A tunnel lining apparent crack integrity diagnosis method and system based on digital images

Abstract

This invention provides a method and system for the holistic diagnosis of apparent cracks in tunnel lining based on digital images, belonging to the field of tunnel lining maintenance, inspection, and safety assessment technology. Based on the core concept of decomposable and superimposed crack geometry, this invention decomposes complex bifurcated cracks into standardized single-crack units to extract multi-dimensional feature parameters. Through hierarchical clustering for objective delimitation and partial least squares regression to fuse two core indicators—the Tunnel Crack Index (TCI) and the fractal dimension D—it constructs a comprehensive quantitative diagnostic system covering image acquisition, parameter extraction, indicator fusion, and risk classification. This completely eliminates the strong reliance on manual experience in traditional tunnel crack assessment, solving industry problems such as distorted parameter extraction of complex cracks, local assessments disrupting overall connectivity, and one-sided assessments using single indicators. The assessment results show better consistency with technical specifications than traditional methods, providing standardized and reusable quantitative criteria for the safe operation and maintenance of various types of transportation tunnels.

Description

Technical Field

[0001] This invention relates to the field of tunnel lining maintenance, inspection, and safety assessment technology, and in particular to a method and system for the overall diagnosis of apparent cracks in tunnel lining based on digital images. Background Technology

[0002] As a permanent support structure for traffic tunnels, tunnel lining is a core component ensuring the normal operation and maintenance of tunnels. With increasing service life, the lining is highly susceptible to cracking due to the combined effects of internal and external factors such as surrounding rock deformation, stress concentration, and temperature changes. The extent of crack development directly affects the stability and operational safety of the tunnel structure. If timely and accurate detection and scientific assessment are not carried out, the continuous expansion of cracks may lead to major safety accidents such as lining spalling and collapse.

[0003] Existing tunnel crack detection and assessment technologies have several shortcomings: First, the diagnostic scale is vague: traditional methods are mostly based on local crack images for assessment, which severs the overall connectivity of the cracks, leading to one-sided assessment results. However, industry standards clearly require overall assessment based on tunnel rings or lining sections. Second, the extraction of complex crack parameters is not standardized: existing technologies lack unified quantitative standards for core parameters such as length and width of complex cracks such as bifurcations and networks, which can easily lead to parameter distortion. Third, the assessment indicators have limitations: relying solely on the Tunnel Crack Index (TCI) or fractal dimension D makes it difficult to fully reflect the actual impact of cracks on the structure. Fourth, traditional manual inspection relies on experience-based judgment, which is inefficient and highly subjective. Single-device inspection can only target specific types of cracks and cannot achieve batch intelligent inspection. Summary of the Invention

[0004] In view of this, to address the aforementioned shortcomings of existing tunnel crack detection and assessment technologies, this invention provides a method and system for the holistic diagnosis of apparent cracks in tunnel lining based on digital images. Based on the core concept of decomposable and superimposed crack geometry, complex bifurcated cracks are decomposed into standardized single-crack units to extract multi-dimensional feature parameters. Through hierarchical clustering for objective delimitation and partial least squares regression to fuse two core indicators—the Tunnel Crack Index (TCI) and the fractal dimension D—a comprehensive quantitative diagnostic system covering image acquisition, parameter extraction, indicator fusion, and risk classification is constructed. This completely eliminates the strong reliance on manual experience in traditional tunnel crack assessment, solving industry problems such as distorted parameter extraction of complex cracks, local assessments disrupting overall connectivity, and one-sided assessments using single indicators. The assessment results show better consistency with the judgments in the "Technical Specifications for Highway Tunnel Maintenance" than traditional assessment methods, providing standardized and reusable quantitative criteria for the safe operation and maintenance of various types of traffic tunnels.

[0005] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for diagnosing the overall integrity of apparent cracks in tunnel lining based on digital images, comprising the following steps: Step S01: Image acquisition: Survey and obtain the basic parameters of the target tunnel, acquire the original image of the tunnel lining appearance, and ensure that the overlap of adjacent images meets the requirements for panoramic stitching. Step S02: Image preprocessing: The original image of the tunnel lining appearance acquired in step S01 is stitched together by feature matching to obtain a panoramic image of the tunnel lining, and a binary image of the lining crack is obtained by semantic segmentation. Step S03: Quantitative extraction of crack parameters: Based on the concept of decomposable superposition of crack geometry, the crack skeleton is extracted and the endpoints and bifurcation points are identified. Complex bifurcation cracks are decomposed into single crack units, and the core parameters of crack length, width, angle, area, density, and spatial distribution characteristics are quantitatively extracted. Step S04: Calculation of core indicators: Calculate the tunnel crack index (TCI) and crack fractal dimension (D) respectively; Step S05: Unified Assessment and Risk Classification: Based on hierarchical clustering, objectively determine the boundary values ​​between TCI and D. Use partial least squares regression to construct a unified TCI-D assessment model, classifying risk levels from I to IV according to TCI-D values, corresponding to different degrees of crack development. The model expression is as follows: Among them, TCI-D is the lining crack assessment index. and The TCI and D weights for lining cracks, For constant terms; Step S06: Results Output and Application: Compare and verify the assessment results with the current highway tunnel maintenance specifications, and output corresponding maintenance and treatment recommendations based on the risk classification.

[0006] Secondly, the present invention provides a digital image-based system for diagnosing the overall integrity of apparent cracks in tunnel lining, used to perform the above-mentioned diagnostic method, including: The basic acquisition module is used to investigate and obtain the basic parameters of the target tunnel, acquire the original images of the tunnel lining appearance, and ensure that the overlap of adjacent images meets the requirements for panoramic stitching. The image preprocessing module is used to obtain a panoramic image of the tunnel lining by feature matching and stitching the original image of the tunnel lining acquired by the basic acquisition module, and to obtain a binary image of the lining cracks by semantic segmentation. The crack parameter extraction module, based on the concept of decomposable superposition of crack geometry, is used to extract crack skeleton and identify endpoints and bifurcation points. It decomposes complex bifurcation cracks into single crack units and quantitatively extracts the core parameters of crack length, width, angle, area, density, and spatial distribution characteristics. The core index calculation module calculates the tunnel crack index (TCI) and crack fractal dimension (D) respectively. The unified assessment and grading module objectively determines the boundary values ​​between TCI and D based on hierarchical clustering. A unified TCI-D assessment model is constructed using partial least squares regression, classifying risk levels from I to IV based on TCI-D values, corresponding to different degrees of crack development. The model expression is as follows: Among them, TCI-D is the lining crack assessment index. and The TCI and D weights for lining cracks, For constant terms; The results application module compares and verifies the assessment results with current highway tunnel maintenance standards, and outputs corresponding maintenance and treatment recommendations based on risk classification.

[0007] Compared with the prior art, the present invention has the following beneficial effects: Advantages of objectivity in assessment: It completely eliminates the strong reliance on human experience in traditional tunnel crack assessment. The core boundary value is objectively determined based on 156 sets of measured tunnel lining samples through hierarchical clustering. There is no need to adjust the parameter weights according to specific projects. The assessment process is free from subjective intervention and the results are reproducible.

[0008] Advantages in the accuracy of complex crack assessment: The innovative core concept of decomposable and superimposed crack geometry can decompose complex cracks such as bifurcations and networks into standardized single crack units, solving the pain point of distortion and inaccuracy in the calculation of complex cracks by traditional parameter extraction methods. The parameter extraction covers multi-dimensional features such as length, width, angle, density, and spatial distribution, and fully characterizes the actual development state of cracks.

[0009] The comprehensive advantages of this assessment include: the innovative fusion of two core indicators—the Tunnel Crack Index (TCI), which is highly sensitive to the geometric characteristics of local cracks, and the fractal dimension D, which characterizes the overall distribution of cracks—and the goodness of fit of the unified TCI-D model constructed through partial least squares regression. R 2 ≥0.997, while taking into account both local high-risk cracks and overall disease distribution characteristics, avoiding the one-sidedness of evaluation by a single indicator.

[0010] Advantages in engineering practicality: The assessment results are more consistent with the judgment of the "Technical Specification for Highway Tunnel Maintenance" JTGH12-2015 than the traditional TDI-C / HSR and TFRI assessment methods. It will not have the problem of over-rating caused by the TDI-C / HSR method relying too much on crack length, nor the problem of under-rating caused by improper weight adjustment of the TFRI method. The output of the four-level risk classification can directly correspond to different maintenance and treatment strategies, and is suitable for the routine maintenance and inspection scenarios of various traffic tunnels. Attached Figure Description

[0011] Figure 1 This is a flowchart of the present invention; Figure 2The fitting result between the observed values ​​and the model evaluation values ​​is the actual value calculated based on the normalized Euclidean distance, and the evaluation value is the calculation result output by the model. Figure 3 The residual value is the difference between the observed value and the model's evaluation value. The maximum residual value does not exceed 10% of the corresponding evaluation value of the model, reflecting the stability of the model. Figure 4 The grading chart is divided into four categories by taking the midpoint of the intersection interval as the dividing value. The upper part is the clustering result of the corresponding lining segment, and the lower part is the scoring and classification result. Detailed Implementation

[0012] The technical solutions of the specific embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the specific embodiments described are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0013] like Figure 1 As shown, the present invention provides a method for diagnosing the overall integrity of apparent cracks in tunnel lining based on digital images, comprising the following steps: Step S01: Image acquisition: Survey and obtain the basic parameters of the target tunnel, acquire the original image of the tunnel lining appearance, and ensure that the overlap of adjacent images meets the requirements for panoramic stitching. Step S02: Image preprocessing: The original image of the tunnel lining appearance acquired in step S01 is stitched together by feature matching to obtain a panoramic image of the tunnel lining, and a binary image of the lining crack is obtained by semantic segmentation. Step S03: Quantitative extraction of crack parameters: Based on the concept of decomposable superposition of crack geometry, the crack skeleton is extracted and the endpoints and bifurcation points are identified. Complex bifurcation cracks are decomposed into single crack units, and the core parameters of crack length, width, angle, area, density, and spatial distribution characteristics are quantitatively extracted. Step S04: Calculation of core indicators: Calculate the tunnel crack index (TCI) and crack fractal dimension (D) respectively; Step S05: Unified Assessment and Risk Classification: Based on hierarchical clustering, objectively determine the boundary values ​​between TCI and D. Use partial least squares regression to construct a unified TCI-D assessment model, classifying risk levels from I to IV according to TCI-D values, corresponding to different degrees of crack development. The model expression is as follows: Among them, TCI-D is the lining crack assessment index. and The TCI and D weights for lining cracks, For constant terms; Step S06: Result Output and Application: Compare and verify the assessment results with current highway tunnel maintenance standards, and output corresponding maintenance and treatment recommendations based on risk classification. For example, apply the established diagnostic method to actual tunnel engineering. By acquiring images of the target tunnel lining, complete parameter extraction, assessment, and classification according to the above steps, and verify the effectiveness of the results by comparing them with industry standards ("Technical Specification for Highway Tunnel Maintenance" JTGH12-2015); based on the risk classification results, provide targeted recommendations for tunnel crack repair and reinforcement.

[0014] This technical solution constructs a complete end-to-end diagnostic logic, encompassing image acquisition, preprocessing, parameter extraction, index calculation, risk classification, and result application. Its core principle is based on the decomposable and superimposed nature of crack geometry to accurately extract complex crack parameters. Through hierarchical clustering for objective delimitation and partial least squares regression fusing TCI and D dual indicators, it achieves a quantitative assessment of crack development status from local to global perspectives. This completely eliminates the reliance on human experience in traditional assessments, addressing the core pain points of existing technologies such as distorted complex crack parameters, one-sided single-indicator assessments, and local assessments severing overall connectivity. The assessment process is standardized, reproducible, and adaptable to various traffic tunnel scenarios.

[0015] In this invention, step S01 may specifically include the following steps: S01-01: Obtain and organize quantifiable information such as tunnel type, design dimensions, service life, construction technology, and geological conditions through surveys. The evaluation results can be adapted to the tunnel's own engineering background to further enhance the engineering reference value of the results and avoid mechanical evaluations that are divorced from the actual scenario. S01-02: An image acquisition system consisting of a mobile tunnel inspection vehicle equipped with a CCD camera array is used. The CCD camera array has a field of view of 210°, covering the tunnel arch, left and right shoulders, and left and right waist areas. The inspection vehicle moves at a constant speed along the tunnel, acquiring original images of the tunnel lining surface. The acquired image resolution meets the requirements for sub-millimeter level crack identification, and during the acquisition process, the overlapping areas of adjacent images meet the pixel-level stitching requirements. A single trip can complete full-section, blind-spot-free acquisition without the need for re-shooting. The high resolution can identify minute cracks, balancing acquisition efficiency and detection accuracy, and meeting the efficiency requirements of large-scale tunnel maintenance and inspection.

[0016] In this invention, step S02 specifically includes the following steps: Step S02-01: The original image of the tunnel lining appearance acquired in step S01 is preprocessed, and orientation correction, tilt correction and orthorectification are completed in sequence to eliminate image distortion caused by the tunnel arch structure and restore the true size of the cracks. Step S02-02: The SURF (Speeded Up Robust Features) algorithm is used to extract similar feature operators from adjacent images. High-resolution panoramic images of tunnel lining are stitched together through feature matching, alignment and fusion. Steps S02-03: Use a semantic segmentation model to segment the cracks in the panoramic image to obtain a binary image of the lining cracks; connect the fracture cracks through morphological closing operations, filter out small target noise with an area of ​​less than 10 pixels, and ensure the connectivity and subjectivity of the cracks.

[0017] This technical solution eliminates arch structure distortion and restores the true size of cracks through three-level correction. The SURF algorithm is used to achieve high-precision panoramic stitching, and morphological closing operations simultaneously connect fracture cracks and filter out low-noise. The preprocessed data closely matches the actual state of the lining, with minimal stitching error. The denoising process does not damage the connectivity and main morphology of the cracks, providing a high-quality data source for subsequent parameter extraction.

[0018] This invention addresses the industry pain points of distorted parameter extraction and fragmented assessment scales in complex cracks such as bifurcated and mesh-like cracks in tunnel linings. It proposes a method for standardized processing and quantitative analysis of crack geometry. The core logic is that any complex lining crack can be decomposed into several standardized single-crack basic units. After accurate parameter quantification at the unit level, the overall characteristics of the crack are restored through regular superposition, simultaneously ensuring both local calculation accuracy and the completeness of overall defect assessment. Therefore, in this invention, step S03 may specifically include the following steps:

[0019] Step S03-01: Obtain the crack morphology skeleton using the skeleton extraction method, subtract burr branches with a length of less than 5 pixels, and obtain the crack skeleton point set. S ; Traverse the skeleton point set to construct an 8-neighbor adjacency list, and detect endpoints N8(S( x , y ))=1 and the bifurcation point N8(S(·))>3; Step S03-02: Based on the endpoints and bifurcation points, classify the cracks into single cracks and bifurcation cracks, and decompose the bifurcation cracks into multiple single crack units. Step S03-03: Based on the decomposed single crack element, quantitatively extract multidimensional parameters: Length: Calculated by multiplying the sum of Euclidean distances between adjacent points along the skeleton path by the image resolution. r The formula is:

[0020] in( x , y） The coordinates of the skeleton points; Length is a fundamental quantitative indicator of crack development scale, directly reflecting the extent and connectivity of lining damage extension, and is a core input item of TCI. The present invention uses a detachable superposition method to accurately calculate the length, solving the technical problem of distortion in the calculation of complex crack length, and providing a basis for the overall crack damage assessment.

[0021] Width: Calculates the maximum distance from the skeleton pixel to the edge of the connected component. W max The actual maximum width is w max =2rW max; Width is the core control threshold indicator for the safety classification of lining structures. It directly corresponds to the stress concentration degree of the lining, the waterproof integrity and the reduction of the bearing capacity. It provides core weighting parameters for TCI calculation and solves the technical problem that the existing assessment is insufficient in distinguishing the risk of cracks.

[0022] Angle: The first principal component vector is calculated using principal component analysis and determined using the four-quadrant arctangent function. The formula is: θ=mod ( arctan2 ( v y , v x ) × π / 180 , 180 ), in( v x , v y () represents the eigenvector corresponding to the largest eigenvalue of the covariance matrix. arctan2 Represents the arctangent function in the four quadrants. mod The modulus operation limits the crack angle to 0°~180°; Crack angle is the only parameter that quantifies the relative relationship between crack direction and principal stress direction of the lining. It can solve the problem of misjudging the risk of cracks of the same size caused by neglecting this dimension in existing technologies, and achieve accurate mapping of the risk nature of cracks and correction of TCI calculation weights. At the same time, it can predict crack propagation trends and invert the causes of defects. Crack distribution characteristics are the only spatial vector parameters that match the stress characteristics of the annular zone of the tunnel lining. It can solve the problem that the existing average assessment of the whole section is prone to masking local high risks, and achieve limit state assessment based on structural weak areas. At the same time, it can identify the systemic failure risk of the lining force transmission system and locate systemic structural defects, providing a targeted basis for refined maintenance.

[0023] Area: Calculated based on the number of crack pixels N in the binary image, using the following formula: A = r²N; The area is used to quantify the effective bearing area loss rate of the lining in a two-dimensional dimension, which makes up for the limitations of one-dimensional geometric parameter evaluation. It can accurately distinguish the degree of damage of different crack morphologies, provide a basis for the normalization evaluation of crack damage, and ensure the comparability of evaluation results of lining sections of different sizes.

[0024] Density: The formula is D= L total / H×W×r 2, in L total H represents the total crack length, and H and W represent the image height and width. The density conforms to the legal definition in the "Technical Specification for Highway Tunnel Maintenance", quantifies the overall degree of disease development in the lining section, realizes the overall assessment by lining ring / entire section, solves the technical problem of traditional local assessment of the overall connectivity of fracture cracks, and provides core input indicators for risk classification and clustering.

[0025] Distribution characteristics: The image was divided into 5 regions in a ratio of 12:7:4:7:12, and the crack distribution was represented by Boolean vectors.

[0026] Step S03 first extracts the crack skeleton and removes burrs. Endpoints and bifurcation points are identified through 8-neighborhood detection, decomposing complex bifurcation / network cracks into standardized single crack units. Then, six core parameters—length, width, angle, area, density, and distribution characteristics—are calculated. Based on the concept of decomposable and superimposed crack geometry, endpoint and bifurcation point detection transforms bifurcation cracks into single crack units, ensuring the precision and accuracy of length, width, and other parameter calculations, and avoiding errors caused by global parameters. A single crack is defined as a crack without any overall bifurcation structure. The crack type is determined by detecting the endpoints / bifurcation points obtained from the 8-neighborhood of the crack skeleton. When the crack is complex, the feature point set composed of endpoints and bifurcation points is traversed. When there is one and only one path between two points that does not contain any other feature points, this path is a single crack. The parameters of each single crack are then calculated, and the parameters of the complex crack are obtained by superimposing the parameters of each single crack. It solves the industry pain point of traditional parameter extraction methods being distorted and inaccurate in calculating cracks with complex shapes. The multi-dimensional parameters fully cover the geometric features, stress features and distribution features of cracks, and the characterization accuracy is far higher than that of single-parameter evaluation methods.

[0027] In this invention, step S04 specifically includes the following steps: Step S04-01: TCI Refinement Calculation: Decompose the crack into several small segments according to the bifurcation point, calculate the length and width of each crack segment, and substitute them into the TCI calculation formula: ; in, AThe area to be lining; n The number of cracks; t （k） For the first k The width of the crack; l （k） For the first k The length of the crack, For the first k Strip crack normal and x i The included angle of the axis; For the first k Strip crack normal and x j The included angle of the axis; α The weighted correlation coefficient for crack width; β The weighted correlation coefficient is the crack length. Step S04-02: Fractal dimension D calculation: The box dimension method is used to capture the self-similarity of the crack. The maximum box size is selected as 64. The fractal dimension D of the crack is calculated and verified by classical fractal structures (such as Koch curves). The error is controlled within 0.03.

[0028] In step S04, the TCI calculation decomposes the crack into small segments, and the combined length, width, and angle parameters are superimposed to obtain the total value. The fractal dimension is calculated using the box-counting method, with the error controlled within 0.03. TCI is highly sensitive to the stress concentration of local cracks, and the fractal dimension can effectively capture the self-similarity characteristics of the overall crack distribution. Both types of indicators have high calculation accuracy, characterizing crack damage from two dimensions: local high risk and overall development degree, avoiding one-sided assessment.

[0029] In this invention, in step S05, crack density, total area, number of intersections, and maximum width are selected as indicators. After Z-score standardization, hierarchical clustering is used to determine the TCI and D boundary values, clarifying the crack characteristics of each zone. The boundary values ​​are obtained based on clustering of measured data, without any subjective human intervention. There is no need to adjust parameter weights for different engineering scenarios, and the evaluation results are highly universal and objective, avoiding the over- or under-rated problems caused by improper weight adjustment in traditional methods.

[0030] In this invention, in step S05, =0.5569, =0.8942, =2.0371, balancing the sensitivity of TCI to crack geometry parameters with the ability of D to characterize distribution characteristics, the risk classification standard was validated through 156 tunnel lining samples, showing good consistency with the actual crack development state. The TCI cutoff value is 7.56×10. -5 The cutoff value for D is 1.17, indicating a good model fit. R²With a score of ≥0.997, the evaluation accuracy is extremely high. The parameters are fixed values ​​that can be directly called, eliminating the need to retrain the model for specific projects. This makes it easy to implement and highly user-friendly.

[0031] In this invention, in step S05: When 0 ≤ TCI-D < 2.3, it is classified as Level I risk, corresponding to very mild crack development; When 2.3 ≤ TCI-D < 3.6, it is classified as Level II risk, corresponding to a small number of dispersed cracks; When 3.6 ≤ TCI-D < 4.9, it is classified as Level III risk, corresponding to significant crack expansion; When TCI-D≥4.9, it is classified as Level IV risk, corresponding to the degree of crack development.

[0032] In step S05, the hierarchical logic is clear, with each level directly corresponding to different maintenance and treatment strategies. Frontline maintenance personnel can carry out their work directly based on the hierarchical results without needing a professional background, making the project highly practical.

[0033] The above step S05 can be specifically as follows: S05-01: Four-zone delineation: Crack density, total area, number of intersections, and maximum width are selected as indicators. After Z-score standardization, hierarchical clustering is used to determine the TCI boundary value (7.56×10). -5 The boundary value between TCI and D (1.17) clarifies the crack characteristics of each zone. Using the four-zone division of TCI and D, the crack characteristics within the lining section can be described. For example... Figure 4 As shown, the grades are the results of a comprehensive evaluation based on TCI_D.

[0034] S05-02: TCI-D Model Construction: Based on partial least squares regression, fusing the advantages of TCI and D, the model expression is as follows: ,in =0.5569, =0.8942, =2.0371; The goodness of fit is calculated from the observations using Euclidean distance. R² ≥0.997, as shown in Figure 2-3. Here, the Euclidean distance is a value calculated based on the spatial distance of each sample parameter, called the observed value, which can be understood as the actual calculation result. The accuracy of the model's calculation is verified by fitting the observed values ​​with the evaluation values ​​calculated by the model.

[0035] S05-03: Risk Classification: Based on the distribution of TCI-D values, risks are classified into levels I-IV: Level I (0≤TCI-D<2.3), Level II (2.3≤TCI-D<3.6), Level III (3.6≤TCI-D<4.9), and Level IV (TCI-D≥4.9), corresponding to extremely light crack development, a small number of scattered cracks, significant expansion, and highly developed cracks, respectively.

[0036] The present invention also provides a digital image-based system for diagnosing the overall integrity of apparent cracks in tunnel lining, used to perform the above-mentioned diagnostic method, comprising: The basic acquisition module is used to investigate and obtain the basic parameters of the target tunnel, acquire the original images of the tunnel lining appearance, and ensure that the overlap of adjacent images meets the requirements for panoramic stitching. The image preprocessing module is used to obtain a panoramic image of the tunnel lining by feature matching and stitching the original image of the tunnel lining acquired by the basic acquisition module, and to obtain a binary image of the lining cracks by semantic segmentation. The crack parameter extraction module, based on the concept of decomposable superposition of crack geometry, is used to extract crack skeleton and identify endpoints and bifurcation points. It decomposes complex bifurcation cracks into single crack units and quantitatively extracts the core parameters of crack length, width, angle, area, density, and spatial distribution characteristics. The core index calculation module calculates the tunnel crack index (TCI) and crack fractal dimension (D) respectively. The unified assessment and grading module objectively determines the boundary values ​​between TCI and D based on hierarchical clustering. A unified TCI-D assessment model is constructed using partial least squares regression, classifying risk levels from I to IV based on TCI-D values, corresponding to different degrees of crack development. The model expression is as follows: Among them, TCI-D is the lining crack assessment index. and The TCI and D weights for lining cracks, For constant terms; The results application module compares and verifies the assessment results with current highway tunnel maintenance standards, and outputs corresponding maintenance and treatment recommendations based on risk classification.

[0037] This diagnostic system systematizes and productizes the aforementioned diagnostic methods, and can be directly integrated into existing tunnel mobile detection equipment to achieve integrated automated operation from image acquisition to evaluation result output. It eliminates the need for manual step-by-step processing and significantly improves detection efficiency compared to traditional manual evaluation.

[0038] The technical solution of the present invention will be clearly and thoroughly described below with reference to specific embodiments.

[0039] Example 1 This embodiment is a core model construction example of the above-mentioned method for the overall diagnosis of apparent cracks in tunnel lining based on digital images. It aims to establish a unified TCI-D assessment index and supporting risk classification system through the construction of a tunnel lining crack sample library, multi-dimensional parameter correlation analysis, objective zoning, and multi-index fusion modeling. This addresses the technical problems of traditional assessment methods, such as the reliance on engineering experience for boundary values, limitations of single-index assessment, and the inability to simultaneously consider the geometric characteristics and distribution features of cracks. The specific implementation process is as follows: 1. Spatial construction of tunnel lining crack samples To construct a comprehensive feature sample space covering different crack development morphologies and varying degrees of damage, this embodiment selects two highway tunnels in Japan with well-developed cracks during their service life as data sources: 1.1 Riyue Tunnel: Built in 1982, the tunnel is 780m long, and cracks of varying degrees have appeared in all 65 lining sections along the line; 1.2 Yiyuan Tunnel: There are a total of 94 lining sections along the line, and each lining section has cracks and defects of varying degrees.

[0040] A total of 159 crack images and corresponding parameter data of the lining sections of the two tunnels were collected to form the initial sample space; among them, no crack development was found in the lining sections No. 28, 89 and 90, which were removed as abnormal data, and finally 156 groups of valid lining crack samples were obtained.

[0041] In this embodiment, core crack parameters are selected as the basic indicators for the sample, including: Tunnel Crack Index (TCI), fractal dimension D, D-fit degree, total number of cracks, number of single cracks, number of intersecting cracks, number of endpoints, number of intersections, total crack length, crack area, maximum crack width, crack density, and crack distribution characteristics, totaling 12 indicators. Among them, D-fit degree, total number of cracks that can be converted from other indicators, and non-numerical crack distribution characteristics are removed, and 9 effective quantitative indicators are retained for subsequent analysis. The basic data of the sample are shown in Table 1 below.

[0042] Table 1 Basic Parameters of Tunnel Lining Crack Samples

[0043] Data distribution verification showed that the values ​​of the nine effective indicators were mostly concentrated between the upper and lower quartiles. The outliers outside the 1.5 IQR range all corresponded to lining sections with severe crack development. The data distribution characteristics were completely consistent with the actual crack development pattern of tunnel lining, and the sample library was sufficiently representative.

[0044] 2. Sample data preprocessing and correlation analysis 2.1 Data Standardization Processing Because the numerical values ​​of the various indicators differ significantly in terms of dimensions and magnitudes, they cannot be directly used in clustering and regression calculations. Therefore, this embodiment employs the Z-score standardization method to process all indicator data, eliminating scale differences between different data points while preserving the original sample distribution characteristics. The standardization formula is as follows:

[0045] In the formula, x represents the original data, μ represents the sample mean, and σ represents the sample standard deviation.

[0046] 2.2 Pearson Correlation Analysis Pearson correlation analysis was used to quantify the linear correlation of crack parameters. The formula for calculating the Pearson correlation coefficient is as follows:

[0047] Where r is the Pearson correlation coefficient, which represents the degree of linear correlation between the two variables. and Let be the i-th data point of variables X and Y.

[0048] Correlation analysis results show that a correlation coefficient of 0.8–1 indicates a strong correlation, while 0.6–0.8 indicates a relatively strong correlation. Specifically, TCI shows a strong correlation with crack density, total crack length, total crack area, number of intersections, number of endpoints, and maximum crack width; fractal dimension D shows a relatively strong correlation with maximum crack width; crack density is completely correlated with total crack length, consistent with the definition of linear density. These results provide data support for subsequent selection of zoning indicators and the complementary integration of TCI and D.

[0049] 3. Objective partitioning of TCI and D four-partitions based on hierarchical clustering 3.1 Selection of Core Indicators for Each Zone Traditional four-zone assessment methods use engineering experience values ​​to define the boundary between TCI and D, which suffers from strong subjectivity and is prone to information loss. To achieve an objective definition of the boundary values, this embodiment selects four core indicators that are highly correlated with both TCI and D as the basis for zoning, based on correlation analysis results: crack density, total crack area, number of intersections, and maximum crack width.

[0050] 3.2 Hierarchical Clustering Partition Calculation This embodiment employs a hierarchical clustering algorithm based on the average linkage method, using the squared Euclidean distance as the basis for calculating sample distance to achieve objective classification of sample data. The formula for calculating the average distance between clusters is as follows:

[0051] in, The distance between clusters,A and B There are two clusters. x and y These are samples within the cluster.

[0052] Based on the silhouette coefficient and DBI index verification, when the TCI-related indicators are classified into two categories, the silhouette coefficient is 0.71, indicating the highest sample similarity within the cluster, and the binary classification result demonstrates optimal rationality. Based on the hierarchical clustering results, the final TCI cutoff value is determined to be 7.56 × 10⁻⁶. -5 By combining the fitting functions of TCI and D, the boundary value of D was calculated to be 1.17, thus completing the objective definition of the four zones of TCI-D. The crack characteristics of each zone are as follows:

[0053] First interval: TCI < 7.56 × 10 -5 Furthermore, D < 1.17, indicating that the cracks are single cracks or a few bifurcated cracks with obvious directionality; Second interval: TCI ≥ 7.56 × 10 -5 Furthermore, D < 1.17, and the cracks are mainly fine, long, narrow, and densely networked branching cracks; Third interval: TCI < 7.56 × 10 -5 Furthermore, D≥1.17 indicates a larger crack width and fewer bifurcated structures; Fourth interval: TCI ≥ 7.56 × 10 -5 Furthermore, D≥1.17 indicates highly developed cracks, forming a complex network-like intersecting structure.

[0054] 4. Construction of a unified TCI-D evaluation model based on partial least squares regression To address the technical issues that a single TCI or D index is insufficient to comprehensively assess crack damage and that the traditional four-zone method lacks a unified quantitative index, this embodiment is based on partial least squares regression, integrating the high sensitivity of TCI to the core geometric parameters of cracks and the strong characterization ability of D to crack distribution characteristics to construct a unified TCI-D assessment model.

[0055] 4.1 Basic Expressions of the Model The basic expression of the TCI-D unified evaluation model is as follows:

[0056] Among them, TCI-D is the lining crack assessment index. and The TCI and D weights for lining cracks, This is a constant term.

[0057] 4.2 Establishment of Model Observations Based on 156 valid samples, two outlier samples with D < 1 were removed, retaining 154 valid samples. A disease-free baseline sample was added; the theoretical fractal dimension of the disease-free lining is 1, and the TCI value is 0. Based on the Z-score standardized data, the Euclidean distance between the 154 valid samples and the disease-free baseline sample was calculated as the observed value of TCI-D. The Euclidean distance calculation formula is as follows:

[0058]

[0059] in, For Euclidean distance and For the i-th sample and the disease-free sample in the standardized TCI sample space, and Let i be the i-th sample and the disease-free sample in the standardized D-sample space.

[0060] Weight coefficients calculated by partial least squares regression and The values ​​are 0.5569 and 0.8942 respectively, with a constant of 2.0371. To ensure the objectivity of the crack assessment index, the observed values ​​of TCI-D were statistically analyzed based on Euclidean distance, and the calculation results are as follows: Figure 2-3 As shown. The calculated TCI-D observation range is [0, 6.19], and the evaluation value is the result of partial least squares regression calculation. The two have a good fit. R 2 When the slope K is 0.997, the slope K is 0.9968. The residual is the difference between the observed value and the estimated value. The average residual is -6.45 × 10⁻⁶. -7 The maximum value is 0.35, which is less than 10% of the corresponding sample evaluation value, indicating a good fit.

[0061] 5. Establishment of risk classification standards for tunnel lining cracks (Levels I-IV) The tunnel was divided into four categories based on the midpoint of the intersection intervals, as shown in the figure. Levels I-IV represent tunnel crack risk from high to low. Level I indicates very minor crack development or only very fine cracks, with the overall structure stable. Level II indicates a small number of scattered cracks in the tunnel lining, showing no obvious signs of expansion. Level III indicates the presence of densely intersecting cracks or wide cracks in the tunnel lining, with a clear trend of crack expansion requiring close monitoring to prevent further expansion. Level IV represents highly developed tunnel cracks forming a complex network, requiring urgent repair of the tunnel lining to prevent further risks. The figure shows that the scores for linings in the second and third intervals, such as S65, S111, S139, and S149, are more reasonable than single-index evaluation, and this aligns with the objective fact that the widest single crack has a greater impact on the lining than multiple small cracks.

[0062] Analysis of Comparative Results in Practical Applications Based on the model established in Embodiment 1 above, another tunnel is applied in practice, specifically: After obtaining the binary image of the tunnel lining, the development status of cracks in the tunnel lining was quantitatively analyzed and risk-classified based on the TCI-D model proposed in this invention. Most cracks in the tunnel lining were single cracks, and the discontinuity of cracks was caused by mutual obstruction from multiple pipelines. The quantitative results of the cracks are shown in Table 2. In this tunnel section lining, lining No. 11 had the highest TCI-D score of 2.37, classifying it as Class II. Linings No. 6, 9, and 18 were also Class II, with multiple scattered cracks detected. Overall, the crack development in this section was low, with only a few cracks observed, and the overall structural state of the tunnel remained stable.

[0063] Table 2 Crack parameters and evaluation results

[0064] The following comparative analysis was conducted against other assessment methods to verify the practicality and effectiveness of the TCI-D method. Table 3 shows the statistical results of the disease. TDI-C and HSR are assessment methods based on fractal theory combined with basic crack parameters, while TFRI is an assessment index of multidimensional crack parameters constructed based on the crack tensor theory of TCI. The standards described in the table refer to the lining crack assessment methods specified in the Chinese "Technical Specification for Highway Tunnel Maintenance" (JTG H12-2015). Due to the different diagnostic scales, the crack width and fractal dimension used by TDI-C and HSR are cumulative values, and similarity conversion was performed based on image size in this study.

[0065] The evaluation results indicate that all TDI-C and HSR assessments were classified as Level 1, primarily due to over-rating caused by biases in the diagnostic scales. Furthermore, the lack of standardization led to an over-reliance on crack length, thus reducing the sensitivity of crack assessment. TFRI assessments typically fall between Level 3 and Level 4. Although it comprehensively considers the influence of multidimensional crack parameters on the lining failure process and can describe cracks in more detail, the allocation of parameter weighting factors needs to be analyzed in conjunction with specific engineering conditions. TCI-D assessment results are highly consistent with the standard, typically falling between Level 1 and Level 2. TCI-D integrates the advantages of both TCI and D assessment schemes, fully considering the local differences and overall spatial distribution characteristics of complex cracks in the lining section, thus providing a systematic diagnostic assessment standard for tunnel crack evaluation.

[0066] Table 3 Comparison of Evaluation Results

[0067] The above are merely preferred embodiments of the present invention; however, the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concept, should be covered within the scope of protection of the present invention.

Claims

1. A method for diagnosing integrity of apparent cracks in a tunnel lining based on digital images, characterized by, Includes the following steps: Step S01: Image acquisition: Survey and obtain the basic parameters of the target tunnel, acquire the original image of the tunnel lining appearance, and ensure that the overlap of adjacent images meets the requirements for panoramic stitching. Step S02: Image preprocessing: The original image of the tunnel lining appearance acquired in step S01 is stitched together by feature matching to obtain a panoramic image of the tunnel lining, and a binary image of the lining crack is obtained by semantic segmentation. Step S03: Quantitative extraction of crack parameters: Based on the concept of decomposable superposition of crack geometry, the crack skeleton is extracted and the endpoints and bifurcation points are identified. Complex bifurcation cracks are decomposed into single crack units, and the core parameters of crack length, width, angle, area, density, and spatial distribution characteristics are quantitatively extracted. Step S04: Calculation of core indicators: Calculate the tunnel crack index (TCI) and crack fractal dimension (D) respectively; Step S05: Unified assessment and risk classification: Based on hierarchical clustering, objectively determine the boundary values ​​of TCI and D, and use partial least squares regression to construct a unified TCI-D assessment model. Divide the risk levels into I-IV according to the TCI-D values, corresponding to different degrees of crack development. The model expression is where TCI-D is a lining crack assessment index, and TCI and D weight of the lining crack, is a constant term; Step S06: Results Output and Application: Compare and verify the assessment results with the current highway tunnel maintenance specifications, and output corresponding maintenance and treatment recommendations based on the risk classification.

2. The method according to claim 1, wherein, A mobile tunnel inspection vehicle equipped with an image acquisition device consisting of a CCD camera array is used to acquire original images of the tunnel lining surface. The field of view of the CCD camera array covers the tunnel arch, left and right shoulders, and left and right waist areas, and the resolution of the acquired images meets the requirements for sub-millimeter level crack identification.

3. The method according to claim 1, wherein, Step S02 specifically includes the following steps: Step S02-01: Preprocess the original image of the tunnel lining appearance acquired in step S01, and sequentially complete the orientation correction, tilt correction and orthorectification to eliminate image distortion caused by the tunnel arch structure and restore the true size of the cracks. Step S02-02: The SURF algorithm is used to extract similar feature operators from adjacent images, and high-resolution panoramic images of tunnel lining are stitched together through feature matching, alignment and fusion. Steps S02-03: Use a semantic segmentation model to segment the cracks in the panoramic image to obtain a binary image of the lining cracks; connect the fracture cracks through morphological closing operations, filter out small target noise with an area of ​​less than 10 pixels, and ensure the connectivity and subjectivity of the cracks.

4. The method for diagnosing the overall integrity of apparent cracks in tunnel lining based on digital images according to claim 1, characterized in that, Step S03 specifically includes the following steps: Step S03-01: Obtain the crack morphology skeleton using the skeleton extraction method, subtract burr branches with a length of less than 5 pixels, and obtain the crack skeleton point set. S ; Traverse the skeleton point set to construct an 8-neighbor adjacency list, and detect the endpoints N8(S( x , y ))=1 and the bifurcation point N8(S(·))>3; Step S03-02: Based on the endpoints and bifurcation points, classify the cracks into single cracks and bifurcation cracks, and decompose the bifurcation cracks into multiple single crack units. Step S03-03: Based on the decomposed single crack element, quantitatively extract multidimensional parameters: Length: sum of Euclidean distance between adjacent points of the skeleton path, multiplied by the image resolution r The formula is: ； wherein (A) is x , y） is the backbone point coordinate; Width: Calculate the maximum distance of a skeleton pixel to the edge of the connected component W max The actual maximum width is w max =2rW max; Angle: The first principal component vector is calculated using principal component analysis and determined using the four-quadrant arctangent function. The formula is: θ =mod ( arctan2 ( v y , v x ) × π / 180 , 180 ), in( v x , v y () represents the eigenvector corresponding to the largest eigenvalue of the covariance matrix. arctan2 Represents the arctangent function in the four quadrants. mod The modulus operation limits the crack angle to 0°~180°; Area: Calculated based on the number of crack pixels N in the binary image, using the following formula: A = r²N; Density: The formula is D= L total / H×W×r 2, in L total H represents the total crack length, and H and W represent the image height and width. Distribution characteristics: The image was divided into 5 regions in a ratio of 12:7:4:7:12, and the crack distribution was represented by Boolean vectors.

5. The method for diagnosing the overall integrity of apparent cracks in tunnel lining based on digital images according to claim 1, characterized in that, Step S04 specifically includes the following steps: Step S04-01: TCI Calculation: Decompose the crack into several smaller segments according to the bifurcation point, calculate the length and width of each segment, and substitute them into the TCI calculation formula: ； in, A The area to be lining; n The number of cracks; t （k） For the first k The width of the crack; l （k）k For the first k The length of the crack, For the first k Strip crack normal and x i The included angle of the axis; For the first k Strip crack normal and x j The included angle of the axis; α The weighted correlation coefficient for crack width; β The weighted correlation coefficient is the crack length. Step S04-02: Fractal dimension D calculation: The box dimension method is used to capture the self-similarity of the crack. The maximum box size is selected as 64. The fractal dimension D of the crack is calculated and verified by classical fractal structure. The error is controlled within 0.

03.

6. The method for diagnosing the overall integrity of apparent cracks in tunnel lining based on digital images according to claim 1, characterized in that, In step S05, crack density, total area, number of intersections, and maximum width are selected as indicators. After Z-score standardization, hierarchical clustering is used to determine the TCI boundary value and D boundary value to clarify the crack characteristics of each zone.

7. The method for diagnosing the overall integrity of apparent cracks in tunnel lining based on digital images according to claim 1, characterized in that, In step S05, =0.5569, =0.8942, =2.0371, TCI cutoff value is 7.56×10 -5 The threshold value for D is 1.

17.

8. The method for diagnosing the overall integrity of apparent cracks in tunnel lining based on digital images according to claim 1, characterized in that, In step S05: When 0 ≤ TCI-D < 2.3, it is classified as Level I risk, corresponding to very mild crack development; When 2.3 ≤ TCI-D < 3.6, it is classified as Level II risk, corresponding to a small number of dispersed cracks; When 3.6 ≤ TCI-D < 4.9, it is classified as Level III risk, corresponding to significant crack expansion; When TCI-D≥4.9, it is classified as Level IV risk, corresponding to the degree of crack development.

9. A method for diagnosing the overall integrity of apparent cracks in tunnel lining based on digital images according to any one of claims 1-8, characterized in that, In step S01, the basic parameters include tunnel type, design dimensions, service life, construction technology, and geological conditions.

10. A digital image-based system for diagnosing the overall integrity of apparent cracks in tunnel lining, used to perform the diagnostic method according to any one of claims 1-9, characterized in that, include: The basic acquisition module is used to investigate and obtain the basic parameters of the target tunnel, acquire the original images of the tunnel lining appearance, and ensure that the overlap of adjacent images meets the requirements for panoramic stitching. The image preprocessing module is used to obtain a panoramic image of the tunnel lining by feature matching and stitching the original image of the tunnel lining acquired by the basic acquisition module, and to obtain a binary image of the lining cracks by semantic segmentation. The crack parameter extraction module, based on the concept of decomposable superposition of crack geometry, is used to extract crack skeleton and identify endpoints and bifurcation points. It decomposes complex bifurcation cracks into single crack units and quantitatively extracts the core parameters of crack length, width, angle, area, density, and spatial distribution characteristics. The core index calculation module calculates the tunnel crack index (TCI) and crack fractal dimension (D) respectively. The unified assessment and grading module objectively determines the boundary value between TCI and D based on hierarchical clustering, and constructs a unified TCI-D assessment model using partial least squares regression. Based on the TCI-D value, it is divided into risk levels of I-IV corresponding to different degrees of crack development. The model expression is Among them, TCI-D is the lining crack assessment index. and The TCI and D weights for lining cracks, For constant terms; The results application module compares and verifies the assessment results with current highway tunnel maintenance standards, and outputs corresponding maintenance and treatment recommendations based on risk classification.

Images