A method for intelligent identification and extraction of surface cracks in soft rock roadway based on machine vision

Abstract

The present application relates to the technical field of mine soft rock roadway structure health monitoring and machine vision detection, and provides a kind of soft rock roadway surface crack intelligent identification and extraction method based on machine vision.The method uses industrial camera to continuously image collection on the surface of soft rock roadway roof and two sides, combines camera calibration and illumination adaptive preprocessing, realizes image distortion correction and noise suppression;The preprocessed image is input into the deep learning crack identification model to output the crack pixel-level segmentation result, and based on the scale calibration, the pixel features are converted into engineering scale parameters, and the geometric indexes such as crack length, width and strike direction are automatically extracted;Further, a multi-index risk assessment model is constructed, which comprehensively considers factors such as crack width grade, length expansion rate, direction and angle between soft rock roadway principal stress direction to calculate risk grade and output early warning information.The method realizes the automatic identification, parameterization extraction and risk classification of cracks, and improves the efficiency of soft rock roadway inspection and the objectivity of evaluation.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring and structural health monitoring in mine soft rock tunnel engineering, and in particular to an intelligent identification and extraction method that uses machine vision and deep learning to automatically identify surface cracks in soft rock tunnels, extract geometric parameters, and complete risk level assessment. Background Technology

[0002] As a crucial component of mine production, transportation, and ventilation systems, the stability of the surrounding rock in soft rock roadways directly impacts the safety of underground personnel and the continuity of production. Under the influence of factors such as mining disturbance, stress redistribution, groundwater action, and aging support structures, the roof and sidewalls of soft rock roadways are prone to various defects, including cracks, spalling, and exfoliation. The generation and propagation of cracks are often among the early signs of surrounding rock instability. Currently, crack inspection relies heavily on manual visual inspection and experience-based judgment, which suffers from low detection efficiency, high subjectivity, difficulty in quantifying and recording cracks, and inability to continuously track crack evolution. Furthermore, some soft rock roadways have environments with high dust levels, poor lighting, and limited space, posing safety risks to close-range manual observation.

[0003] Existing image-based crack detection technologies have been applied to surface structures, but they still face several technical bottlenecks in underground soft rock tunnels: First, the complex surface texture, uneven lighting, and presence of wet spots and coal dust in soft rock tunnels easily lead to false positives and false negatives; second, cracks are long and narrow with a wide scale range, making it difficult to reliably extract them using single thresholds or traditional edge detection algorithms; third, the identification results are mostly limited to a "present / absent" level, lacking automatic extraction of engineering quantitative indicators such as crack length, width, and direction; fourth, there is a lack of linkage with engineering risk assessment logic, making it impossible to generate operable hierarchical early warning outputs. Therefore, it is necessary to propose an intelligent crack identification, parameter extraction, and risk classification method suitable for underground soft rock tunnel environments to achieve objective quantification and continuous monitoring of cracks. Summary of the Invention

[0004] This invention aims to solve the problems existing in the inspection of cracks in soft rock roadways, such as heavy reliance on manual labor, low detection efficiency, difficulty in quantifying crack parameters, insufficient stability in identification under complex lighting and noise conditions, and lack of risk classification and early warning output.

[0005] To achieve the above objectives, this invention provides a machine vision-based intelligent identification and extraction method for surface cracks in soft rock tunnels, comprising at least the following technical points:

[0006] (1) An industrial camera is used to continuously acquire images of the roof and sidewall surfaces of the soft rock roadway. Combined with a supplementary lighting device and pose recording, a traceable "mileage / location-image" mapping is achieved.

[0007] (2) Perform camera calibration and distortion correction on the acquired images, and perform illumination adaptive enhancement, noise reduction and contrast enhancement to obtain standardized input suitable for deep learning inference;

[0008] (3) Construct and train a deep learning model for crack recognition, and output crack probability maps or crack masks in a pixel-level segmentation manner to improve the robustness of crack detection in complex backgrounds.

[0009] (4) Based on the crack mask, perform connected domain decomposition, skeletonization and boundary extraction to obtain the crack center line and boundary line, and form the crack geometric representation;

[0010] (5) Based on the calibration scale or external scale identifier, convert the crack pixel features into engineering scale, calculate the crack length, maximum / average width and direction, and form a crack parameter set;

[0011] (6) Establish a crack risk assessment model, which should at least comprehensively consider indicators such as crack width level, crack length, crack density, and the angle between crack direction and the principal stress direction of soft rock roadway, calculate risk score and output risk level, and link early warning strategy and data visualization display.

[0012] Furthermore, the present invention also provides a system implementation corresponding to the above method. The system includes an image acquisition module, a preprocessing module, a fracture identification module, a fracture extraction module, a parameter calculation module, a risk assessment and early warning module, and a data management and visualization module. The modules work together to form a closed loop for intelligent downhole fracture detection.

[0013] Compared with the prior art, the present invention has at least the following beneficial effects:

[0014] (1) Deep learning pixel-level segmentation improves crack recognition accuracy and noise resistance, adapting to uneven lighting and complex textures in underground environments;

[0015] (2) To achieve automated quantitative extraction of crack length, width and direction, forming engineering parameters that can be used for trend analysis;

[0016] (3) Introduce a multi-indicator risk assessment and graded early warning mechanism to transform the “identification results” into “decisive risk levels”;

[0017] (4) Supports location mapping and visualization, which facilitates rapid location, verification and long-term evolution tracking of cracks, significantly improving the efficiency and safety of soft rock roadway inspection. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the module structure of the intelligent identification and extraction system for surface cracks in soft rock tunnels based on machine vision, according to the present invention.

[0019] Figure 2 is a flowchart illustrating the intelligent identification and extraction method for surface cracks in soft rock tunnels based on machine vision according to the present invention. Detailed Implementation

[0020] This embodiment provides a method for intelligent crack identification and extraction suitable for fixed-point inspection in underground soft rock roadways. The system includes:

[0021] (1) Image acquisition module: industrial camera, lens, mounting bracket, supplementary lighting device, and trigger. The industrial camera is installed on the side wall support of the soft rock tunnel or at the inspection point of the soft rock tunnel. The focal length of the lens is selected according to the cross-sectional size of the soft rock tunnel so that the camera's field of view covers the target area of ​​the roof and both sides. The supplementary lighting device uses ring LED lights or strip array LED lights, installed around the lens or on both sides of the camera to reduce shadow interference. The trigger is used for timed shooting or triggered by the inspection personnel.

[0022] (2) Calibration and preprocessing module: including camera calibration unit and image preprocessing unit. After installation, the camera calibration unit uses the calibration board to complete the intrinsic parameter calibration and obtain parameters such as focal length, principal point, and distortion coefficient; the preprocessing unit performs distortion correction, noise reduction and illumination equalization on the acquired images.

[0023] (3) Crack recognition module: Deployed on a downhole edge computing terminal or portable industrial control computer, with a pre-trained deep learning crack recognition model built in, used to infer the input image and output a crack segmentation probability map.

[0024] (4) Crack extraction module: used to perform thresholding, connected component analysis, skeletonization and boundary extraction on the segmentation results, and output the crack center line and crack boundary line.

[0025] (5) Parameter calculation module: used to calculate crack length, width and direction with the support of scale conversion relationship, and output crack parameter set.

[0026] (6) Risk assessment and early warning module: used to calculate risk score and give risk level based on crack parameter set.

[0027] (7) Data management and visualization module: used to store the original crack image, segmentation mask, crack center line / boundary line, crack parameters and risk level, and generate reports.

[0028] (A) Image Acquisition and Location Information Recording: Cameras are deployed at mileage markers or fixed points in the soft rock tunnel, and the mileage, cross-sectional position (roof / left side / right side), and camera orientation at each point are recorded. Data acquisition is timed, with images acquired at intervals of, for example, 1 to 5 minutes. To reduce dust impact, a transparent protective cover and a simple cleaning mechanism can be installed in front of the lens.

[0029] (B) Calibration and Preprocessing: Camera calibration involves capturing multiple images using a standard calibration board and solving for the camera intrinsic parameter matrix and distortion coefficients. Preprocessing includes:

[0030] Distortion correction: Radial and tangential distortion correction is performed on the original image using calibration parameters;

[0031] Noise reduction: Bilateral filtering or nonlocal mean filtering is used to suppress coal dust particle noise;

[0032] Lighting equalization: Adaptive histogram equalization or Retinex enhancement is used to make details in shadow areas and overexposed areas clearer;

[0033] Standardization: The image is scaled to the model input size and the mean and variance are normalized.

[0034] (C) Crack Recognition Model Inference: The crack recognition module inputs a standardized image into the deep learning model, and the model outputs a crack probability map of the same size as the input. Thresholding is performed on the probability map (the threshold can be fixed or adaptive) to obtain a binary crack mask. To enhance the recognition of fine cracks, multi-scale inference can be introduced, cropping the image at different scales and fusing the output results.

[0035] (D) Crack geometry extraction: Connectivity analysis is performed on the crack mask to separate different crack regions; the area, perimeter, and aspect ratio of the bounding rectangle are calculated for each connected region, and small or approximately circular noise regions are removed. Skeletonization is performed on the remaining crack regions to obtain the centerline; the centerline is then connected by breakpoints and pruned to generate a continuous main centerline; at the same time, the boundary lines of the crack regions are extracted.

[0036] (E) Crack parameter calculation: Length: Calculate the arc length of the centerline according to the pixel sequence and convert it into the actual length; Width: Sample the boundary spacing in the normal direction of the centerline to obtain the width sequence, and output the maximum width and average width; Direction: Perform principal component analysis on the centerline point set to obtain the principal direction vector and convert it into direction angle; If the crack is a curve, the segmented direction can be calculated and the principal direction and direction dispersion can be output.

[0037] (F) Risk Assessment and Early Warning: Establish a risk scoring function that integrates multiple indicators, comprehensively considering crack width, length, density, location, and orientation. The specific calculation model is as follows: First, calculate the severity score of the foundation cracks. :

[0038] In the formula, The maximum width of the crack. The length of the crack. To monitor the crack density within the area; , , These are the engineering safety benchmark thresholds for the corresponding soft rock roadway levels; , , The weight coefficient of a single indicator, and satisfies + + = 1. Since crack width has the strongest indicative effect on surrounding rock instability, it is usually set... .

[0039] Secondly, a spatial location additional coefficient is introduced. Additional coefficient for directional risk Calculate the final comprehensive risk score. :

[0040] Among them, the positional additional coefficient Differentiated weights are set based on the location of the crack; for example, when the crack is located in a critical area of ​​the roof or near the anchor bolt holes, a different weight is applied. to When located in the regular area of ​​the two sides, take Directional risk factor The calculation formula is:

[0041] In the formula, The angle between the principal direction of the crack and the direction of the maximum principal stress in the soft rock tunnel is given. This is the directional sensitivity adjustment coefficient (ranging from 0.1 to 0.3). This formula increases the directional additional coefficient when the angle between the crack orientation and the principal stress direction is close to 0° or 90°, thereby automatically improving the overall risk score. .

[0042] System preset judgment threshold set And based on the rating The risk is classified into four levels: low, medium, high, and extremely high, and early warning strategies are output accordingly. When the risk is deemed low, the system only makes a routine record; when... When the risk level is determined to be medium, the system will display a pop-up message prompting a retest and a shorter shooting interval; when... When a high-risk condition is identified, the system recommends increased monitoring and activation of downhole support inspection equipment; when If the risk is deemed extremely high, the system will immediately trigger an audible and visual alarm and report the highest priority alarm to the ground platform.

[0043] This embodiment enables automated output of crack identification, parameterized extraction, and risk classification at fixed inspection points. Compared with manual recording, it has the advantages of strong consistency, traceability, and ease of trend analysis.

Claims

1. A method for intelligent identification and extraction of surface cracks in soft rock roadways based on machine vision, characterized in that, Includes the following steps: S1. Image acquisition: Industrial cameras and supplementary lighting devices are deployed on the soft rock tunnel inspection platform or fixed support to acquire images of the soft rock tunnel roof and sidewall surfaces according to the preset shooting interval, and the camera position information is recorded. S2. Calibration and preprocessing: The industrial camera is calibrated for intrinsic parameters and distortion is corrected. The acquired images are then subjected to illumination equalization, noise reduction and contrast enhancement to obtain standardized input images. S3, Crack Recognition: Input the standardized input image into the trained deep learning crack recognition model, and output a pixel-level crack segmentation mask; S4. Crack Extraction: Perform connected component analysis and skeletonization on the crack segmentation mask to obtain the crack centerline and boundary line. S5. Parameter calculation: Based on the camera calibration scale or external scale identifier, convert the crack center line and boundary line from the pixel scale to the actual scale, and calculate the crack length, width and direction. S6. Risk Assessment: Construct a crack risk assessment index system, calculate the risk score based on at least the crack width, length, direction and location information, and output the crack risk level and early warning results accordingly.

2. The intelligent identification and extraction method for surface cracks in soft rock tunnels based on machine vision according to claim 1, characterized in that, In step S1, the industrial camera is a global shutter camera, and the distance between the camera and the surface of the soft rock tunnel is kept within a preset range. Equal-interval acquisition is achieved through trigger synchronization.

3. The intelligent identification and extraction method for surface cracks in soft rock tunnels based on machine vision according to claim 1, characterized in that, In step S1, the supplementary lighting device is a ring-shaped LED supplementary light or a strip array supplementary light, and the supplementary lighting brightness is adaptively adjusted according to the image grayscale histogram.

4. The intelligent identification and extraction method for surface cracks in soft rock tunnels based on machine vision according to claim 1, characterized in that, The illumination equalization in step S2 includes either Retinex-based luminance decomposition or enhancement processing based on adaptive histogram equalization.

5. The intelligent identification and extraction method for surface cracks in soft rock tunnels based on machine vision according to claim 1, characterized in that, The deep learning crack identification model in step S3 is a semantic segmentation network or an instance segmentation network, and it is trained using a soft rock tunnel crack sample library.

6. The intelligent identification and extraction method for surface cracks in soft rock tunnels based on machine vision according to claim 5, characterized in that, The input of the deep learning crack recognition model is a multi-scale image patch, and the output is a crack probability map of the same size as the input. The crack segmentation mask is obtained by thresholding.

7. The intelligent identification and extraction method for surface cracks in soft rock tunnels based on machine vision according to claim 1, characterized in that, In step S4, the skeletonization process uses a thinning algorithm to extract the crack centerline, and then performs breakpoint connection and branch pruning on the centerline to obtain a continuous crack orientation.

8. The intelligent identification and extraction method for surface cracks in soft rock tunnels based on machine vision according to claim 1, characterized in that, In step S5, the crack length is the arc length of the crack centerline at the actual scale, and the crack width is a statistical measure of the distance between crack boundary lines, including at least the maximum width and the average width.

Images