Non-contact intelligent measurement method and system for average opening inside high-position dangerous rock mass crack

Abstract

The invention belongs to the field of slope engineering, and discloses a non-contact intelligent measurement method and system for the internal average opening degree of a high-position dangerous rock mass crack, and the method comprises the steps: obtaining the stress data of the surface of a high-position dangerous rock mass and the image data of the surface of the high-position dangerous rock mass, and carrying out the preprocessing of the image data of the surface of the high-position dangerous rock mass; according to the preprocessed surface image data of the high-position dangerous rock mass, using a convolutional neural network to identify and classify crack features, and according to the crack features, extracting crack contour coordinates; according to the extracted fracture contour coordinates, fracture displacement data are analyzed through a distance transformation algorithm, and an extended hyperbolic model used for describing the closing behavior of the rock fracture under the action of external force is constructed based on the fracture displacement data and stress data of the surface of the high-position dangerous rock mass; and calculating opening numerical values of different positions of the crack through the established extended hyperbolic model, and carrying out statistical analysis on all the opening numerical values by adopting a weighted average algorithm to obtain a final measurement result of the average opening in the crack.

Description

Technical Field

[0001] This invention belongs to the field of slope engineering, specifically relating to a non-contact intelligent measurement method and system for the average opening of tensile cracks inside high-altitude dangerous rock masses. Background Technology

[0002] Monitoring tension cracks in high-altitude unstable rock masses is a crucial aspect of geological disaster prevention and control, directly impacting the accuracy of rock mass stability assessments and disaster early warnings. Precise measurement of crack opening provides vital information on the risk of rock mass instability, playing an irreplaceable role in ensuring engineering safety and the safety of people's lives and property. However, existing measurement methods often face limitations in accuracy or operational difficulties in complex terrain and harsh environments. For example, traditional contact measurement is limited by terrain complexity and difficult to implement on high-altitude, steep rock masses. While non-contact methods based on remote sensing technology can cover large areas, they lack the fine detail required to capture crack contours, making it difficult to accurately reflect the dynamic changes in crack opening. These limitations make it difficult for existing technologies to meet the demands for high precision and real-time performance in practical engineering applications.

[0003] In non-contact measurement, accurately acquiring the crack profile is the primary technical challenge. Due to the complex terrain of high-altitude unstable rock masses, interference factors such as light, shadow, and rock surface texture make it difficult for optical equipment to reliably capture clear crack boundary information. For example, under strong light or cloudy weather, the image of the crack edge may be blurred, leading to deviations in feature point identification. This problem further makes it difficult to accurately establish the mathematical mapping relationship between crack profile and internal aperture. Because of the lack of stable profile data, the measurement results often have large errors when calculating the aperture values ​​at different locations of the crack, especially when the crack morphology is complex or the width varies drastically. The statistical analysis results of the average aperture may be distorted, affecting the reliability of subsequent risk assessments.

[0004] Therefore, how to stably capture the crack profile features using high-precision optical equipment in complex environments and establish an accurate mathematical mapping relationship between the profile and the internal aperture has become a key issue in the non-contact measurement of the average aperture inside tensile cracks in high-altitude dangerous rock masses. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a non-contact intelligent measurement method and system for the average aperture of tensile cracks inside high-altitude unstable rock masses. By integrating multiple technologies, it achieves high-precision automated measurement of cracks in complex environments, significantly improving the reliability and efficiency of unstable rock mass stability assessment, and providing important technical support for geological disaster prevention and control.

[0006] To achieve the above objectives, the present invention provides the following solution: A non-contact intelligent measurement method for the average aperture inside tensile fractures in high-altitude unstable rock masses, the method comprising: Acquire stress data and image data of the surface of the high-level dangerous rock mass, and preprocess the image data of the surface of the high-level dangerous rock mass; Based on the preprocessed surface image data of the high-altitude unstable rock mass, a convolutional neural network was used to identify and classify crack features, and crack contour coordinates were extracted based on the crack features. Based on the extracted crack contour coordinates, the crack displacement data is analyzed using a distance transformation algorithm. Based on the crack displacement data and the stress data of the surface of the high-altitude dangerous rock mass, an extended hyperbolic model is constructed to describe the closing behavior of rock cracks under external forces. The aperture values ​​at different locations of the crack were calculated using an extended hyperbolic model. A weighted average algorithm was then used to perform statistical analysis on all aperture values ​​to obtain the final measurement result of the average aperture inside the crack.

[0007] Preferably, the method for obtaining stress data and image data of the surface of the high-level unstable rock mass includes: The impact energy generated by the air-blast detonator disturbs the high-level unstable rock mass, causing displacement of the rock mass in the normal direction of the crack, which in turn causes fluctuations in the opening of the tensile crack. The normal stress perpendicular to the crack plane is determined by the explosive equivalent of the detonator and the distance between the detonation location and the unstable rock mass. The image data of the crack is obtained by a high-definition camera mounted on a drone.

[0008] Preferably, the method for preprocessing image data of high-altitude unstable rock masses includes: Multispectral imaging technology was used to perform spectral analysis on the original image set to determine areas of illumination variation and shadow interference. If changes in illumination or areas of shadow interference are detected, a corrected image is generated using a multispectral data fusion algorithm to obtain a denoised image set. Based on the denoised image set, an image normalization algorithm is used to adjust the image brightness and contrast to generate a normalized image sequence.

[0009] Preferably, the method for identifying and classifying crack features using a convolutional neural network based on preprocessed high-altitude unstable rock mass surface image data, and extracting crack contour coordinates based on crack features, includes: Based on the principles of deep convolutional networks, large convolutional kernels are used to classify and locate crack images, and a lightweight semantic segmentation MobileNetv2-GCN model is constructed based on crack features. The GCN module adopts the concept of GoogleNet, splitting large convolutional kernels into convolutional combinations. The size of the convolutional kernel is determined by parameters. Control, use and Convolutional combinations; The crack image global segmentation model uses a pre-trained ResNet152 as the feature network and a fully convolutional improved network FCN4 as the segmentation framework, which are fused to form the ResNet-GCN model. Multi-scale feature maps are extracted from different stages of the feature network. The global convolutional network structure generates multi-scale semantic score maps for each class of image. The low-resolution score map is upsampled by the deconvolution layer and then superimposed with the high-resolution score map to generate a new score map. After the last upsampling, the final semantic score map is generated and the prediction result is output. Refine the boundary The module is integrated into the network and trained end-to-end. Boundary alignment is modeled as a residual structure. The boundary refinement BR module uses small convolutional kernels to extract detailed semantic information, obtain the spatial structural details of the crack, and reuses the boundary refinement module. The module aims to refine crack boundary prediction, resulting in a refined fractional map. The expression is: In the formula, This is a coarse fractional graph. This is a residual branch.

[0010] Preferably, a method for constructing an extended hyperbolic model to describe the closure behavior of rock fractures under external forces, based on fracture displacement data and stress data on the surface of high-lying unstable rock masses, includes: ; in, σ n The normal stress is perpendicular to the plane of the crack. u m This is the maximum normal displacement. k ni For the initial normal stiffness, n To adjust the parameters; Normal stiffness of the crack k n The calculation formula is as follows: .

[0011] The present invention also provides a non-contact intelligent measurement system for the average aperture inside tensile fractures in high-altitude unstable rock masses. The system is used to implement the aforementioned method and includes: an acquisition module, an extraction module, a construction module, and an analysis module. The acquisition module is used to acquire stress data and image data of the surface of the high-level dangerous rock mass, and to preprocess the image data of the surface of the high-level dangerous rock mass. The extraction module is used to identify and classify crack features using a convolutional neural network based on the preprocessed surface image data of the high-altitude dangerous rock mass, and to extract crack contour coordinates based on the crack features. The construction module is used to analyze crack displacement data using a distance transformation algorithm based on the extracted crack contour coordinates, and to construct an extended hyperbolic model to describe the closing behavior of rock fissures under external forces based on the crack displacement data and stress data on the surface of the high-altitude dangerous rock mass. The analysis module is used to calculate the aperture value at different locations of the crack using the established extended hyperbolic model, and to perform statistical analysis on all aperture values ​​using a weighted average algorithm to obtain the final measurement result of the average aperture inside the crack.

[0012] Preferably, the process of acquiring stress data and image data of the surface of the high-level unstable rock mass includes: The impact energy generated by the air-blast detonator disturbs the high-level unstable rock mass, causing displacement of the rock mass in the normal direction of the crack, which in turn causes fluctuations in the opening of the tensile crack. The normal stress perpendicular to the crack plane is determined by the explosive equivalent of the detonator and the distance between the detonation location and the unstable rock mass. The image data of the crack is obtained by a high-definition camera mounted on a drone.

[0013] Preferably, the acquisition module includes: a spectral analysis unit, a calibration unit, and an adjustment unit; The spectral analysis unit is used to perform spectral analysis on the original image set using multispectral imaging technology to determine areas of illumination variation and shadow interference. The correction unit is used to generate a corrected image through a multispectral data fusion algorithm to obtain a denoised image set if changes in illumination or shadow interference areas are detected. The adjustment unit is used to adjust the brightness and contrast of images based on the denoised image set using an image normalization algorithm to generate a normalized image sequence.

[0014] Preferably, based on the preprocessed surface image data of the high-altitude unstable rock mass, a convolutional neural network is used to identify and classify crack features. The process of extracting crack contour coordinates based on these features includes: Based on the principles of deep convolutional networks, large convolutional kernels are used to classify and locate crack images, and a lightweight semantic segmentation MobileNetv2-GCN model is constructed based on crack features. The GCN module adopts the concept of GoogleNet, splitting large convolutional kernels into convolutional combinations. The size of the convolutional kernel is determined by parameters. Control, use and Convolutional combinations; The crack image global segmentation model uses a pre-trained ResNet152 as the feature network and a fully convolutional improved network FCN4 as the segmentation framework, which are fused to form the ResNet-GCN model. Multi-scale feature maps are extracted from different stages of the feature network. The global convolutional network structure generates multi-scale semantic score maps for each class of image. The low-resolution score map is upsampled by the deconvolution layer and then superimposed with the high-resolution score map to generate a new score map. After the last upsampling, the final semantic score map is generated and the prediction result is output. Refine the boundary The module is integrated into the network and trained end-to-end. Boundary alignment is modeled as a residual structure. The boundary refinement BR module uses small convolutional kernels to extract detailed semantic information, obtain the spatial structural details of the crack, and reuses the boundary refinement module. The module aims to refine crack boundary prediction, resulting in a refined fractional map. The expression is: In the formula, This is a coarse fractional graph. This is a residual branch.

[0015] Preferably, the process of constructing an extended hyperbolic model to describe the closure behavior of rock fractures under external forces, based on fracture displacement data and stress data on the surface of high-lying unstable rock masses, includes: ; in, σ n The normal stress is perpendicular to the plane of the crack. u m This is the maximum normal displacement. k ni For the initial normal stiffness, n To adjust the parameters; Normal stiffness of the crack k n The calculation formula is as follows: .

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1) Based on the independently developed hyperbolic contact model for rock mass fracture propagation and combined with UAV photography technology, this invention proposes an innovative method for measuring the opening information inside tensile fractures in high-altitude dangerous rock masses; 2) This invention proposes an indirect method for measuring the internal opening of fractures by considering the closure law of tensile fractures in unstable rock masses under normal load, thus avoiding the interference of rock block obstruction on other direct measurement methods. 3) By integrating multiple technologies, high-precision automated measurement of cracks in complex environments has been achieved, which significantly improves the reliability and efficiency of stability assessment of unstable rock masses and provides important technical support for geological disaster prevention and control. Attached Figure Description

[0017] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the 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.

[0018] Figure 1 This is a schematic diagram of a non-contact intelligent measurement method for the average aperture of tensile fractures in high-altitude dangerous rock masses, as described in an embodiment of the present invention. Figure 2 This is an overall structural diagram of the ResNet-GCN model according to an embodiment of the present invention, wherein (a) is a structural schematic diagram of the entire framework; (b) is a structural schematic diagram of the GCN; and (c) is a schematic diagram of the boundary refinement module. Figure 3 This is a schematic diagram of the stress-displacement extended hyperbolic model according to an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] Example 1 like Figure 1 As shown, this invention provides a non-contact intelligent measurement method for the average aperture inside tensile fractures in high-altitude unstable rock masses, the method comprising: Acquire stress data and image data of the surface of the high-level dangerous rock mass, and preprocess the image data of the surface of the high-level dangerous rock mass; Based on the preprocessed surface image data of the high-altitude unstable rock mass, a convolutional neural network was used to identify and classify crack features, and crack contour coordinates were extracted based on the crack features. Based on the extracted crack contour coordinates, the crack displacement data is analyzed using a distance transformation algorithm. Based on the crack displacement data and the stress data of the surface of the high-altitude dangerous rock mass, an extended hyperbolic model is constructed to describe the closing behavior of rock cracks under external forces. The aperture values ​​at different locations of the crack were calculated using an extended hyperbolic model. A weighted average algorithm was then used to perform statistical analysis on all aperture values ​​to obtain the final measurement result of the average aperture inside the crack.

[0022] In this embodiment, the method for obtaining stress data and image data of the surface of a high-level unstable rock mass includes: After completing the initial identification of the unstable rock mass, detonators were deployed to induce ground motion based on local geological parameters. Detonating the detonators generated vibrations that exerted forces on the unstable rock mass and the subsequent slope. This force caused displacement of the unstable rock mass in the normal direction of the fractures, thus inducing fluctuations in the opening of the tension fractures. The normal stress perpendicular to the fracture plane... σ n The explosive equivalent of the detonator and the distance between the detonation location and the unstable rock mass are determined. The normal displacement of the fissure is estimated by the change in the opening of the fissure outcrop measured by a high-definition camera mounted on a drone.

[0023] Specifically: Crack normal displacement: ; in, The normal displacement of the crack (m) The normal stress (Pa) caused by the explosion wave, Normal stiffness of the crack ( (), estimated by experiment or experience.

[0024] Estimation method (based on explosive yield): ; in, Explosion yield ( TNT), For the distance of the bursting center , Rock mass wave impedance , Peak particle velocity (PPV) ), It is a direction factor (0.1-0.3, depending on the angle between the crack and the wave propagation direction).

[0025] Fracture aperture was extracted using high-resolution drone images and SfM photogrammetry. Establish fracture aperture variation and Geometric relationships: ; in, The angle between the fracture surface and the camera's line of sight (extracted from the 3D model).

[0026] Install a controllable detonator array on the slope; High-speed cameras and drones are used to simultaneously monitor crack opening; Establish Empirical regression model: ; in, The initial closure stress of the crack , The crack flexibility coefficient .

[0027] In this embodiment, the method for preprocessing surface image data of high-altitude unstable rock masses includes: Multispectral imaging technology was used to perform spectral analysis on the original image set to determine areas of illumination variation and shadow interference. If changes in illumination or areas of shadow interference are detected, a corrected image is generated using a multispectral data fusion algorithm to obtain a denoised image set. Based on the denoised image set, an image normalization algorithm is used to adjust the image brightness and contrast to generate a normalized image sequence.

[0028] In this embodiment, based on the preprocessed surface image data of the high-altitude unstable rock mass, a convolutional neural network is used to identify and classify crack features. The method for extracting crack contour coordinates based on these crack features includes: Based on the principles of deep convolutional networks, large convolutional kernels are used to classify and locate crack images, and a lightweight semantic segmentation MobileNetv2-GCN model is constructed based on crack features. As shown in Figure 2(b), the GCN module does not directly use larger convolutional kernels. Instead, it adopts the idea of ​​GoogleNet, splitting large convolutional kernels into convolutional combinations. The size of the convolutional kernel is determined by parameters. Control, use and Convolutional combinations; resulting in a larger feature map Close regional connectivity becomes possible. Compared to ordinary... Compared to convolution, GCN structures have fewer parameters, which makes them more practical for larger convolution kernel sizes.

[0029] The overall segmentation model for crack images is shown in Figure 2(a). It uses a pre-trained ResNet 152 as the feature network and a fully convolutional network (FCN4) as the segmentation framework, fusing them to form the ResNet-GCN model. Multi-scale feature maps are extracted from different stages of the feature network. The global convolutional network structure generates multi-scale semantic score maps for each image class. Low-resolution score maps are upsampled by deconvolutional layers and then superimposed with high-resolution score maps to generate new score maps. After a final upsampling, the final semantic score map is generated, and the prediction result is output.

[0030] Accurate boundary identification and segmentation are crucial steps in image segmentation. Common boundary prediction methods include Conditional Random Fields (CRF), and later derivatives such as DenseCRF and CRFAsRNN. DenseCRF is a CRF variant built on fully connected graphs, serving as a post-deep convolution processing method. CRFAsRNN models DenseCRF as RNN-style operators and proposes an end-to-end processing procedure, but it involves excessive CPU computation. To improve the accuracy of crack boundary identification, a boundary refinement (BR) module is added. Unlike CRF-based processing methods, the experiment will... The module is integrated into the network and trained end-to-end, as shown in Figure 2(c). Unlike the GCN module, which uses large convolutional kernels to obtain the overall semantic connectivity features of the region, the experiment models boundary alignment as a residual structure. The BR module uses small convolutional kernels to extract richer detailed semantic information, obtain more details of the crack spatial structure, and reuse them. This module aims to improve the resolution of crack feature maps, thereby refining crack boundary prediction. (Refined fractional map) The expression is: In the formula, This is a coarse fractional graph. This is a residual branch.

[0031] In this embodiment, the method for constructing an extended hyperbolic model to describe the closure behavior of rock fissures under external forces, based on crack displacement data and stress data on the surface of high-altitude unstable rock masses, includes: By measuring the stress state and displacement data of the tensile fractures in the unstable rock mass, the normal stress-displacement curve of the fracture was plotted. Combined with the extended hyperbolic model of fracture closure, the opening-closing curve of the entire fracture process was inferred, thus determining the average aperture of the fracture under steady-state conditions. The extended hyperbolic model is as follows: ; in, σn The normal stress is perpendicular to the plane of the crack. u m This is the maximum normal displacement. k ni For the initial normal stiffness, n To adjust the parameters; Normal stiffness of the crack k n The calculation formula is as follows: .

[0032] In this model, adjust the parameters n Due to the influence of rock mechanical properties, laboratory tests should be carried out simultaneously in this method. n Determination: Rock blocks with the same lithology as the unstable rock mass to be tested are collected and prepared into standard cubic rock fracture specimens through cutting and splitting methods. Normal compression tests are conducted, and stress-normal displacement (closure) curves are plotted to determine the corresponding adjustment parameters for the tensile fractures of the unstable rock mass. n .

[0033] By using the time history curves of the disturbance load obtained during the test and the observed real-time closure quantities, a plot can be drawn as follows: Figure 3 As shown, this is a segment of the crack normal displacement-stress curve. Furthermore, the extended hyperbolic model is used to determine the maximum normal displacement, which is difficult to obtain directly in actual field testing. u m This is equivalent to the internal characteristic opening degree of the tensile fractures in the unstable rock mass.

[0034] Example 2 The present invention also provides a non-contact intelligent measurement system for the average aperture inside tensile fractures in high-altitude unstable rock masses. The system is used to implement the method described in Embodiment 1. The system includes: an acquisition module, an extraction module, a construction module, and an analysis module. The acquisition module is used to acquire stress data and image data of the surface of the high-level dangerous rock mass, and to preprocess the image data of the surface of the high-level dangerous rock mass. The extraction module is used to identify and classify crack features based on the preprocessed surface image data of the high-altitude dangerous rock mass using a convolutional neural network, and extract the crack contour coordinates based on the crack features. The module is used to analyze crack displacement data using a distance transformation algorithm based on the extracted crack contour coordinates. Based on the crack displacement data and stress data on the surface of the high-altitude dangerous rock mass, an extended hyperbolic model is constructed to describe the closing behavior of rock cracks under external forces. The analysis module is used to calculate the aperture value at different locations of the crack using the established extended hyperbolic model. It then uses a weighted average algorithm to perform statistical analysis on all aperture values ​​to obtain the final measurement result of the average aperture inside the crack.

[0035] In this embodiment, the process of acquiring stress data and image data of the surface of the high-level unstable rock mass includes: The impact energy generated by the air-blast detonator disturbs the high-level unstable rock mass, causing displacement of the rock mass in the normal direction of the crack, which in turn causes fluctuations in the opening of the tensile crack. The normal stress perpendicular to the crack plane is determined by the explosive equivalent of the detonator and the distance between the detonation location and the unstable rock mass. The image data of the crack is obtained by a high-definition camera mounted on a drone.

[0036] In this embodiment, the acquisition module includes: a spectral analysis unit, a calibration unit, and an adjustment unit; The spectral analysis unit is used to perform spectral analysis on the original image set using multispectral imaging technology to determine areas of illumination variation and shadow interference. The correction unit is used to generate a corrected image through a multispectral data fusion algorithm if changes in illumination or shadow interference areas are detected, thereby obtaining a set of denoised images. The adjustment unit is used to adjust the brightness and contrast of images based on the denoised image set using an image normalization algorithm to generate a normalized image sequence.

[0037] In this embodiment, based on the preprocessed surface image data of the high-altitude unstable rock mass, a convolutional neural network is used to identify and classify crack features. The process of extracting crack contour coordinates based on these features includes: Based on the principles of deep convolutional networks, large convolutional kernels are used to classify and locate crack images, and a lightweight semantic segmentation MobileNetv2-GCN model is constructed based on crack features. The GCN module adopts the concept of GoogleNet, splitting large convolutional kernels into convolutional combinations. The size of the convolutional kernel is determined by parameters. Control, use and Convolutional combinations; The crack image global segmentation model uses a pre-trained ResNet152 as the feature network and a fully convolutional improved network FCN4 as the segmentation framework, which are fused to form the ResNet-GCN model. Multi-scale feature maps are extracted from different stages of the feature network. The global convolutional network structure generates multi-scale semantic score maps for each class of image. The low-resolution score map is upsampled by the deconvolution layer and then superimposed with the high-resolution score map to generate a new score map. After the last upsampling, the final semantic score map is generated and the prediction result is output. Refine the boundary The module is integrated into the network and trained end-to-end. Boundary alignment is modeled as a residual structure. The boundary refinement BR module uses small convolutional kernels to extract detailed semantic information, obtain the spatial structural details of the crack, and reuses the boundary refinement module. The module aims to refine crack boundary prediction, resulting in a refined fractional map. The expression is: In the formula, This is a coarse fractional graph. This is a residual branch.

[0038] In this embodiment, the process of constructing an extended hyperbolic model to describe the closure behavior of rock fissures under external forces, based on crack displacement data and stress data on the surface of high-altitude unstable rock masses, includes: ; in, σ n The normal stress is perpendicular to the plane of the crack. u m This is the maximum normal displacement. k ni For the initial normal stiffness, n To adjust the parameters; Normal stiffness of the crack k n The calculation formula is as follows: .

[0039] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A non-contact intelligent measurement method for the average aperture of tensile fractures inside high-altitude unstable rock masses, characterized in that, The method includes: Acquire stress data and image data of the surface of the high-level dangerous rock mass, and preprocess the image data of the surface of the high-level dangerous rock mass; Based on the preprocessed surface image data of the high-altitude unstable rock mass, a convolutional neural network was used to identify and classify crack features, and crack contour coordinates were extracted based on the crack features. Based on the extracted crack contour coordinates, the crack displacement data is analyzed using a distance transformation algorithm. Based on the crack displacement data and the stress data of the surface of the high-altitude dangerous rock mass, an extended hyperbolic model is constructed to describe the closing behavior of rock cracks under external forces. The aperture values ​​at different locations of the crack were calculated using an extended hyperbolic model. A weighted average algorithm was then used to perform statistical analysis on all aperture values ​​to obtain the final measurement result of the average aperture inside the crack.

2. The method according to claim 1, characterized in that, Methods for obtaining stress data and image data of the surface of high-lying unstable rock masses include: The impact energy generated by the air-blast detonator disturbs the high-level unstable rock mass, causing displacement of the rock mass in the normal direction of the crack, which in turn causes fluctuations in the opening of the tensile crack. The normal stress perpendicular to the crack plane is determined by the explosive equivalent of the detonator and the distance between the detonation location and the unstable rock mass. The image data of the crack is obtained by a high-definition camera mounted on a drone.

3. The method according to claim 1, characterized in that, Methods for preprocessing surface image data of high-altitude unstable rock masses include: Multispectral imaging technology was used to perform spectral analysis on the original image set to determine areas of illumination variation and shadow interference. If changes in illumination or areas of shadow interference are detected, a corrected image is generated using a multispectral data fusion algorithm to obtain a denoised image set. Based on the denoised image set, an image normalization algorithm is used to adjust the image brightness and contrast to generate a normalized image sequence.

4. The method according to claim 1, characterized in that, Based on the preprocessed surface image data of the high-altitude unstable rock mass, a convolutional neural network was used to identify and classify crack features. The method for extracting crack contour coordinates based on these features includes: Based on the principles of deep convolutional networks, large convolutional kernels are used to classify and locate crack images, and a lightweight semantic segmentation MobileNetv2-GCN model is constructed based on crack features. The GCN module adopts the concept of GoogleNet, splitting large convolutional kernels into convolutional combinations. The size of the convolutional kernel is determined by parameters. Control, use and Convolutional combinations; The crack image global segmentation model uses a pre-trained ResNet152 as the feature network and a fully convolutional improved network FCN4 as the segmentation framework, which are fused to form the ResNet-GCN model. Multi-scale feature maps are extracted from different stages of the feature network. The global convolutional network structure generates multi-scale semantic score maps for each class of image. The low-resolution score map is upsampled by the deconvolution layer and then superimposed with the high-resolution score map to generate a new score map. After the last upsampling, the final semantic score map is generated and the prediction result is output. Refine the boundary The module is integrated into the network and trained end-to-end. Boundary alignment is modeled as a residual structure. The boundary refinement BR module uses small convolutional kernels to extract detailed semantic information, obtain the spatial structural details of the crack, and reuses the boundary refinement module. The module aims to refine crack boundary prediction, resulting in a refined fractional map. The expression is: In the formula, This is a coarse fractional graph. This is a residual branch.

5. The method according to claim 1, characterized in that, Methods for constructing an extended hyperbolic model to describe the closure behavior of rock fractures under external forces, based on fracture displacement data and stress data on the surface of high-lying unstable rock masses, include: ； in, σ n The normal stress is perpendicular to the plane of the crack. u m This is the maximum normal displacement. k ni For the initial normal stiffness, n To adjust the parameters; Normal stiffness of the crack k n The calculation formula is as follows: 。 6. A non-contact intelligent measurement system for the average aperture of tensile fractures inside a high-altitude unstable rock mass, the system being used to implement the method described in any one of claims 1-5, characterized in that... The system includes: an acquisition module, an extraction module, a construction module, and an analysis module; The acquisition module is used to acquire stress data and image data of the surface of the high-level dangerous rock mass, and to preprocess the image data of the surface of the high-level dangerous rock mass. The extraction module is used to identify and classify crack features using a convolutional neural network based on the preprocessed surface image data of the high-altitude dangerous rock mass, and to extract crack contour coordinates based on the crack features. The construction module is used to analyze crack displacement data using a distance transformation algorithm based on the extracted crack contour coordinates, and to construct an extended hyperbolic model to describe the closing behavior of rock fissures under external forces based on the crack displacement data and stress data on the surface of high-altitude dangerous rock mass. The analysis module is used to calculate the aperture value at different locations of the crack using the established extended hyperbolic model, and to perform statistical analysis on all aperture values ​​using a weighted average algorithm to obtain the final measurement result of the average aperture inside the crack.

7. The system according to claim 6, characterized in that, The process of acquiring stress data and image data of the surface of high-lying unstable rock masses includes: The impact energy generated by the air-blast detonator disturbs the high-level unstable rock mass, causing displacement of the rock mass in the normal direction of the crack, which in turn causes fluctuations in the opening of the tensile crack. The normal stress perpendicular to the crack plane is determined by the explosive equivalent of the detonator and the distance between the detonation location and the unstable rock mass. The image data of the crack is obtained by a high-definition camera mounted on a drone.

8. The system according to claim 6, characterized in that, The acquisition module includes: a spectral analysis unit, a calibration unit, and an adjustment unit; The spectral analysis unit is used to perform spectral analysis on the original image set using multispectral imaging technology to determine areas of illumination variation and shadow interference. The correction unit is used to generate a corrected image through a multispectral data fusion algorithm to obtain a denoised image set if changes in illumination or shadow interference areas are detected. The adjustment unit is used to adjust the brightness and contrast of images based on the denoised image set using an image normalization algorithm to generate a normalized image sequence.

9. The system according to claim 6, characterized in that, Based on the preprocessed surface image data of the high-altitude unstable rock mass, a convolutional neural network is used to identify and classify crack features. The process of extracting crack contour coordinates based on crack features includes: Based on the principles of deep convolutional networks, large convolutional kernels are used to classify and locate crack images, and a lightweight semantic segmentation MobileNetv2-GCN model is constructed based on crack features. The GCN module adopts the concept of GoogleNet, splitting large convolutional kernels into convolutional combinations. The size of the convolutional kernel is determined by parameters. Control, use and Convolutional combinations; The crack image global segmentation model uses a pre-trained ResNet152 as the feature network and a fully convolutional improved network FCN4 as the segmentation framework, which are fused to form the ResNet-GCN model. Multi-scale feature maps are extracted from different stages of the feature network. The global convolutional network structure generates multi-scale semantic score maps for each class of image. The low-resolution score map is upsampled by the deconvolution layer and then superimposed with the high-resolution score map to generate a new score map. After the last upsampling, the final semantic score map is generated and the prediction result is output. Refine the boundary The module is integrated into the network and trained end-to-end. Boundary alignment is modeled as a residual structure. The boundary refinement BR module uses small convolutional kernels to extract detailed semantic information, obtain the spatial structural details of the crack, and reuses the boundary refinement module. The module aims to refine crack boundary prediction, resulting in a refined fractional map. The expression is: In the formula, This is a coarse fractional graph. This is a residual branch.

10. The system according to claim 6, characterized in that, The process of constructing an extended hyperbolic model to describe the closure behavior of rock fractures under external forces, based on fracture displacement data and stress data on the surface of high-altitude unstable rock masses, includes: ； in, σ n The normal stress is perpendicular to the plane of the crack. u m This is the maximum normal displacement. k ni For the initial normal stiffness, n To adjust the parameters; Normal stiffness of the crack k n The calculation formula is as follows: 。

Images