraint field increment of the grid node is determined based on the features. The initial geological constraint field is iteratively optimized using the gradient descent method based on the constraint field increment to obtain the final geological constraint field.

5. The method according to claim 4, characterized in that, The PCA algorithm is used to extract features from the grid point cloud data, and the constraint field increment of the grid nodes is determined based on the features, including: The features of the grid point cloud data are extracted based on the PCA algorithm to obtain the covariance matrix corresponding to the grid point cloud data; The largest eigenvalue of the covariance matrix is ​​determined, and the constraint field increment is determined based on the largest eigenvalue and the constraint value of the grid node.

6. The method according to claim 1, characterized in that, The terrain complexity index of each sub-region in the alleyway region is determined based on the point cloud data, including: The elevation difference, slope, and curvature of each sub-region in the tunnel area are determined based on the point cloud data. The terrain complexity index of each sub-region is determined based on the fusion results of the elevation difference, slope, and curvature.

7. The method according to claim 1, characterized in that, Based on the terrain model, the geological constraint field, and the corresponding time information, differentiated modeling is performed for different regions and at different times to obtain the three-dimensional modeling results of the tunnel, including: The key structural priorities of different sub-regions in the tunnel area are determined based on the terrain model, and Gaussian atom densities are allocated differentially based on the key structural priorities. The terrain model, the geological constraint field, and the corresponding time information are used to construct 3D projection spatial parameters. Based on the Gaussian splashing method, differentiated modeling of different sub-regions is performed according to the Gaussian atomic density and the 3D projection space parameters corresponding to each sub-region, resulting in a three-dimensional modeling result of the tunnel.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the tunnel three-dimensional modeling method incorporating geological features as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the three-dimensional tunnel modeling method incorporating geological features as described in any one of claims 1-7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the three-dimensional tunnel modeling method incorporating geological features according to any one of claims 1-7.

Images