Tunnel parametric modeling method and system based on binocular vision

Abstract

The invention discloses a tunnel parametric modeling method and system based on binocular vision, and relates to the field of tunnel modeling. Comprising the steps that tunnel images shot by a binocular camera with known characteristic parameters are collected, the images are preprocessed, and a training image set is generated; establishing a neural network model based on the training image set, and realizing prediction functions of inputting tunnel images and outputting characteristic parameters through model training; then, a binocular camera is calibrated, camera parameters are obtained, the calibrated camera is used for collecting an image of a target tunnel, the image is input into the model after preprocessing and splicing, and a characteristic parameter predicted value is generated; and establishing a finite element model based on tunnel characteristic parameters, analyzing rock mass strength parameters, and generating a shear slip index, a surrounding rock strength ratio index and a comprehensive strength evaluation coefficient for judging tunnel strength and stability. The mechanical property and the potential instability risk of the tunnel are comprehensively reflected, and a scientific basis is provided for tunnel construction and subsequent maintenance.

Description

Technical Field

[0001] The present invention relates to the field of tunnel modeling, and in particular to a tunnel parameterized modeling method and system based on binocular vision. Background Art

[0002] With the rapid development of tunnel engineering and underground space development, tunnel construction plays an important role in transportation, mineral resource mining and urban underground space utilization. However, the design and construction process of tunnel engineering faces complex geological environment and diverse rock structure characteristics, especially in unfavorable geological areas, such as weak surrounding rock, high ground stress areas or fault fracture zones, where the instability of surrounding rock often leads to safety hazards. Traditional tunnel rock analysis and modeling methods rely on manual measurement, empirical formulas and numerical simulations, which are time-consuming, low-precision, and difficult to obtain and analyze the key characteristic parameters of the tunnel in real time and automatically.

[0003] Binocular vision technology has been widely used in engineering surveying in recent years. As a three-dimensional reconstruction technology similar to the principle of the human eye, binocular vision can calculate the geometric characteristics of the target, such as depth, inclination, width, etc., through the captured images. However, the way of obtaining geometric information by relying solely on binocular vision is limited by the complex lighting conditions in the tunnel construction environment, the complexity of the rock surface texture, and noise interference, and it is often difficult to directly extract the precise parameters of the tunnel surrounding rock. At the same time, the large amount of tunnel image data captured by binocular vision requires efficient processing and modeling. The traditional parameter extraction process is highly dependent on image quality, resulting in limited application in actual engineering. In addition, the parametric modeling of rock strength and tunnel stability assessment is a multivariate, multidisciplinary problem. The existing methods have a low level of automation for tunnel surrounding rock strength and stability evaluation, and it is difficult to perform real-time analysis based on dynamic information on site.

[0004] In order to solve the above problems, tunnel parametric modeling technology based on artificial intelligence and binocular vision has gradually been studied and applied. By combining deep learning algorithms and binocular vision technology, not only can key feature parameters be extracted from tunnel images more efficiently, but also rock strength and tunnel stability analysis can be performed by establishing a tunnel mechanics model based on the finite element method. However, there is still a lack of a feasible and integrated technical solution that can directly extract the geometric parameters of the tunnel from the tunnel images collected by binocular vision, and generate a comprehensive comprehensive stability evaluation index in combination with the surrounding rock mechanics parameters.

[0005] In the prior art, publication number CN114707220A discloses a tunnel BIM intelligent modeling method based on Revit macro program. In view of the fact that the current long tunnel BIM model has many structures and components, the modeling process is complex, the modeling difficulty and workload are large, how to quickly and intelligently arrange each tunnel structure and component accurately according to the design and construction requirements to replace the manual creation of the long tunnel BIM model, this method uses Revit macro program-based visual programming, and with the help of self-encoding macro program, the traffic engineering line can be directly fitted in Revit, and the self-encoding macro program is used to call the sub-family model to drive the automatic creation of the tunnel BIM model, so that the process of creating the BIM model no longer depends on software such as Civil3D, reduces the modeling steps, creates the sub-family model in parameter form, and drives the tunnel BIM model to be assembled and adjusted with parameter data, automatically improves the attribute content, and replaces the existing manual input of attribute content to complete the modeling, thereby improving the accuracy and efficiency of tunnel BIM modeling. However, in this method, Revit is mainly good at the component layout and geometric model generation of the building information model (BIM). However, in tunnel modeling, the geometry and layout of the tunnel are usually significantly affected by geological conditions, such as the inclination and inclination of the tunnel structure surface. At the same time, there is a lack of direct support for the mechanical properties (such as compressive strength, shear strength) and stability analysis of the tunnel surrounding rock. The safety and stability of tunnel engineering often require a comprehensive evaluation based on rock mass strength, geological parameters and tunnel geometry. As a result, the accuracy and effectiveness of the model are reduced.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0007] The purpose of the present invention is to provide a coal mine supervision method and system based on video images to solve the problems raised in the above background technology.

[0008] To achieve the above object, the present invention provides the following technical solutions: A tunnel parameter modeling method and system based on binocular vision, the specific steps include: Using a binocular camera to shoot several tunnels with known characteristic parameters, and preprocessing the tunnel images, mapping the preprocessed tunnel images with corresponding characteristic parameters one by one to generate a training image set, wherein the preprocessing includes distortion correction and noise reduction enhancement, and the characteristic parameters include tunnel structural surface inclination, inclination, average waveform frequency, average surface fractal dimension, tunnel length and tunnel width; A neural network model is established, the tunnel image in the training image set is used as the input of the neural network model, and the characteristic parameters corresponding to the tunnel image are used as labels to train the neural network model, so as to obtain a tunnel parameter prediction model whose input is the tunnel image and output is the tunnel characteristic parameters; Based on the mobile intelligent body driving the binocular camera to collect images of the target tunnel, several target tunnel images are obtained by compound acquisition, and they are pre-processed based on the same method; After preprocessing, the collected target tunnel images are spliced, the spliced ​​target tunnel images are input into the trained tunnel parameter prediction model, characteristic parameter prediction values ​​of the target tunnel are output, and a finite element model of the target tunnel is established based on the obtained characteristic parameter prediction values; Based on the established finite element model of the target tunnel, the rock strength parameters of the target tunnel are analyzed and obtained. According to the rock strength parameters of the target tunnel and combined with the predicted values ​​of the characteristic parameters of the target tunnel, a shear sliding index and a surrounding rock strength ratio index are generated. A comprehensive strength evaluation coefficient is generated through the shear sliding index and the surrounding rock strength ratio index. The strength of the target tunnel is judged based on the comprehensive strength evaluation coefficient. The rock strength parameters include rock compressive strength, rock tensile strength and rock shear strength.

[0009] Furthermore, a neural network model is established, and a neural network prediction model is established based on the LSTM model. The activation function and optimization algorithm are selected, wherein the Tanh function is selected as the activation function, and Adam is selected as the optimization algorithm of the LSTM model; the formula of the Tanh function is: ; In the formula, Represents the Tanh function, independent variable Represents the weighted sum of the neuron's input, that is, the result of weighted summation of the input received by the neuron from the previous layer; At the same time, the hyperparameters of the LSTM model are set, and the hyperparameters of the LSTM model include: the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the batch processing number, and the number of hidden layer neurons; The number of network layers is set to a 3-layer network structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training times is set to 100, the batch size is set to 256, and the number of hidden layer neurons is 32.

[0010] Furthermore, the collected tunnel image is preprocessed, and the preprocessing includes distortion correction and noise reduction enhancement, wherein the distortion correction includes radial distortion correction and tangential distortion correction, and the specific logic of the radial distortion correction and the tangential distortion correction is: establish three radial distortion coefficients, calibrated as , and , the formula for obtaining the ideal image coordinates without tangential distortion is: ; Establish two tangential distortion coefficients and calibrate them as and , the model expression without radial distortion is: ; By combining the two distortions, the effects of both can be eliminated at the same time. The parameter expression after superposition is: ; In the formula, , and are three different radial distortion coefficients, and are two different tangential distortion coefficients, represents the undistorted image coordinates under the ideal optical center system, is the image coordinate of a pixel in the captured image, is the distance from the calibration point to the imaging center, where the image coordinate system takes the principal point position coordinate as the image center, the origin is the intersection of the camera optical axis and the imaging plane, and the right is Axis positive direction, downward The positive direction of the axis is established, where the radial distortion coefficient, tangential distortion coefficient and principal point position coordinates are obtained through camera calibration; The method for performing noise reduction and enhancement processing on the tunnel image is as follows: using the wavelet transform denoising method to denoise the image after distortion correction, and using bilateral filtering to enhance the details of the tunnel image.

[0011] Furthermore, a mobile intelligent agent is used to carry a calibrated binocular camera to collect images of the target tunnel, wherein A * The algorithm generates the motion path of the mobile agent and dynamically corrects the brightness of the shot according to the environmental parameters in the target tunnel, where the environmental parameters include the ambient brightness and the horizontal distance between the tunnel structure surface and the binocular camera. The specific formula for dynamically correcting the brightness of the shot is: ; In the formula, is the corrected light intensity, is the initial light brightness, is the ambient brightness, is the horizontal distance between the tunnel structure surface and the binocular camera, is the reflectivity of the tunnel structure surface.

[0012] Furthermore, the collected target tunnel images are pre-processed and then stitched, wherein the logic of image stitching is: a number of marker points are randomly set on the structural surface of the target tunnel, and the image is stitched based on the marker points, specifically, the SIFT algorithm in the feature point-based matching algorithm is selected to extract the features of the marker points; a Gaussian difference pyramid is established by selecting the target area image, the position of the feature point is determined, the direction is assigned to the feature point, and a descriptor is constructed for the determined feature point, wherein the descriptor is a vector, and the descriptor in the SIFT algorithm is generated by calculating the gradient direction histogram in the area around the feature point; After the landmark points of the image are extracted, the landmark point pairs are searched and matched. After the initial matching pairs are obtained, the double detection strategy coupling algorithm is used to eliminate the wrong matches, and the best matching results of the landmark points are obtained. The images are spliced ​​based on the matching results.

[0013] Furthermore, according to the rock mass strength parameters of the target tunnel and the characteristic parameters of the target tunnel, the shear sliding index and the surrounding rock strength ratio index are generated, wherein the formula for calculating the shear sliding index is: ; In the formula, is the shear sliding index, is the shear strength of rock mass, is the structural surface roughness coefficient, is the predicted value of the inclination angle of the target tunnel structure surface, is the normal stress of the target tunnel, is the proportional constant set; Among them, the structural surface roughness coefficient It is calculated by the predicted values ​​of the average waveform frequency and average surface fractal dimension of the target tunnel structure surface. The specific calculation formula is as follows: ; In the formula, is the predicted value of the average surface fractal dimension, is the predicted value of the average waveform frequency; Normal stress of target tunnel Characterized by the target tunnel burial depth, the specific formula is: ; In the formula, is the unit weight of rock and soil, is the target tunnel depth, and are the predicted values ​​of the tunnel length and tunnel width of the target tunnel respectively.

[0014] Furthermore, the formula for calculating the surrounding rock strength ratio index is: ; In the formula, is the surrounding rock strength ratio index, and are the rock mass compressive strength and rock mass tensile strength, respectively. is the predicted value of the inclination of the target tunnel; The comprehensive strength evaluation coefficient is generated by the shear sliding index and the surrounding rock strength ratio index, where the formula for calculating the comprehensive strength evaluation coefficient is: ; In the formula, is the comprehensive strength assessment coefficient, and are the weight coefficients of shear sliding index and surrounding rock strength ratio index, respectively. and and All are greater than 0; The strength of the target tunnel is determined based on the comprehensive strength evaluation coefficient, wherein the specific logic for determining the strength of the target tunnel is as follows: when When , the target tunnel strength is judged to be excellent, indicating that the target tunnel has no risk of collapse and sliding; when When , the target tunnel is judged to have good strength, indicating that the target tunnel has the risk of sliding and should be reinforced; when When the target tunnel strength is judged to be poor, it means that the target tunnel is at risk of collapse and is not suitable for entry; in It is the preset tunnel strength judgment threshold.

[0015] The present invention also provides a tunnel parameter modeling system based on binocular vision, and the tunnel parameter modeling system based on binocular vision is used to execute the above-mentioned tunnel parameter modeling method based on binocular vision, including: A sample image processing module is used to shoot several tunnels with known characteristic parameters based on a binocular camera, and preprocess the tunnel images obtained by shooting, and map the preprocessed tunnel images with corresponding characteristic parameters one by one to generate a training image set, wherein the preprocessing includes distortion correction and noise reduction enhancement, and the characteristic parameters include tunnel structural surface inclination, inclination, average waveform frequency, average surface fractal dimension, tunnel length and tunnel width; The parameter prediction model training module is used to establish a neural network model, using the tunnel image in the training image set as the input of the neural network model, and using the characteristic parameters corresponding to the tunnel image as labels to train the neural network model, so as to obtain a tunnel parameter prediction model whose input is the tunnel image and output is the tunnel characteristic parameters; The target tunnel image acquisition module is used to acquire images of the target tunnel by driving the binocular camera based on the mobile intelligent body, obtain a number of target tunnel images by means of compound acquisition, and pre-process them based on the same method; A tunnel model building module is used to splice the collected target tunnel images after preprocessing, input the spliced ​​target tunnel images into the trained tunnel parameter prediction model, output the characteristic parameter prediction values ​​of the target tunnel, and establish the target tunnel finite element model based on the obtained characteristic parameter prediction values; The tunnel strength judgment module is used to analyze and obtain the rock strength parameters of the target tunnel based on the established finite element model of the target tunnel, generate a shear sliding index and a surrounding rock strength ratio index according to the rock strength parameters of the target tunnel and the predicted values ​​of the characteristic parameters of the target tunnel, generate a comprehensive strength evaluation coefficient through the shear sliding index and the surrounding rock strength ratio index, and judge the strength of the target tunnel based on the comprehensive strength evaluation coefficient, wherein the rock strength parameters include rock compressive strength, rock tensile strength and rock shear strength.

[0016] Compared with the prior art, the present invention has the following beneficial effects: First, this scheme uses binocular vision technology to directly collect tunnel image information, and combines image preprocessing technology, including distortion correction and noise reduction enhancement, to effectively solve the problems of complex lighting conditions and irregular rock surface texture in the tunnel construction environment. The accuracy of subsequent parameter extraction is ensured by high-quality image processing. At the same time, through the compound acquisition and image stitching method, the geometric structure of the tunnel space can be completely restored, laying the foundation for the establishment of an accurate tunnel finite element model. Secondly, by constructing a neural network model, the trained data set is input into the model, and a direct mapping relationship from the tunnel image to the characteristic parameters is established. Compared with traditional empirical formulas or manual inference methods, the neural network model can efficiently learn and fit complex data, thereby accurately predicting the structural parameters of the tunnel, such as the inclination and inclination of the structural surface. It improves the efficiency of tunnel parametric modeling and can dynamically adapt to the diversity of different tunnel environments. In addition, the finite element model is established through the predicted values ​​of the tunnel characteristic parameters, and further combined with the rock mass strength parameters to generate comprehensive stability evaluation indicators, including the shear sliding index and the surrounding rock strength ratio index. It fully reflects the mechanical properties and potential instability risks of the tunnel, providing a scientific basis for tunnel construction and subsequent maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the overall method flow of the present invention; Figure 2 It is a schematic diagram of the overall system structure of the present invention. DETAILED DESCRIPTION

[0018] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0019] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] Example: See also Figure 1 , the present invention provides a technical solution: A tunnel parameter modeling method based on binocular vision, the specific steps include: Step 1: Using a binocular camera, shoot several tunnels with known characteristic parameters, preprocess the tunnel images, map the preprocessed tunnel images with the corresponding characteristic parameters one by one, and generate a training image set. The preprocessing includes distortion correction and noise reduction enhancement. The characteristic parameters include the inclination angle, tilt, average waveform frequency, average surface fractal dimension, tunnel length and tunnel width of the tunnel structure surface.

[0021] The collected tunnel image is preprocessed, and the preprocessing includes distortion correction and noise reduction enhancement, wherein the distortion correction includes radial distortion correction and tangential distortion correction, and the specific logic of radial distortion correction and tangential distortion correction is: establish three radial distortion coefficients, calibrated as , and , the formula for obtaining the ideal image coordinates without tangential distortion is: ; Establish two tangential distortion coefficients and calibrate them as and , the model expression without radial distortion is: ; By combining the two distortions, the effects of both can be eliminated at the same time. The parameter expression after superposition is: ; In the formula, , and are three different radial distortion coefficients, and are two different tangential distortion coefficients, represents the undistorted image coordinates under the ideal optical center system, is the image coordinate of a pixel in the captured image, is the distance from the calibration point to the imaging center, where the image coordinate system takes the principal point position coordinate as the image center, the origin is the intersection of the camera optical axis and the imaging plane, and the right is Axis positive direction, downward The positive direction of the axis is established, where the radial distortion coefficient, tangential distortion coefficient and principal point position coordinates are obtained through camera calibration; The method for performing noise reduction and enhancement processing on the tunnel image is as follows: using the wavelet transform denoising method to denoise the image after distortion correction, and using bilateral filtering to enhance the details of the tunnel image.

[0022] The specific steps of the wavelet transform denoising method include: decomposing the distortion-corrected image through wavelet transform to obtain wavelet coefficients of the image at different scales and directions; thresholding the wavelet coefficients to set the low-amplitude wavelet coefficients to zero and retain the high-amplitude wavelet coefficients; inversely transforming the wavelet coefficients after thresholding to reconstruct the processed coefficients into an image to complete the image denoising; Bilateral filtering is used to enhance the details of the tunnel image. The specific filter transformation is based on the formula: ; In the formula, is the coordinate vector in the image coordinate system, is the coordinate vector The gray value at Gray value The gray value after bilateral filtering transformation is are all Gaussian functions, where The formula is: ; ; In the formula, is the coordinate vector in the image coordinate system, is the coordinate vector The gray value at and They are The standard deviation of .

[0023] Step 2: Establish a neural network model, use the tunnel image in the training image set as the input of the neural network model, and use the feature parameters corresponding to the tunnel image as labels to train the neural network model, so as to obtain a tunnel parameter prediction model with tunnel image as input and tunnel feature parameters as output.

[0024] Establish a neural network model, establish a neural network prediction model based on the LSTM model, select the activation function and optimization algorithm, among which the Tanh function is selected as the activation function, and Adam is selected as the optimization algorithm of the LSTM model; the formula of the Tanh function is: ; In the formula, Represents the Tanh function, independent variable Represents the weighted sum of the neuron's input, that is, the result of weighted summation of the input received by the neuron from the previous layer; At the same time, the hyperparameters of the LSTM model are set, and the hyperparameters of the LSTM model include: the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the batch processing number, and the number of hidden layer neurons; The number of network layers is set to a 3-layer network structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training times is set to 100, the batch size is set to 256, and the number of hidden layer neurons is 32.

[0025] In the tunnel parameter prediction task, tunnel images and corresponding characteristic parameters (such as structural surface inclination, tilt, length, width, etc.) not only have spatial correlation, but also may have certain temporal correlation. Especially in tunnel construction monitoring, the changes between continuous image data reflect the evolution of construction dynamics and geological conditions.

[0026] In tunnel construction or monitoring, the image data collected by the binocular camera is usually a continuous, time-related data sequence. Since LSTM can effectively capture time correlation, when processing these continuous image sequences, LSTM can learn the dynamic change characteristics between image frames.

[0027] During the acquisition process, tunnel images may be disturbed by complex lighting, noise, or rock surface texture, which may cause the features extracted from a single image to be unstable. LSTM can smooth out the noise in individual images and extract more stable and robust features by aggregating information in time series.

[0028] Tunnel parameter prediction involves not only a single parameter, but a comprehensive prediction problem of multiple parameters (such as inclination, tilt, width, and length). LSTM can solve complex multivariate prediction tasks by learning the nonlinear mapping relationship between multidimensional input and multidimensional output.

[0029] Step 3: Based on the mobile intelligent body, the binocular camera is driven to collect images of the target tunnel. Several target tunnel images are obtained through compound collection, and they are pre-processed based on the same method.

[0030] A binocular camera consists of two cameras with exactly the same parameters and placed in parallel. In this ideal state, the optical axes of the two cameras remain parallel and the projection planes are in the same plane. The binocular camera is calibrated using Zhang Zhengyou's calibration method. The specific steps include: making a checkerboard calibration plate; moving the calibration plate and collecting images; extracting corner points; obtaining focal length, principal point position coordinates and distortion coefficients, and the distortion coefficients include radial distortion coefficients and tangential distortion coefficients. The binocular camera first performs single-target calibration, and then performs binocular calibration based on the single-target calibration data. Zhang Zhengyou's calibration method is a commonly used camera calibration method and will not be described here.

[0031] The duplex acquisition method is a method of acquiring the same target image at different angles, specifically acquiring images of the tunnel wall at different angles.

[0032] A mobile intelligent agent is used to carry a calibrated binocular camera to collect images of the target tunnel, where A * The algorithm generates the motion path of the mobile agent and dynamically corrects the brightness of the shot according to the environmental parameters in the target tunnel, where the environmental parameters include the ambient brightness and the horizontal distance between the tunnel structure surface and the binocular camera. The specific formula for dynamically correcting the brightness of the shot is: ; In the formula, is the corrected light intensity, is the initial light brightness, is the ambient brightness, is the horizontal distance between the tunnel structure surface and the binocular camera, is the reflectivity of the tunnel structure surface.

[0033] Among them, the corrected light brightness Used to indicate the brightness of light provided by an external light source during shooting. The larger the value, the greater the light brightness that needs to be provided.

[0034] Among them, the ambient brightness The smaller the value, the lower the brightness of the light in the tunnel, which will reduce the clarity of the captured image. With corrected light intensity Inversely proportional to the ambient brightness The smaller it is, the greater the light intensity needs to be provided to ensure the quality of the captured image. Exponential function , used to indicate the ambient brightness Nonlinear suppression of lighting adjustment. The increase of exponential function will rise rapidly, causing the denominator to increase, thereby reducing the correction for the additional light.

[0035] In tunnel engineering, the intensity of the light source will decay as the distance of light propagation increases. Therefore, the horizontal distance between the tunnel structure surface and the binocular camera The larger the value is, the clearer the image should be made by increasing the brightness of the light. and Proportional, using the logarithmic function The characteristics of the model make the effect of distance on light intensity decrease, which can avoid excessively magnifying the effect of long distance on light intensity and enhance the physical stability of the model.

[0036] Indicates the ability of the tunnel structure surface to reflect light, which is mainly affected by the tunnel material (such as rock, concrete, etc.) and the surface color (such as dark or light). The higher it is, the stronger the reflected light is, and the brightness of the external light required for shooting is reduced. The lower the reflection, the weaker the light, and the higher the light intensity required to ensure a clear image. and Inversely proportional, by introducing Dynamically adjust the light intensity to suit different surface reflectance characteristics. For example, increase the light intensity when shooting rough surfaces and reduce the light intensity when shooting smooth surfaces, thus ensuring a balanced brightness of the tunnel image.

[0037] The initial light brightness can be set according to the actual situation combined with expert experience, and the reflectivity of the tunnel structure surface can be obtained through experiments using a standard reference plate with known reflectivity (such as a white plate or a grayscale plate) as a calibration benchmark for measurement.

[0038] The specific steps and path planning for the mobile agent to drive the binocular camera to collect images of the target tunnel include: placing the agent at the entrance of the tunnel, centered on the cross section of the tunnel; adjusting the binocular camera so that its field of view covers the top and side walls of the tunnel, ensuring that the key feature areas are within the field of view; path description: driving in a straight line along the center line of the tunnel, from the entrance to the exit, the agent should pause at predetermined key nodes (such as tunnel turns, support change areas). Or the mobile agent can be planned for the path through intelligent algorithms, such as algorithms, etc.; the intelligent agent is equipped with obstacle avoidance sensors to automatically adjust the path to avoid possible obstacles. According to the real-time feedback data, the intelligent agent's travel speed and shooting frequency are adjusted to adapt to complex terrain or different tunnel conditions.

[0039] Step 4: After preprocessing, the collected target tunnel images are spliced, and the spliced ​​target tunnel images are input into the trained tunnel parameter prediction model, and the characteristic parameter prediction values ​​of the target tunnel are output, and the target tunnel finite element model is established based on the obtained characteristic parameter prediction values.

[0040] The finite element model of the target tunnel is established based on the predicted values ​​of the characteristic parameters of the output target tunnel. According to the design drawings, geological survey reports and relevant design specifications of the tunnel, the geometric parameters, surrounding rock types and support structure parameters of the tunnel are preliminarily determined to establish the finite element model. The model is then corrected and optimized through field data. Finite element software such as ABAQUS, ANSYS, MIDAS, etc. are used to directly generate the target tunnel model based on the geological and structural characteristics of the target tunnel through the modeling tools provided by the software (such as geometric drawing, parametric modeling, etc.), combined with the input characteristic parameter values. The direction of the tunnel is determined by the moving path of the movable intelligent body.

[0041] After the collected target tunnel images are preprocessed, they are stitched together. The logic of image stitching is as follows: a number of marker points are randomly set on the structural surface of the target tunnel, and the image is stitched together based on the marker points. Specifically, the SIFT algorithm in the feature point-based matching algorithm is used to extract the features of the marker points; a Gaussian difference pyramid is established by selecting the target area image, determining the position of the feature points, assigning directions to the feature points, and constructing descriptors for the determined feature points. The descriptor is a vector. The descriptor in the SIFT algorithm is generated by calculating the gradient direction histogram in the area around the feature points. After the landmark points of the image are extracted, the landmark point pairs are searched and matched. After the initial matching pairs are obtained, the double detection strategy coupling algorithm is used to eliminate the wrong matches, and the best matching results of the landmark points are obtained. The images are spliced ​​based on the matching results.

[0042] The SIFT (Size Invariant Feature Transform) algorithm is a computer vision processing algorithm, and its implementation steps are decomposed into: selecting a target area image to establish a Gaussian difference pyramid, constructing an image pyramid by Gaussian filtering at different scales, so as to detect features at different scales, and finding local extreme points by Gaussian difference in each scale space, which may correspond to key points in the image; determining the position of the feature point, and finding a stable key point position and scale by fitting the pixels around the extreme point on the basis of the local extreme point; assigning a direction to the feature point, and determining the direction of the key point by calculating the Gaussian gradient of the scale space to ensure that the feature is invariant to image rotation; constructing a descriptor for the determined feature point, and establishing a descriptor in the area around the key point, which encodes the local image structure, and the descriptor is usually based on the gradient direction histogram around the key point, as well as the gradient amplitude and direction of the feature point; using the distance or similarity measure (such as Euclidean distance or cosine similarity) between the descriptors to match the feature points, and after obtaining the initial matching pair, using the dual detection strategy coupling (PROSAC) algorithm to eliminate the wrong matching, and obtain the best matching result of the landmark point.

[0043] The descriptor is a vector that can uniquely represent the local appearance and characteristics of the feature point, and is generated by calculating the gradient direction histogram in the area around the feature point in the SIFT (Scale Invariant Feature Transform) algorithm.

[0044] After the feature points of the image are extracted, it is necessary to search and match the feature point pairs. After obtaining the initial matching pairs, the dual detection strategy coupling (PROSAC) algorithm is used to eliminate false matches and obtain the best matching results and centroid image coordinate positions of the landmark points. The (PROSAC) algorithm does not randomly extract samples, but first sorts the individuals according to the similarity of the feature matching point pairs, selects the inliers from the sorted matching points, and then determines the model parameters based on the inlier data. The more inliers, the better the model parameters. Through continuous hypothesis and verification, the best model parameters are finally obtained as the best matching solution.

[0045] The marking points can be designed as colored areas of different colors, shapes and sizes for easy identification and splicing.

[0046] Step 5: Based on the established finite element model of the target tunnel, the rock strength parameters of the target tunnel are analyzed and obtained. According to the rock strength parameters of the target tunnel and the predicted values ​​of the characteristic parameters of the target tunnel, a shear sliding index and a surrounding rock strength ratio index are generated. A comprehensive strength evaluation coefficient is generated through the shear sliding index and the surrounding rock strength ratio index. The strength of the target tunnel is judged based on the comprehensive strength evaluation coefficient. The rock strength parameters include rock compressive strength, rock tensile strength and rock shear strength.

[0047] According to the rock mass strength parameters of the target tunnel and the characteristic parameters of the target tunnel, the shear sliding index and the surrounding rock strength ratio index are generated. The formula for calculating the shear sliding index is: ; In the formula, is the shear sliding index, is the shear strength of rock mass, is the structural surface roughness coefficient, is the predicted value of the inclination angle of the target tunnel structure surface, is the normal stress of the target tunnel, is the proportional constant set; The rock mass strength parameters of the target tunnel are obtained by performing finite element analysis on the established finite element model of the target tunnel.

[0048] Among them, the shear sliding index Comprehensive rock mass shear strength and structural surface roughness coefficient, etc., are used to evaluate the ability of the structural surface to resist shear sliding, shear sliding index The larger the value is, the stronger the ability of the structural surface to resist shear sliding is, the more stable the tunnel structure is, and the higher the safety is.

[0049] Normal stress of target tunnel , is the normal stress acting on the structural surface in the normal direction, and its magnitude directly affects the friction and anti-slip ability. The higher the normal stress, the greater the friction between the structural surfaces, thereby improving the ability to resist shear sliding. Therefore, the higher the normal stress, the stronger the anti-slip ability. Therefore, the normal stress of the target tunnel is Shear Slip Index However, the higher the normal stress, the greater the pressure on the structural surface. Therefore, as the normal stress of the target tunnel increases The shear sliding index Also with the decrease, through Implementation, where the proportional constant is set Generally it is between 1 and 2.

[0050] Rock shear strength It is the inherent property of rock mass material, reflecting the ability of rock mass to resist shear failure. The higher the shear strength of rock mass, the stronger the ability of structural surface to resist shear sliding. Shear Slip Index is directly proportional to , and therefore acts as a direct proportionality factor in the formula.

[0051] Structural surface roughness coefficient It is a parameter that describes the degree of concavity and convexity of the tunnel structure surface. The larger its value is, the rougher the structure surface is. The rougher the structure surface is, the stronger the mechanical bite effect between the surfaces is, which significantly improves the shear-slip resistance. Therefore, the structure surface roughness coefficient is Shear Slip Index Proportional, using the logarithmic function The effect of roughness can be smoothed, and the contribution of roughness to the shear sliding index is adjusted to make its influence more stable.

[0052] Predicted value of the inclination angle of the target tunnel structure It is a parameter that describes the geometric state of the tunnel structure surface in space, reflecting the influence of the inclination angle of the structure surface on its ability to resist shear slip. The larger the inclination angle, the higher the sliding risk. It can be naturally described that the larger the inclination angle, the weaker the shear-slip resistance of the structural surface.

[0053] Among them, the structural surface roughness coefficient It is calculated by the predicted values ​​of the average waveform frequency and average surface fractal dimension of the target tunnel structure surface. The specific calculation formula is as follows: ; In the formula, is the predicted value of the average surface fractal dimension, is the predicted value of the average waveform frequency; Among them, the average waveform frequency prediction value It is the frequency of surface ripples, which is used to describe the regularity of surface concavity and convexity changes. The higher the frequency, the rougher the surface, so it is related to the structural surface roughness coefficient. Proportional to, and through the logarithmic function It means that when the predicted value of the average waveform frequency increases, the influence on the structural surface roughness coefficient gradually weakens.

[0054] Average surface fractal dimension prediction It is a dimensionless parameter that describes the complexity of the surface. The more complex and irregular the surface is, the higher the fractal dimension is. Therefore, it is related to the structural surface roughness coefficient. Proportional, through the square form Indicates the roughness coefficient of the structural surface significant contribution.

[0055] Normal stress of target tunnel Characterized by the target tunnel burial depth, the specific formula is: ; In the formula, is the unit weight of rock and soil, is the target tunnel depth, and are the predicted values ​​of the tunnel length and tunnel width of the target tunnel respectively. It reflects the influence of the structural surface inclination on the normal stress component: the larger the inclination, the smaller the normal stress. The target tunnel burial depth is obtained according to the actual exploration data report, and the tunnel burial depth is specifically the depth of the soil layer at the upper end of the tunnel.

[0056] The formula for calculating the surrounding rock strength ratio index is: ; In the formula, is the surrounding rock strength ratio index, and are the rock mass compressive strength and rock mass tensile strength, respectively. is the predicted value of the inclination of the target tunnel; Surrounding rock strength ratio index The comprehensive rock compressive strength, rock tensile strength and predicted inclination value of the target tunnel are used to characterize the overall strength reserve of the tunnel surrounding rock. The larger the value, the greater the strength of the target tunnel, the higher the stability and the higher the safety.

[0057] Among them, the higher the compressive strength, the greater the overall strength of the tunnel, the higher the tensile strength, the stronger the tunnel's ability to resist cracking, and therefore the corresponding tunnel stability and safety are higher. Therefore, the rock mass compressive strength and rock mass tensile strength to surrounding rock strength ratio index Proportional, through the form of square root sum Indicates the limiting effect on tunnel stability and safety.

[0058] Normal stress of target tunnel , is the normal stress acting on the structural surface in the normal direction, and its magnitude directly affects the friction and anti-slip ability. The higher the normal stress, the greater the pressure on the structural surface. Therefore, as the normal stress of the target tunnel increases, The normal stress of the target tunnel is Ratio index to surrounding rock strength Inversely proportional, and through the exponential function It means that the normal stress of the target tunnel As the pressure increases, the risk of collapse increases, so the surrounding rock strength ratio index Rapidly decrease.

[0059] Geomechanical studies have shown that the inclination angle of the surrounding rock has a direct impact on its stability. When the inclination angle is close to vertical, the tunnel is more susceptible to shear sliding or instability. When the inclination angle is close to horizontal, the tunnel is more stable. When the inclination angle is small, Close to 0, the tunnel strength reserve is higher and the stability is better. When the inclination angle is larger, Close to 1, the strength reserve of the surrounding rock is low and the stability is poor. Therefore, the inclination prediction value of the target tunnel is proportional to the surrounding rock strength index. Inversely proportional.

[0060] Predicted tunnel length of target tunnel The larger the value, the longer the tunnel, the larger the exposed area of ​​the surrounding rock, the wider the stress range, and the lower the overall strength reserve. Therefore, the predicted value of the tunnel length of the target tunnel is Ratio index to surrounding rock strength Inversely proportional, the inverse relationship is expressed by fraction.

[0061] The rock strength parameters include rock compressive strength, rock tensile strength and rock shear strength, which are obtained through finite element model combined with rock elastic modulus, Poisson's ratio, density and viscoplastic parameters to simulate the deformation and failure behavior of the rock under different stress conditions, based on finite element analysis experiments.

[0062] The comprehensive strength evaluation coefficient is generated by the shear sliding index and the surrounding rock strength ratio index, where the formula for calculating the comprehensive strength evaluation coefficient is: ; In the formula, is the comprehensive strength assessment coefficient, and are the weight coefficients of shear sliding index and surrounding rock strength ratio index, respectively. and and All are greater than 0; Since the larger the shear sliding index and the surrounding rock strength ratio index, the higher the stability and safety of the tunnel, the shear sliding index and the surrounding rock strength ratio index are proportional to the comprehensive strength evaluation coefficient. The continuity and smoothness of the formula make the formula have good stability and applicability under different input parameters, avoiding possible numerical mutation problems. When smaller, will grow rapidly, highlighting the impact of low-strength surrounding rock on comprehensive assessment. When larger, The value will tend to be flat, avoiding the high value of high-quality surrounding rock from over-dominantly affecting the evaluation results.

[0063] The shear slip index directly reflects the ability of the tunnel structure surface to resist shear slip and is an important control factor for tunnel stability and safety. In engineering practice, shear slip is often the main failure mode of tunnel instability. Even if the surrounding rock strength reserve is high, once shear slip occurs, the overall safety of the tunnel may still be seriously threatened. Therefore, the shear slip index should contribute more to the comprehensive strength evaluation coefficient. and and Both are greater than 0.

[0064] The strength of the target tunnel is determined based on the comprehensive strength evaluation coefficient, wherein the specific logic for determining the strength of the target tunnel is as follows: when When , the target tunnel strength is judged to be excellent, indicating that the target tunnel has no risk of collapse and sliding; when When , the target tunnel is judged to have good strength, indicating that the target tunnel has the risk of sliding and should be reinforced; when When the target tunnel strength is judged to be poor, it means that the target tunnel is at risk of collapse and is not suitable for entry; in It is the preset tunnel strength judgment threshold.

[0065] See also Figure 2 The present invention also provides a tunnel parameter modeling system based on binocular vision, and the tunnel parameter modeling system based on binocular vision is used to execute the above-mentioned tunnel parameter modeling method based on binocular vision, including: A sample image processing module is used to shoot several tunnels with known characteristic parameters based on a binocular camera, and preprocess the tunnel images obtained by shooting, and map the preprocessed tunnel images with corresponding characteristic parameters one by one to generate a training image set, wherein the preprocessing includes distortion correction and noise reduction enhancement, and the characteristic parameters include tunnel structural surface inclination, inclination, average waveform frequency, average surface fractal dimension, tunnel length and tunnel width; The parameter prediction model training module is used to establish a neural network model, using the tunnel image in the training image set as the input of the neural network model, and using the characteristic parameters corresponding to the tunnel image as labels to train the neural network model, so as to obtain a tunnel parameter prediction model whose input is the tunnel image and output is the tunnel characteristic parameters; The target tunnel image acquisition module is used to acquire images of the target tunnel by driving the binocular camera based on the mobile intelligent body, obtain a number of target tunnel images by means of compound acquisition, and pre-process them based on the same method; A tunnel model building module is used to splice the collected target tunnel images after preprocessing, input the spliced ​​target tunnel images into the trained tunnel parameter prediction model, output the characteristic parameter prediction values ​​of the target tunnel, and establish the target tunnel finite element model based on the obtained characteristic parameter prediction values; The tunnel strength judgment module is used to analyze and obtain the rock strength parameters of the target tunnel based on the established finite element model of the target tunnel, generate a shear sliding index and a surrounding rock strength ratio index according to the rock strength parameters of the target tunnel and the predicted values ​​of the characteristic parameters of the target tunnel, generate a comprehensive strength evaluation coefficient through the shear sliding index and the surrounding rock strength ratio index, and judge the strength of the target tunnel based on the comprehensive strength evaluation coefficient, wherein the rock strength parameters include rock compressive strength, rock tensile strength and rock shear strength.

[0066] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0067] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. Those skilled in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0068] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0069] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A tunnel parameter modeling method based on binocular vision, characterized in that: The specific steps include: Using a binocular camera to shoot several tunnels with known characteristic parameters, and preprocessing the tunnel images, mapping the preprocessed tunnel images with corresponding characteristic parameters one by one to generate a training image set, wherein the preprocessing includes distortion correction and noise reduction enhancement, and the characteristic parameters include tunnel structural surface inclination, inclination, average waveform frequency, average surface fractal dimension, tunnel length and tunnel width; A neural network model is established, the tunnel image in the training image set is used as the input of the neural network model, and the characteristic parameters corresponding to the tunnel image are used as labels to train the neural network model, so as to obtain a tunnel parameter prediction model whose input is the tunnel image and output is the tunnel characteristic parameters; Based on the mobile intelligent body driving the binocular camera to collect images of the target tunnel, several target tunnel images are obtained by compound acquisition, and they are pre-processed based on the same method; After preprocessing, the collected target tunnel images are spliced, the spliced ​​target tunnel images are input into the trained tunnel parameter prediction model, characteristic parameter prediction values ​​of the target tunnel are output, and a finite element model of the target tunnel is established based on the obtained characteristic parameter prediction values; Based on the established finite element model of the target tunnel, the rock strength parameters of the target tunnel are analyzed and obtained. According to the rock strength parameters of the target tunnel and combined with the predicted values ​​of the characteristic parameters of the target tunnel, a shear sliding index and a surrounding rock strength ratio index are generated. A comprehensive strength evaluation coefficient is generated through the shear sliding index and the surrounding rock strength ratio index. The strength of the target tunnel is judged based on the comprehensive strength evaluation coefficient. The rock strength parameters include rock compressive strength, rock tensile strength and rock shear strength.

2. The method for tunnel parameterization modeling based on binocular vision according to claim 1, characterized in that: Establish a neural network model, establish a neural network prediction model based on the LSTM model, select the activation function and optimization algorithm, among which the Tanh function is selected as the activation function, and Adam is selected as the optimization algorithm of the LSTM model; the formula of the Tanh function is: ； In the formula, Represents the Tanh function, independent variable Represents the weighted sum of the neuron's input, that is, the result of weighted summation of the input received by the neuron from the previous layer; At the same time, the hyperparameters of the LSTM model are set, and the hyperparameters of the LSTM model include: the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the batch processing number, and the number of hidden layer neurons; The number of network layers is set to a 3-layer network structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training times is set to 100, the batch size is set to 256, and the number of hidden layer neurons is 32.

3. The method for tunnel parameterization modeling based on binocular vision according to claim 2 is characterized in that: The collected tunnel image is preprocessed, and the preprocessing includes distortion correction and noise reduction enhancement, wherein the distortion correction includes radial distortion correction and tangential distortion correction, and the specific logic of radial distortion correction and tangential distortion correction is: establish three radial distortion coefficients, calibrated as , and , the formula for obtaining the ideal image coordinates without tangential distortion is: ； Establish two tangential distortion coefficients and calibrate them as and , the model expression without radial distortion is: ； By combining the two distortions, the effects of both can be eliminated at the same time. The parameter expression after superposition is: ； In the formula, , and are three different radial distortion coefficients, and are two different tangential distortion coefficients, represents the undistorted image coordinates under the ideal optical center system, is the image coordinate of a pixel in the captured image, is the distance from the calibration point to the imaging center, where the image coordinate system takes the principal point position coordinate as the image center, the origin is the intersection of the camera optical axis and the imaging plane, and the right is Axis positive direction, downward The positive direction of the axis is established, where the radial distortion coefficient, tangential distortion coefficient and principal point position coordinates are obtained through camera calibration; The method for performing noise reduction and enhancement processing on the tunnel image is as follows: using the wavelet transform denoising method to denoise the image after distortion correction, and using bilateral filtering to enhance the details of the tunnel image.

4. The method for tunnel parameterization modeling based on binocular vision according to claim 1, characterized in that: A mobile intelligent agent is used to carry a calibrated binocular camera to collect images of the target tunnel, where A * The algorithm generates the motion path of the mobile agent and dynamically corrects the brightness of the shot according to the environmental parameters in the target tunnel, where the environmental parameters include the ambient brightness and the horizontal distance between the tunnel structure surface and the binocular camera. The specific formula for dynamically correcting the brightness of the shot is: ； In the formula, is the corrected light intensity, is the initial light intensity, is the ambient brightness, is the horizontal distance between the tunnel structure surface and the binocular camera, is the reflectivity of the tunnel structure surface.

5. The method for tunnel parameterization modeling based on binocular vision according to claim 1, characterized in that: After the collected target tunnel images are preprocessed, they are stitched together. The logic of image stitching is as follows: a number of marker points are randomly set on the structural surface of the target tunnel, and the image is stitched together based on the marker points. Specifically, the SIFT algorithm in the feature point-based matching algorithm is used to extract the features of the marker points; a Gaussian difference pyramid is established by selecting the target area image, determining the position of the feature points, assigning directions to the feature points, and constructing descriptors for the determined feature points. The descriptor is a vector. The descriptor in the SIFT algorithm is generated by calculating the gradient direction histogram in the area around the feature points. After the landmark points of the image are extracted, the landmark point pairs are searched and matched. After the initial matching pairs are obtained, the double detection strategy coupling algorithm is used to eliminate the wrong matches, and the best matching results of the landmark points are obtained. The images are spliced ​​based on the matching results.

6. The method for tunnel parameterization modeling based on binocular vision according to claim 1, characterized in that: According to the rock mass strength parameters of the target tunnel and the characteristic parameters of the target tunnel, the shear sliding index and the surrounding rock strength ratio index are generated. The formula for calculating the shear sliding index is: ； In the formula, is the shear sliding index, is the shear strength of rock mass, is the structural surface roughness coefficient, is the predicted value of the inclination angle of the target tunnel structure surface, is the normal stress of the target tunnel, is the proportional constant set; Among them, the structural surface roughness coefficient It is calculated by the predicted values ​​of the average waveform frequency and average surface fractal dimension of the target tunnel structure surface. The specific calculation formula is as follows: ； In the formula, is the predicted value of the average surface fractal dimension, is the predicted value of the average waveform frequency; Normal stress of target tunnel Characterized by the target tunnel burial depth, the specific formula is: ； In the formula, is the unit weight of rock and soil, is the target tunnel depth, and are the predicted values ​​of the tunnel length and tunnel width of the target tunnel respectively.

7. The method for tunnel parameterization modeling based on binocular vision according to claim 6, characterized in that: The formula for calculating the surrounding rock strength ratio index is: ； In the formula, is the surrounding rock strength ratio index, and are the rock mass compressive strength and rock mass tensile strength, respectively. is the predicted value of the inclination of the target tunnel; The comprehensive strength evaluation coefficient is generated by the shear sliding index and the surrounding rock strength ratio index, where the formula for calculating the comprehensive strength evaluation coefficient is: ； In the formula, is the comprehensive strength assessment coefficient, and are the weight coefficients of shear sliding index and surrounding rock strength ratio index, respectively. and and All are greater than 0; The strength of the target tunnel is determined based on the comprehensive strength evaluation coefficient, wherein the specific logic for determining the strength of the target tunnel is as follows: when When , the target tunnel strength is judged to be excellent, indicating that the target tunnel has no risk of collapse and sliding; when When , the target tunnel is judged to have good strength, indicating that the target tunnel has the risk of sliding and should be reinforced; when When the target tunnel strength is judged to be poor, it means that the target tunnel is at risk of collapse and is not suitable for entry; in It is the preset tunnel strength judgment threshold.

8. A tunnel parameter modeling system based on binocular vision, characterized by: The binocular vision-based tunnel parameter modeling system is used to execute the binocular vision-based tunnel parameter modeling method according to any one of claims 1 to 7, comprising: A sample image processing module is used to shoot several tunnels with known characteristic parameters based on a binocular camera, and preprocess the tunnel images obtained by shooting, and map the preprocessed tunnel images with corresponding characteristic parameters one by one to generate a training image set, wherein the preprocessing includes distortion correction and noise reduction enhancement, and the characteristic parameters include tunnel structural surface inclination, inclination, average waveform frequency, average surface fractal dimension, tunnel length and tunnel width; The parameter prediction model training module is used to establish a neural network model, using the tunnel image in the training image set as the input of the neural network model, and using the characteristic parameters corresponding to the tunnel image as labels to train the neural network model, so as to obtain a tunnel parameter prediction model whose input is the tunnel image and output is the tunnel characteristic parameters; The target tunnel image acquisition module is used to acquire images of the target tunnel by driving the binocular camera based on the mobile intelligent body, obtain a number of target tunnel images by means of compound acquisition, and pre-process them based on the same method; A tunnel model building module is used to splice the collected target tunnel images after preprocessing, input the spliced ​​target tunnel images into the trained tunnel parameter prediction model, output the characteristic parameter prediction values ​​of the target tunnel, and establish the target tunnel finite element model based on the obtained characteristic parameter prediction values; The tunnel strength judgment module is used to analyze and obtain the rock strength parameters of the target tunnel based on the established finite element model of the target tunnel, generate a shear sliding index and a surrounding rock strength ratio index according to the rock strength parameters of the target tunnel and the predicted values ​​of the characteristic parameters of the target tunnel, generate a comprehensive strength evaluation coefficient through the shear sliding index and the surrounding rock strength ratio index, and judge the strength of the target tunnel based on the comprehensive strength evaluation coefficient, wherein the rock strength parameters include rock compressive strength, rock tensile strength and rock shear strength.

Images