Working Face Fracture Surrounding Rock Identification System
By combining a lightweight deep convolutional neural network with a traditional edge detection algorithm, and utilizing multi-scale decomposition and adaptive illumination correction techniques, the real-time performance and accuracy issues of crack identification in tunnel construction were resolved, achieving efficient crack identification and judgment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHONG JIAO YI GONG JU QIAO SUI GONG CHENG YOU XIAN GONG SI
- Filing Date
- 2026-03-26
- Publication Date
- 2026-05-26
AI Technical Summary
Existing technologies struggle to simultaneously achieve high-precision identification and real-time judgment of face cracks in complex construction environments. Traditional methods are susceptible to environmental interference and have poor fitting effects, while deep learning methods involve large computational loads and cannot meet real-time requirements.
Image data is acquired using a high-definition industrial camera, an LED ring light source, and an environmental sensor. Combined with a lightweight deep convolutional neural network and a traditional edge detection algorithm, the fine localization and complete connection of the region of interest in the crack are achieved through multi-scale decomposition, adaptive illumination correction, adaptive threshold edge detection, and Hough transform.
In complex construction environments, it achieves the unification of high-precision fracture identification and real-time judgment, reduces computational complexity, and improves the practicality and reliability of intelligent geological logging technology.
Smart Images
Figure CN122089706A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of geological exploration technology for tunnels and underground engineering, and in particular to a system for identifying and judging fractured surrounding rock at the tunnel face. Background Technology
[0002] In tunnel construction, the tunnel face refers to the fresh rock mass exposed during tunnel excavation. The fissures developed on its surface are key geological information for evaluating the stability of the surrounding rock, determining support parameters, and predicting geological hazards. Accurately identifying the development characteristics of fissures at the tunnel face is of significant engineering importance for guiding safe tunnel construction and optimizing support design.
[0003] Currently, the identification and judgment of face fractures mainly relies on two technical approaches. One is based on traditional digital image processing. This method first performs illumination correction preprocessing on the acquired face images, then uses edge detection operators such as Canny and Sobel to extract the edges of rock mass structural surfaces, then filters out false edge noise through mathematical morphology processing, and uses Hough transform to fit straight edges. Finally, it achieves local edge connection based on distance and angle criteria, thereby obtaining complete structural surface traces. The other is an image segmentation method based on deep learning. This method constructs a semantic segmentation dataset containing two types of labels: fractures and background. It uses deep convolutional neural network models such as DeepLabv3+ for pixel-level fracture segmentation, and uses an encoder-decoder structure to automatically learn fracture features, achieving end-to-end fracture identification.
[0004] However, existing technologies have significant limitations in practical engineering applications. While traditional image processing methods offer high computational efficiency, their edge detection results are easily affected by factors such as complex rock surface textures, uneven lighting, and interference from dust and water mist, generating a large amount of unstructured surface noise. Furthermore, the Hough transform performs poorly in fitting non-linear cracks, making it difficult to guarantee the completeness and accuracy of crack identification. Although deep learning-based methods can achieve pixel-level fine segmentation, their model training relies on large-scale, high-quality labeled datasets. However, the tunnel construction site environment is complex and variable, with dust concentration, lighting conditions, and water mist interference severely affecting image quality, resulting in insufficient generalization ability of the model in real-world engineering environments. Moreover, deep learning models have high computational requirements, making it difficult to meet the real-time identification needs of construction sites. Therefore, existing technologies struggle to simultaneously achieve high-precision and real-time identification of face cracks in complex construction environments, hindering the widespread application of intelligent geological logging technology in tunnel engineering. Summary of the Invention
[0005] The purpose of this invention is to provide a system for identifying and judging fractured surrounding rock at the working face, so as to solve the technical problem that existing technologies are unable to simultaneously achieve both accuracy and real-time judgment of fracture identification in complex construction environments.
[0006] This invention provides a system for identifying and judging fractured surrounding rock at the tunnel face, comprising: An image acquisition module is used to acquire image data of the rock surface in the tunnel face construction environment. The image acquisition module includes a high-definition industrial camera, an LED ring light source, and an environmental sensor. The LED ring light source is set around the lens of the high-definition industrial camera. The environmental sensor is integrated with the high-definition industrial camera. The high-definition industrial camera, the LED ring light source, and the environmental sensor are all electrically connected to the main control module. The main control module is used to receive image data and environmental parameters transmitted by the image acquisition module, and to preprocess the image data and identify and judge the cracks. The main control module includes an image preprocessing unit, a crack feature extraction unit and a crack identification unit. The image preprocessing unit, the crack feature extraction unit and the crack identification unit are electrically connected in sequence. The main control module is electrically connected to the storage module and the display module respectively. A storage module is used to store intermediate data and final result data during the crack identification and judgment process. The storage module is electrically connected to the main control module. The display module is used to display the crack identification and judgment results in real time, and the display module is electrically connected to the main control module. The main control module is configured to perform the following steps: The image preprocessing unit receives raw image data transmitted by the high-definition industrial camera and environmental parameter data transmitted by the environmental sensor. Based on the environmental parameter data, it determines the current ambient lighting conditions, generates a supplementary lighting control signal based on the determination result, and transmits it to the LED ring supplementary lighting source to adjust the supplementary lighting intensity. At the same time, it uses a two-dimensional gamma function to perform adaptive lighting correction on the raw image data to obtain corrected image data. The image preprocessing unit performs multi-scale decomposition on the corrected image data to obtain a multi-scale image sequence, crops the high-resolution image in the multi-scale image sequence into several sub-image blocks, and transmits the sub-image blocks and the low-resolution image in the multi-scale image sequence to the crack feature extraction unit. The crack feature extraction unit performs coarse-grained crack region localization on the low-resolution image based on a lightweight deep convolutional neural network, generates a crack region of interest mask, and transmits the crack region of interest mask to the edge detection subunit of the crack feature extraction unit. The edge detection subunit determines the target processing region in the sub-image block according to the crack region of interest mask, extracts crack edge pixels in the target processing region using an adaptive threshold edge detection algorithm, obtains initial edge image data, and transmits the initial edge image data to the crack identification unit. The crack identification unit performs connected component labeling on the initial edge image data, calculates the aspect ratio and area parameters of each connected component, and filters out unstructured surface noise with an aspect ratio less than a preset threshold and an area less than a preset number of pixels to obtain candidate crack edge data. The fracture identification unit performs line segment fitting on the candidate fracture edge data based on Hough transform, extracts fracture line segment parameters, and locally connects fracture line segments that meet the connection conditions according to distance and angle criteria to generate complete fracture trace data. The crack identification unit fuses and verifies the complete crack trace data with the crack region of interest mask, outputs the final crack identification result, and transmits it to the storage module and the display module respectively.
[0007] Furthermore, the high-definition industrial camera in the image acquisition module adopts a fixed-focus lens. The focal length of the fixed-focus lens is determined according to the typical excavation span of the tunnel face. The high-definition industrial camera is fixed to the tunnel excavation equipment or temporary support structure by a rigid bracket. The rigid bracket is equipped with a shock-absorbing and buffering device, which is used to absorb the mechanical vibration generated during tunnel construction. The color temperature of the LED ring light source is selected based on the typical reflection characteristics of the rock surface. The illumination angle of the LED ring light source is arranged at a fixed angle with the optical axis of the high-definition industrial camera. The fixed angle is used to reduce the specular reflection light from the rock surface entering the lens. The environmental sensor includes a dust concentration detection unit and an air humidity detection unit. The dust concentration detection unit and the air humidity detection unit are electrically connected to the main control module. The main control module determines whether the current environment meets the image acquisition conditions based on the real-time data fed back by the dust concentration detection unit and the air humidity detection unit. When the environmental conditions do not meet the preset threshold range, the main control module generates a waiting instruction or a warning message.
[0008] Furthermore, the illumination adaptive correction process performed by the image preprocessing unit includes: First, the original image data is converted from the RGB color space to the HSV color space, and the luminance component is extracted. Then, the local mean and local variance of the luminance component are calculated. Based on the local mean and local variance, a correction coefficient matrix of the two-dimensional gamma function is constructed. Each element in the correction coefficient matrix is related to the luminance value of the corresponding pixel position. The luminance component is nonlinearly transformed using the correction coefficient matrix. The transformed luminance component is then recombined with the original hue and saturation components and converted back to the RGB color space to obtain the corrected image data. The parameters of the two-dimensional gamma function are pre-calibrated based on the typical color characteristics of the tunnel face rock mass. The pre-calibration process is completed using a standard color chart under a standard light source environment.
[0009] Furthermore, the multi-scale decomposition process performed by the image preprocessing unit adopts the Gaussian pyramid decomposition algorithm. The Gaussian pyramid decomposition algorithm generates a multi-scale image sequence from the bottom layer to the top layer by performing multiple Gaussian blurring and downsampling operations on the corrected image data. The bottom layer image retains the original resolution detail information, and the top layer image retains the global structure information. The image preprocessing unit transmits the top layer image as a low-resolution image to the crack feature extraction unit. The image preprocessing unit performs non-overlapping uniform cropping of the underlying image according to a preset size to generate sub-image blocks. The size of the sub-image blocks is determined according to the dimension requirements of the input layer of the lightweight deep convolutional neural network. The image preprocessing unit assigns a spatial location index identifier to each sub-image block. The spatial location index identifier is used for position correspondence in the subsequent fusion verification process.
[0010] Furthermore, the crack feature extraction unit employs a lightweight deep convolutional neural network based on the MobileNetV2 architecture. The input layer of MobileNetV2 receives a low-resolution image. The inverse residual structure of MobileNetV2 introduces a linear bottleneck layer and a depthwise separable convolutional layer. The linear bottleneck layer is used to reduce the feature dimension, and the depthwise separable convolutional layer is used to reduce the number of parameters and computational cost. The output of MobileNetV2 is connected to a shallow semantic segmentation head. The shallow semantic segmentation head outputs a binary mask with the same size as the low-resolution image. In the binary mask, regions with pixel values of foreground represent crack regions of interest, and regions with pixel values of background represent non-crack regions. The crack feature extraction unit performs a morphological dilation operation on the binary mask to expand the coverage of the crack region of interest and generate a crack region of interest mask. The size of the structuring element of the morphological dilation operation is determined according to the typical width range of cracks on the tunnel face.
[0011] Furthermore, the adaptive threshold edge detection algorithm executed by the edge detection subunit includes: First, the effective processing area in the sub-image block is determined based on the crack region of interest mask. Pixels outside the crack region of interest mask are masked. Within the effective processing area, the gradient magnitude and gradient direction are calculated. The gradient magnitude is calculated using the Sobel operator, and the gradient direction is obtained by calculating the arctangent value based on the horizontal and vertical gradient components. Then, high and low thresholds are determined based on the statistical distribution of the gradient magnitude. The high threshold is the gray value where the cumulative probability in the gradient magnitude histogram reaches a preset proportion, and the low threshold is the product of the high threshold and a preset coefficient. Edge points are filtered based on the high and low thresholds. Pixels with gradient magnitudes greater than the high threshold are retained as confirmed edge points, and pixels with gradient magnitudes between the low and high thresholds and connected to the confirmed edge points are retained as candidate edge points. Finally, the initial edge image data composed of confirmed edge points and candidate edge points is output.
[0012] Furthermore, the connected component labeling process executed by the gap identification unit adopts the eight-connected neighborhood determination criterion. The eight-connected neighborhood determination criterion groups pixels with adjacent pixel positions and the same gray value into the same connected component. The gap identification unit traverses all foreground pixels in the initial edge image data, assigns a unique identifier to each connected component, calculates the minimum bounding rectangle of each connected component, uses the long side of the minimum bounding rectangle as the principal axis length, uses the short side of the minimum bounding rectangle as the principal axis width, calculates the ratio of the principal axis length to the principal axis width as the aspect ratio, and simultaneously counts the total number of pixels contained in each connected component as the area parameter. The preset threshold is set based on the statistical results of the typical aspect ratio of the crack at the working face. The preset number of pixels is calculated based on the minimum identifiable width of the crack and the image resolution. The crack identification unit retains the connected components that simultaneously satisfy the aspect ratio being greater than the preset threshold and the area being greater than the preset number of pixels, and filters out the connected components that do not meet the conditions to obtain candidate crack edge data.
[0013] Furthermore, the Hough transform process performed by the crack identification unit maps the edge pixels in the Cartesian coordinate system to the polar coordinate parameter space. The horizontal axis in the polar coordinate parameter space represents the angle between the line normal and the horizontal axis, and the vertical axis in the polar coordinate parameter space represents the distance from the line normal to the origin. The crack identification unit accumulates and counts the frequency of the quantized parameter combinations in the polar coordinate parameter space, extracts the local peak points as the parameters of the effective line segments, and calculates the endpoint coordinates in the Cartesian coordinate system based on the parameters of the effective line segments to obtain the crack line segment parameters. The distance criterion is set as follows: if the minimum distance between the endpoints of two fracture segments is less than a preset distance threshold, the distance criterion is satisfied. The angle criterion is set as follows: if the absolute value of the difference between the slopes of two fracture segments is less than the tangent value corresponding to the preset angle threshold, the angle criterion is satisfied. The preset distance threshold is determined based on the typical discontinuity length of the fracture trace line on the tunnel face, and the preset angle threshold is determined based on the statistical results of the consistency of the attitude of the same group of structural surfaces.
[0014] Furthermore, the fusion verification process performed by the crack identification unit includes: The complete crack trace data is projected onto the coordinate space corresponding to the low-resolution image and spatially overlaid with the crack region of interest mask. The length ratio of the complete crack trace data falling within the crack region of interest mask is calculated. If the length ratio is greater than the preset verification threshold, the complete crack trace data is deemed valid, retained, and output as the final crack identification result. If the length ratio is less than or equal to the preset verification threshold, the complete crack trace data is deemed to have false detection or missed detection, triggering a feedback correction mechanism. The feedback correction mechanism includes: adjusting the high and low threshold parameters of the adaptive threshold edge detection algorithm, or expanding the expansion range of the crack region of interest mask, and re-executing the edge detection and crack identification process until the verification conditions are met or the preset iteration limit is reached.
[0015] Furthermore, the storage module includes a temporary cache area and a permanent storage area. The temporary cache area is used to store multi-scale image sequences, sub-image blocks, crack region of interest masks, and initial edge image data. The permanent storage area is used to store the final crack identification results and the corresponding tunnel face station number information and acquisition time information. The display module includes a field display terminal and a remote monitoring terminal. The field display terminal is set in the tunnel excavation equipment control room and displays the current tunnel face crack distribution overlay image and crack statistical parameters in real time. The remote monitoring terminal is connected to the main control module via a wired or wireless communication link. The remote monitoring terminal receives and displays the crack identification results of multiple tunnel faces. The main control module is equipped with a data output interface, which is used to export the final crack identification results in a standard format to tunnel geological logging software or surrounding rock classification and evaluation system.
[0016] The beneficial effects of this invention are as follows: The tunnel face fracture surrounding rock identification and judgment system of this invention, through the collaborative work of the image acquisition module, main control module, storage module and display module, adopts a multi-scale decomposition and hierarchical processing strategy. First, a lightweight deep convolutional neural network is used to perform coarse-grained fracture region localization on low-resolution images, generating a fracture region of interest mask to limit the subsequent processing range. Then, adaptive threshold edge detection, connected component morphological filtering, Hough transform linear fitting and local edge connection based on distance and angle criteria are performed on the target region of the high-resolution sub-image block. Finally, the complete fracture trace data and the fracture region of interest mask are fused and verified to output the result. Thus, while ensuring the accuracy of fracture identification, the computational complexity is significantly reduced. It achieves the unity of high-precision identification and real-time judgment of tunnel face fractures in complex construction environments. It effectively overcomes the technical defects of existing technologies, such as the susceptibility of traditional image processing methods to environmental interference and poor fitting effect on non-linear fractures, and the large computational load of deep learning methods and difficulty in meeting real-time requirements. It improves the practicality and reliability of intelligent geological logging technology in tunnel engineering. Attached Figure Description
[0017] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.
[0018] Figure 1 This is a schematic diagram of the tunnel face fracture surrounding rock identification system of the present invention. Detailed Implementation
[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. The technical solutions provided by various embodiments of this invention will be described in detail below with reference to the accompanying drawings.
[0020] Please see Figure 1 This invention provides a system for identifying and judging fractured surrounding rock at the working face. The system includes an image acquisition module, a main control module, a storage module, and a display module. The modules are electrically connected to achieve data transmission and control signal interaction.
[0021] The image acquisition module is used to acquire image data of the rock surface in the tunnel face construction environment. The image acquisition module includes a high-definition industrial camera, an LED ring light source, and an environmental sensor. The LED ring light source is set around the lens of the high-definition industrial camera, and the environmental sensor is integrated with the high-definition industrial camera. The high-definition industrial camera, the LED ring light source, and the environmental sensor are all electrically connected to the main control module.
[0022] In practical engineering applications, high-definition industrial cameras use fixed-focus lenses. The focal length of the fixed-focus lens is determined according to the typical excavation span of the tunnel face to ensure that the overall and local crack features of the tunnel face can be clearly captured at a fixed shooting distance. The high-definition industrial camera is fixed to the tunnel excavation equipment or temporary support structure by a rigid bracket. The rigid bracket is equipped with a shock-absorbing buffer device to absorb the mechanical vibration generated during tunnel construction and avoid image blurring caused by vibration. The color temperature of the LED ring supplementary light source is selected according to the typical reflection characteristics of the rock surface. The illumination angle of the LED ring supplementary light source is arranged at a fixed angle with the optical axis of the high-definition industrial camera. The fixed angle is used to reduce the specular reflection light from the rock surface entering the lens, thereby reducing specular interference.
[0023] The environmental sensor includes a dust concentration detection unit and an air humidity detection unit, which are electrically connected to the main control module. Based on the real-time data fed back by the dust concentration detection unit and the air humidity detection unit, the main control module determines whether the current environment meets the image acquisition conditions. When the environmental conditions do not meet the preset threshold range, the main control module generates a waiting instruction or an early warning message to prompt the construction personnel to improve the acquisition environment or wait for the environmental conditions to recover.
[0024] The above design addresses the special scenario of large fluctuations in dust concentration and severe water mist interference at tunnel construction sites. Through the linkage control of real-time environmental monitoring and supplementary lighting adjustment, it ensures that the acquired raw image data has a sufficient quality foundation, providing reliable input for subsequent crack identification.
[0025] The main control module receives image data and environmental parameters transmitted from the image acquisition module, and performs preprocessing and crack identification on the image data. The main control module includes an image preprocessing unit, a crack feature extraction unit, and a crack identification unit, which are electrically connected sequentially. The main control module is also electrically connected to the storage module and the display module. The image preprocessing unit receives raw image data from a high-definition industrial camera and environmental parameter data from an environmental sensor. Based on the environmental parameter data, it determines the current ambient lighting conditions, generates a supplementary lighting control signal based on the determination result, and transmits it to an LED ring supplementary lighting source to adjust the supplementary lighting intensity. Simultaneously, a two-dimensional gamma function is used to perform adaptive lighting correction on the raw image data to obtain the corrected image data.
[0026] The illumination adaptive correction process performed by the image preprocessing unit includes: first, converting the original image data from RGB color space to HSV color space, extracting the luminance component, then calculating the local mean and local variance of the luminance component, constructing a correction coefficient matrix of a two-dimensional gamma function based on the local mean and local variance, where each element in the correction coefficient matrix is related to the luminance value of the corresponding pixel position, performing a nonlinear transformation on the luminance component using the correction coefficient matrix, and then recombining the transformed luminance component with the original hue and saturation components, converting it back to the RGB color space to obtain the corrected image data. The parameters of the two-dimensional gamma function are pre-calibrated based on the typical color characteristics of the tunnel face rock mass, and the pre-calibration process is completed using a standard color chart under a standard light source environment.
[0027] The above correction method addresses the uneven lighting and severe local shadows at the tunnel face. Through local adaptive adjustment, it effectively compresses overly bright areas and enhances overly dark areas, ensuring that the crack edges maintain consistent contrast throughout the image and avoiding the loss of local details caused by traditional global correction methods.
[0028] The image preprocessing unit performs multi-scale decomposition on the corrected image data to obtain a multi-scale image sequence. The high-resolution image in the multi-scale image sequence is cropped into several sub-image blocks, and the sub-image blocks and the low-resolution image in the multi-scale image sequence are transmitted to the crack feature extraction unit. The multi-scale decomposition process performed by the image preprocessing unit uses the Gaussian pyramid decomposition algorithm. The Gaussian pyramid decomposition algorithm generates a multi-scale image sequence from the bottom layer to multiple layers by performing multiple Gaussian blurring and downsampling operations on the corrected image data. The bottom layer image retains the original resolution detail information, while the top layer image retains global structural information. The image preprocessing unit transmits the top layer image as a low-resolution image to the crack feature extraction unit. The image preprocessing unit then performs non-overlapping uniform cropping of the bottom layer image according to a preset size to generate sub-image blocks. The size of the sub-image blocks is determined according to the dimension requirements of the input layer of the lightweight deep convolutional neural network. The image preprocessing unit assigns a spatial location index identifier to each sub-image block, which is used for positional correspondence in the subsequent fusion verification process.
[0029] The aforementioned multi-scale decomposition strategy addresses the conflict between the resolution of the face image and computational resources. Through layered processing, it preserves the crack details of the high-resolution image while reducing the computational burden on the deep learning model, enabling the system to achieve real-time processing on an embedded hardware platform.
[0030] The crack feature extraction unit uses a lightweight deep convolutional neural network to perform coarse-grained crack region localization on low-resolution images, generating a crack region of interest (ROI) mask. This ROI mask is then transmitted to the edge detection subunit of the crack feature extraction unit. The lightweight deep convolutional neural network used in the crack feature extraction unit is based on the MobileNetV2 architecture. The input layer of MobileNetV2 receives the low-resolution image. The inverse residual structure of MobileNetV2 incorporates a linear bottleneck layer and a depthwise separable convolutional layer. The linear bottleneck layer reduces feature dimensionality, while the depthwise separable convolutional layer reduces the number of parameters and computational cost. The output of MobileNetV2 is connected to a shallow semantic segmentation head, which outputs a binary mask of the same size as the low-resolution image. In the binary mask, regions with pixel values representing the foreground indicate crack ROIs, and regions with pixel values representing the background indicate non-crack regions. The crack feature extraction unit performs a morphological dilation operation on the binary mask to expand the coverage of the crack ROI, generating the crack ROI mask. The size of the structuring element in the morphological dilation operation is determined based on the typical width range of cracks on the tunnel face.
[0031] The above design addresses the problems of complex crack distribution patterns and high computational cost of traditional global edge detection. It utilizes a lightweight network to quickly locate the approximate crack area, limiting subsequent fine processing to a local range. This avoids high-density computation across the entire image. At the same time, the morphological dilation operation ensures the integrity of crack edges and prevents crack breakage caused by inaccurate network segmentation boundaries.
[0032] The edge detection subunit determines the target processing region in the sub-image block based on the crack region of interest mask, uses an adaptive threshold edge detection algorithm to extract crack edge pixels in the target processing region, obtains initial edge image data, and transmits the initial edge image data to the crack identification unit.
[0033] The adaptive threshold edge detection algorithm executed by the edge detection subunit includes: first, determining the effective processing area in the sub-image block based on the crack region of interest mask, and shielding pixels outside the crack region of interest mask range; calculating the gradient magnitude and gradient direction within the effective processing area; the gradient magnitude is calculated using the Sobel operator, and the gradient direction is obtained by calculating the arctangent value based on the horizontal and vertical gradient components; then, determining high and low thresholds based on the statistical distribution of the gradient magnitude; the high threshold is the gray value where the cumulative probability in the gradient magnitude histogram reaches a preset proportion, and the low threshold is the product of the high threshold and a preset coefficient; edge point filtering is performed based on the high and low thresholds, retaining pixels with gradient magnitudes greater than the high threshold as determined edge points, and retaining pixels with gradient magnitudes between the low and high thresholds that are connected to the determined edge points as candidate edge points; finally, outputting the initial edge image data composed of determined edge points and candidate edge points.
[0034] The aforementioned adaptive threshold strategy addresses the complex texture of the rock mass surface at the tunnel face and the large variation in contrast between cracks and the background. It avoids edge omissions or noise introduction caused by fixed thresholds, and significantly reduces invalid computation by limiting the processing area through masking.
[0035] The crack identification unit marks the connected components of the initial edge image data, calculates the aspect ratio and area parameters of each connected component, and filters out unstructured surface noise with aspect ratios less than a preset threshold and area less than a preset number of pixels to obtain candidate crack edge data. The connected component labeling process executed by the crack identification unit adopts the eight-connected neighborhood criterion. The eight-connected neighborhood criterion groups pixels with adjacent pixel positions and the same grayscale value into the same connected component. The crack identification unit traverses all foreground pixels in the initial edge image data, assigns a unique identifier to each connected component, calculates the minimum bounding rectangle of each connected component, uses the long side of the minimum bounding rectangle as the principal axis length, and the short side of the minimum bounding rectangle as the principal axis width, calculates the ratio of the principal axis length to the principal axis width as the aspect ratio, and simultaneously counts the total number of pixels contained in each connected component as the area parameter. The preset threshold is set based on the statistical results of typical aspect ratios of cracks at the tunnel face, and the preset number of pixels is calculated based on the minimum identifiable width of the crack and the image resolution. The crack identification unit retains connected components that simultaneously satisfy the aspect ratio greater than the preset threshold and the area greater than the preset number of pixels, and filters out connected components that do not meet the conditions to obtain candidate crack edge data.
[0036] The aforementioned morphological filtering method addresses noise interference such as spots and textures in the face image by distinguishing between the linear geometric features of the crack and the blocky morphology of the noise, effectively preserving the true crack edges and filtering out a large amount of unstructured surface noise.
[0037] The crack identification unit fits straight line segments to candidate crack edge data based on the Hough transform, extracts crack segment parameters, and locally connects crack segments that meet the connection conditions according to distance and angle criteria to generate complete crack trace data. The Hough transform process performed by the crack identification unit maps edge pixels in the Cartesian coordinate system to polar coordinate parameter space. The horizontal axis in the polar coordinate parameter space represents the angle between the line normal and the horizontal axis, and the vertical axis represents the distance from the line normal to the origin. The crack identification unit accumulates and counts the frequency of quantized parameter combinations in the polar coordinate parameter space, extracts local peak points as parameters of effective straight line segments, and calculates the endpoint coordinates in the Cartesian coordinate system based on the parameters of the effective straight line segments to obtain the crack segment parameters.
[0038] The distance criterion is set as follows: if the minimum distance between the endpoints of two crack segments is less than a preset distance threshold, then the distance criterion is satisfied.
[0039] The angle criterion is set as follows: if the absolute value of the difference between the slopes of two fracture line segments is less than the tangent value corresponding to the preset angle threshold, then the angle criterion is satisfied. The preset distance threshold is determined based on the typical discontinuity length of the fracture trace on the tunnel face, and the preset angle threshold is determined based on the statistical results of the consistency of the attitude of the structural surfaces in the same group.
[0040] The aforementioned local connection strategy addresses the problem of discontinuous traces caused by uneven rock surfaces and local obstruction of fractures at the tunnel face. By using geometric continuity constraints, segmented fractures are connected into complete traces, restoring the true extension morphology of the fractures and providing complete structural surface information for subsequent geological analysis.
[0041] The crack identification unit fuses and verifies the complete crack trace data with the crack region of interest (ROI) mask, outputting the final crack identification result and transmitting it to the storage module and display module respectively. The fusion verification process performed by the crack identification unit includes: projecting the complete crack trace data onto the coordinate space corresponding to the low-resolution image; performing spatial overlay analysis with the crack ROI mask; calculating the proportion of the complete crack trace data falling within the crack ROI mask's range; if the proportion is greater than a preset verification threshold, the complete crack trace data is deemed valid, retained, and output as the final crack identification result; if the proportion is less than or equal to the preset verification threshold, the complete crack trace data is determined to have false detections or missed detections, triggering a feedback correction mechanism.
[0042] The feedback correction mechanism includes: adjusting the high and low threshold parameters of the adaptive threshold edge detection algorithm, or expanding the expansion range of the crack region of interest mask, and re-executing the edge detection and crack identification process until the verification conditions are met or the preset iteration limit is reached.
[0043] The above verification mechanism addresses the potential inconsistency between deep learning segmentation results and traditional edge detection results. It ensures the reliability of the final result through spatial consistency verification, while the feedback correction mechanism enables the system to have self-optimization capabilities and adapt to the crack identification needs under different geological conditions.
[0044] The storage module is used to store intermediate and final data during the crack identification process. It is electrically connected to the main control module. The storage module includes a temporary buffer and a permanent storage area. The temporary buffer stores multi-scale image sequences, sub-image blocks, crack region-of-interest masks, and initial edge image data. The permanent storage area stores the final crack identification results, corresponding face station information, and acquisition time information.
[0045] The above-mentioned partitioned storage design addresses the large amount of intermediate data generated during the identification of fractures at the working face. It saves storage space through a cyclic overwrite mechanism of the temporary cache area, while retaining key results and related information to facilitate subsequent traceability and geological logging.
[0046] The display module is used to display the fracture identification results in real time and is electrically connected to the main control module. The display module includes a field display terminal and a remote monitoring terminal. The field display terminal is located in the tunnel boring machine's control room and displays a superimposed image of the current face fracture distribution and fracture statistical parameters in real time. The remote monitoring terminal is connected to the main control module via a wired or wireless communication link and receives and displays the fracture identification results from multiple faces. The main control module has a data output interface used to export the final fracture identification results in a standard format to tunnel geological logging software or a surrounding rock grading and evaluation system.
[0047] The aforementioned multi-terminal display design meets the information needs of different roles at the tunnel construction site. On-site operators can monitor the geological conditions of the tunnel face in real time, while remote management personnel can monitor geological changes at multiple working faces. The standard format export function enables seamless integration with existing engineering software, improving the efficiency of geological data utilization.
[0048] As can be seen from the above embodiments, this invention constructs a multi-scale hierarchical processing system architecture to address the special construction environment and crack identification technology requirements of tunnel faces. It organically combines the coarse-grained localization capability of lightweight deep learning models with the fine edge extraction capability of traditional image processing algorithms. By using crack region of interest masks, it achieves the complementary advantages of the two methods and a reasonable allocation of computational load. This overcomes the shortcomings of pure deep learning methods, such as large computational load and poor real-time performance, as well as the shortcomings of pure traditional methods, such as weak anti-interference capability and poor adaptability. At the same time, it designs special mechanisms for tunnel face scenarios, such as environment adaptive acquisition control, illumination adaptive correction, morphological noise filtering, and spatial consistency verification, so that the system can maintain stable and reliable crack identification performance in complex and ever-changing construction environments. This provides a practical and feasible technical solution for intelligent geological logging of tunnels.
[0049] The embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention.
Claims
1. A system for identifying and judging fractured surrounding rock at a tunnel face, characterized in that, include: An image acquisition module is used to acquire image data of the rock surface in the tunnel face construction environment. The image acquisition module includes a high-definition industrial camera, an LED ring light source, and an environmental sensor. The LED ring light source is set around the lens of the high-definition industrial camera. The environmental sensor is integrated with the high-definition industrial camera. The high-definition industrial camera, the LED ring light source, and the environmental sensor are all electrically connected to the main control module. The main control module is used to receive image data and environmental parameters transmitted by the image acquisition module, and to preprocess the image data and identify and judge the cracks. The main control module includes an image preprocessing unit, a crack feature extraction unit and a crack identification unit. The image preprocessing unit, the crack feature extraction unit and the crack identification unit are electrically connected in sequence. The main control module is electrically connected to the storage module and the display module respectively. A storage module is used to store intermediate data and final result data during the crack identification and judgment process. The storage module is electrically connected to the main control module. The display module is used to display the crack identification results in real time, and the display module is electrically connected to the main control module.
2. The face fracture surrounding rock identification system according to claim 1, characterized in that, The main control module is configured to perform the following steps: The image preprocessing unit receives raw image data transmitted by the high-definition industrial camera and environmental parameter data transmitted by the environmental sensor. Based on the environmental parameter data, it determines the current ambient lighting conditions, generates a supplementary lighting control signal based on the determination result, and transmits it to the LED ring supplementary lighting source to adjust the supplementary lighting intensity. At the same time, it uses a two-dimensional gamma function to perform adaptive lighting correction on the raw image data to obtain corrected image data. The image preprocessing unit performs multi-scale decomposition on the corrected image data to obtain a multi-scale image sequence, crops the high-resolution image in the multi-scale image sequence into several sub-image blocks, and transmits the sub-image blocks and the low-resolution image in the multi-scale image sequence to the crack feature extraction unit. The crack feature extraction unit performs coarse-grained crack region localization on the low-resolution image based on a lightweight deep convolutional neural network, generates a crack region of interest mask, and transmits the crack region of interest mask to the edge detection subunit of the crack feature extraction unit. The edge detection subunit determines the target processing region in the sub-image block according to the crack region of interest mask, extracts crack edge pixels in the target processing region using an adaptive threshold edge detection algorithm, obtains initial edge image data, and transmits the initial edge image data to the crack identification unit. The crack identification unit performs connected component labeling on the initial edge image data, calculates the aspect ratio and area parameters of each connected component, and filters out unstructured surface noise with an aspect ratio less than a preset threshold and an area less than a preset number of pixels to obtain candidate crack edge data. The fracture identification unit performs line segment fitting on the candidate fracture edge data based on Hough transform, extracts fracture line segment parameters, and locally connects fracture line segments that meet the connection conditions according to distance and angle criteria to generate complete fracture trace data. The crack identification unit fuses and verifies the complete crack trace data with the crack region of interest mask, outputs the final crack identification result, and transmits it to the storage module and the display module respectively. The high-definition industrial camera in the image acquisition module uses a fixed-focus lens. The focal length of the fixed-focus lens is determined according to the typical excavation span of the tunnel face. The high-definition industrial camera is fixed to the tunnel excavation equipment or temporary support structure by a rigid bracket. The rigid bracket is equipped with a shock-absorbing and buffering device, which is used to absorb the mechanical vibration generated during tunnel construction. The color temperature of the LED ring light source is selected based on the typical reflection characteristics of the rock surface. The illumination angle of the LED ring light source is arranged at a fixed angle with the optical axis of the high-definition industrial camera. The fixed angle is used to reduce the specular reflection light from the rock surface entering the lens. The environmental sensor includes a dust concentration detection unit and an air humidity detection unit. The dust concentration detection unit and the air humidity detection unit are electrically connected to the main control module. The main control module determines whether the current environment meets the image acquisition conditions based on the real-time data fed back by the dust concentration detection unit and the air humidity detection unit. When the environmental conditions do not meet the preset threshold range, the main control module generates a waiting instruction or a warning message.
3. The face fracture surrounding rock identification system according to claim 2, characterized in that, The illumination adaptive correction process performed by the image preprocessing unit includes: First, the original image data is converted from the RGB color space to the HSV color space, and the luminance component is extracted. Then, the local mean and local variance of the luminance component are calculated. Based on the local mean and local variance, a correction coefficient matrix of the two-dimensional gamma function is constructed. Each element in the correction coefficient matrix is related to the luminance value of the corresponding pixel position. The luminance component is nonlinearly transformed using the correction coefficient matrix. The transformed luminance component is then recombined with the original hue and saturation components and converted back to the RGB color space to obtain the corrected image data. The parameters of the two-dimensional gamma function are pre-calibrated based on the typical color characteristics of the tunnel face rock mass. The pre-calibration process is completed using a standard color chart under a standard light source environment.
4. The face fracture surrounding rock identification system according to claim 3, characterized in that, The multi-scale decomposition process performed by the image preprocessing unit adopts the Gaussian pyramid decomposition algorithm. The Gaussian pyramid decomposition algorithm generates a multi-scale image sequence from the bottom to the top by performing multiple Gaussian blurring and downsampling operations on the corrected image data. The bottom image retains the original resolution detail information, and the top image retains the global structural information. The image preprocessing unit transmits the top image as a low-resolution image to the crack feature extraction unit. The image preprocessing unit performs non-overlapping uniform cropping of the underlying image according to a preset size to generate sub-image blocks. The size of the sub-image blocks is determined according to the dimension requirements of the input layer of the lightweight deep convolutional neural network. The image preprocessing unit assigns a spatial location index identifier to each sub-image block. The spatial location index identifier is used for position correspondence in the subsequent fusion verification process.
5. The face fracture surrounding rock identification system according to claim 4, characterized in that, The crack feature extraction unit employs a lightweight deep convolutional neural network based on the MobileNetV2 architecture. The input layer of MobileNetV2 receives a low-resolution image. The inverse residual structure of MobileNetV2 introduces a linear bottleneck layer and a depthwise separable convolutional layer. The linear bottleneck layer is used to reduce the feature dimension, and the depthwise separable convolutional layer is used to reduce the number of parameters and computational cost. The output of MobileNetV2 is connected to a shallow semantic segmentation head. The shallow semantic segmentation head outputs a binary mask with the same size as the low-resolution image. In the binary mask, regions with pixel values of foreground represent crack regions of interest, and regions with pixel values of background represent non-crack regions. The crack feature extraction unit performs a morphological dilation operation on the binary mask to expand the coverage of the crack region of interest and generate a crack region of interest mask. The size of the structuring element of the morphological dilation operation is determined according to the typical width range of cracks on the tunnel face.
6. The face fracture surrounding rock identification system according to claim 5, characterized in that, The adaptive threshold edge detection algorithm executed by the edge detection subunit includes: First, the effective processing area in the sub-image block is determined based on the crack region of interest mask. Pixels outside the crack region of interest mask are masked. Within the effective processing area, the gradient magnitude and gradient direction are calculated. The gradient magnitude is calculated using the Sobel operator, and the gradient direction is obtained by calculating the arctangent value based on the horizontal and vertical gradient components. Then, high and low thresholds are determined based on the statistical distribution of the gradient magnitude. The high threshold is the gray value where the cumulative probability in the gradient magnitude histogram reaches a preset proportion, and the low threshold is the product of the high threshold and a preset coefficient. Edge points are filtered based on the high and low thresholds. Pixels with gradient magnitudes greater than the high threshold are retained as confirmed edge points, and pixels with gradient magnitudes between the low and high thresholds and connected to the confirmed edge points are retained as candidate edge points. Finally, the initial edge image data composed of confirmed edge points and candidate edge points is output.
7. The face fracture surrounding rock identification system according to claim 6, characterized in that, The connected component labeling process executed by the crack identification unit adopts the eight-connected neighborhood determination criterion. The eight-connected neighborhood determination criterion groups pixels with adjacent pixel positions and the same gray value into the same connected component. The crack identification unit traverses all foreground pixels in the initial edge image data, assigns a unique identifier to each connected component, calculates the minimum bounding rectangle of each connected component, uses the long side of the minimum bounding rectangle as the principal axis length, uses the short side of the minimum bounding rectangle as the principal axis width, calculates the ratio of the principal axis length to the principal axis width as the aspect ratio, and counts the total number of pixels contained in each connected component as the area parameter. The preset threshold is set based on the statistical results of the typical aspect ratio of the crack at the working face. The preset number of pixels is calculated based on the minimum identifiable width of the crack and the image resolution. The crack identification unit retains the connected components that simultaneously satisfy the aspect ratio being greater than the preset threshold and the area being greater than the preset number of pixels, and filters out the connected components that do not meet the conditions to obtain candidate crack edge data.
8. The face fracture surrounding rock identification system according to claim 7, characterized in that, The Hough transform process performed by the crack identification unit maps the edge pixels in the Cartesian coordinate system to the polar coordinate parameter space. The horizontal axis in the polar coordinate parameter space represents the angle between the line normal and the horizontal axis, and the vertical axis in the polar coordinate parameter space represents the distance from the line normal to the origin. The crack identification unit accumulates and counts the frequency of the quantized parameter combinations in the polar coordinate parameter space, extracts the local peak points as the parameters of the effective line segments, and calculates the endpoint coordinates in the Cartesian coordinate system based on the parameters of the effective line segments to obtain the crack line segment parameters. The distance criterion is set as follows: if the minimum distance between the endpoints of two fracture segments is less than a preset distance threshold, the distance criterion is satisfied. The angle criterion is set as follows: if the absolute value of the difference between the slopes of two fracture segments is less than the tangent value corresponding to the preset angle threshold, the angle criterion is satisfied. The preset distance threshold is determined based on the typical discontinuity length of the fracture trace line on the tunnel face, and the preset angle threshold is determined based on the statistical results of the consistency of the attitude of the same group of structural surfaces.
9. The face fracture surrounding rock identification system according to claim 8, characterized in that, The fusion verification process performed by the crack identification unit includes: The complete crack trace data is projected onto the coordinate space corresponding to the low-resolution image and spatially overlaid with the crack region of interest mask. The length ratio of the complete crack trace data falling within the crack region of interest mask is calculated. If the length ratio is greater than the preset verification threshold, the complete crack trace data is deemed valid, retained, and output as the final crack identification result. If the length ratio is less than or equal to the preset verification threshold, the complete crack trace data is deemed to have false detection or missed detection, triggering a feedback correction mechanism. The feedback correction mechanism includes: adjusting the high and low threshold parameters of the adaptive threshold edge detection algorithm, or expanding the expansion range of the crack region of interest mask, and re-executing the edge detection and crack identification process until the verification conditions are met or the preset iteration limit is reached.
10. The face fracture surrounding rock identification system according to claim 9, characterized in that, The storage module includes a temporary cache area and a permanent storage area. The temporary cache area is used to store multi-scale image sequences, sub-image blocks, crack region of interest masks, and initial edge image data. The permanent storage area is used to store the final crack identification results and the corresponding tunnel face station number information and acquisition time information. The display module includes a field display terminal and a remote monitoring terminal. The field display terminal is set in the tunnel excavation equipment control room and displays the current tunnel face crack distribution overlay image and crack statistical parameters in real time. The remote monitoring terminal is connected to the main control module via a wired or wireless communication link. The remote monitoring terminal receives and displays the crack identification results of multiple tunnel faces. The main control module is equipped with a data output interface, which is used to export the final crack identification results in a standard format to tunnel geological logging software or surrounding rock classification and evaluation system.