Tunnel lining crack parameter quantification and safety evaluation method based on digital image

Abstract

The invention discloses a tunnel lining crack parameter quantification and safety evaluation method based on a digital image, and provides a method for accurately calculating key quantification parameters such as crack length, maximum width, fractal dimension and type and the like based on a principal component analysis (PCA) method and a depth-first search (DFS) algorithm for a crack image detection result. Accurate quantification of crack characteristic parameters is realized; according to the method, a tunnel lining crack health evaluation index is established, the tunnel lining crack health evaluation index TCHI is established, the tunnel lining crack health evaluation index TCHI is established, the tunnel lining crack health evaluation index TCHI is comprehensive, objective and easy to calculate, and therefore systematic evaluation of the overall safety degree of a tunnel is achieved. The method is wide in application range and high in adaptability, and can provide technical support for realizing automation of tunnel disease recognition and detection by using a computer vision technology in the future.

Description

Technical Field

[0001] The present invention belongs to the technical field of tunnel disease detection, and particularly relates to a method for quantifying tunnel lining crack parameters and safety assessment based on digital images. Background Technique

[0002] With the rapid development of the Chinese economy, the urban space demand has expanded sharply and the ground space has become increasingly tense. Reasonably developing and utilizing underground space has become an important topic. As the core lifeline project of urban underground space development, tunnel engineering shows a rapid development trend under the background of accelerating urbanization. However, affected by the long-term coupling effect of the environment and load, typical diseases such as cracks, leakage, and voids behind the lining structure are likely to occur. Among them, lining cracks are the most common manifestation form, which will not only reduce the bearing capacity of the lining structure, but also become the seepage channel of groundwater, causing the corrosion of steel bars and leading to the occurrence of other diseases. Therefore, the detection, assessment, and maintenance of tunnel lining cracks are crucial for ensuring the safe operation of tunnels.

[0003] In recent years, with the rapid development of artificial intelligence methods, crack detection methods based on computer vision and deep learning have begun to be applied to tunnel disease detection. For example, the patent number is CN202411417796.0, and the patent name is a method and system for crack segmentation measurement of a subway shield tunnel. The crack skeleton is extracted from the crack area through a thinning algorithm, and based on the crack skeleton, the crack length and crack width are calculated according to the pixel connection method, providing a method for effectively obtaining and quantifying the detailed characteristics of cracks. However, how to combine the quantification results of these crack characteristics with the overall health status of the tunnel lining to evaluate the impact degree and urgency of cracks is an issue worthy of in-depth research and attention.

[0004] Therefore, a tunnel safety level assessment method that introduces the fractal dimension and comprehensively considers the crack length and maximum width is proposed. For example, the patent number is CN202311863072.4, and the patent name is a method for quantifying and statistically analyzing the apparent diseases of tunnel lining cracks and safety diagnosis. Although the disease level of cracks is determined according to the crack length, the maximum crack width, and the fractal dimension, the information on the crack orientation and type is ignored in the evaluation index. And the type of cracks (circumferential cracks, radial cracks, and inclined cracks) is directly related to the cause of cracks, the development trend, and their impact on structural safety.

[0005] In view of the above, the present application proposes a method for quantifying tunnel lining crack parameters and safety assessment based on digital images to make up for the problems in the prior art. Summary of the Invention

[0006] Aiming at the deficiencies of the above-mentioned existing technologies, the present invention provides a method for quantifying crack parameters and safety assessment of tunnel linings based on digital images. In this method, four key parameters, namely length, maximum width, fractal dimension, and crack type, are selected from the crack image detection and segmentation results as the quantification indexes of cracks. Using these parameter values, the grading levels and standards of crack health are established successively through methods such as hierarchical clustering and regression analysis, so as to realize the systematic assessment of the overall safety degree of the tunnel.

[0007] A method for quantifying crack parameters and safety assessment of tunnel linings based on digital images, characterized by comprising the following steps:

[0008] S1: Obtain digital images of the tunnel lining surface;

[0009] S2: Use the YOLO network model to identify and instance segment the digital images in S1, extract crack mask information, and generate binary images of cracks;

[0010] S3: Identify and extract four key crack quantification parameters, namely the length, maximum width, fractal dimension, and crack type, of the cracks from the binary images in S2;

[0011] S4: Introduce the method of hierarchical clustering, an unsupervised learning method, to automatically classify multiple crack data samples according to the crack quantification parameters in S3, and confirm that the grading levels of crack health are 4 clusters;

[0012] S5: Select a reference sample from the cluster without cracks among the multiple crack data samples in the 4 clusters in S4, and calculate the relative Euclidean distance between the reference sample and the other 3 cluster samples to evaluate the health of the cracks in the other 3 clusters;

[0013] S6: Establish a calculation model for the tunnel crack health diagnosis index (TCHI);

[0014] S7: Verify the accuracy and applicability of the calculation model in S6 through residual analysis and comparison of true values and predicted values;

[0015] S8: Expand the assessment of single cracks to the entire tunnel section and establish the tunnel crack health level standard.

[0016] Further, the step S3 includes the following steps:

[0017] S3-1: Adopt a strategy combining the median axis transformation method (MAT) and the discrete skeleton evolution pruning algorithm (DSE) to accurately obtain the crack skeleton;

[0018] S3-2: Adopt a cross-point removal strategy to remove the cross-points in the complex structure of the crack skeleton and simplify it into linear segments;

[0019] S3-3: Use the principal component analysis method (PCA) to determine the propagation direction of the crack to accurately measure its width;

[0020] S3-4: Use the depth-first search algorithm (DFS) to accurately measure the length of the crack;

[0021] S3-5: Use the rotation angle (r c ) as a quantization index, and determine the classification of the crack by analyzing the angular difference between the crack propagation direction and the tunnel axis. The cracks are divided into three crack types: circumferential cracks, longitudinal cracks, and inclined cracks;

[0022] S3-6: Use the box dimension to quantify the density and distribution of the tunnel lining cracks, that is, the fractal dimension of the cracks. Calculate the box dimension of the crack image according to the following formula:

[0023]

[0024] In the formula, F d is the box dimension; N(s) represents the number of boxes covering the crack part in the image when the box size is s, and s is the side length of the square box.

[0025] Furthermore, the step S4 includes:

[0026] Collect digital images of multiple cracks and generate binary images through the method of step S2 to form a crack data sample space. For each binary image of each crack sample in the crack data sample space, obtain four key crack quantization parameters of the crack of each sample, namely the maximum width, length, fractal dimension, and type, and use a sample without cracks as a reference sample;

[0027] Introduce the hierarchical clustering method to automatically classify the remaining crack data samples according to the four key crack quantization parameters, and reveal the internal structure and classification rules of the data;

[0028] Through the dendrogram analysis of the crack data, since the inter-class distance increases significantly when the number of clusters reaches four main clusters, and adding more clusters will not significantly improve the intra-class homogeneity or the inter-class discrimination, 4 clusters are selected as the grading levels of the crack health grade.

[0029] Further, step S5 specifically includes: among the 4 clusters, assuming the cluster where the reference sample is located as Class 2, calculating the relative Euclidean distance between the reference sample and the samples in the other 3 clusters assumed to be Class 1, Class 3, and Class 4. The calculation results include the maximum value, minimum value, average value, and standard deviation of the four key crack quantification parameters between the samples in Class 1, Class 3, and Class 4 and the reference sample. According to the average value difference of the relative Euclidean distance between the samples in Class 1, Class 3, and Class 4 and the reference sample, they are classified from large to small as: Grade IV, which means the crack severity has reached the limit and immediate countermeasures need to be taken; Grade III, indicating that the crack disease is relatively serious and has potential threats and should be paid attention to; Grade II, the crack disease degree is general and may cause safety problems in the future; Grade I, that is, Class 2 where the reference sample is located, the crack degree is slight or there is no crack and it can be temporarily not processed; thus, the health degrees of the cracks corresponding to the 4 clusters are determined respectively.

[0030] Further, comprehensively considering the influence forms of the maximum width w m and the length l, crack type K, fractal dimension D on the crack health degree, the calculation model formula of the tunnel crack health diagnosis index (TCHI) is proposed as follows:

[0031] TCHI = k1×(w m ×l) k2 ×e k3×D ×(1 + k4×K)

[0032] In the formula, w m is the maximum crack width, l is the crack length, D is the fractal dimension of the crack, K is the crack type. When the crack is a circumferential crack, K takes 0; when it is an inclined crack, K takes 0.5; when it is a longitudinal crack, K takes 1. k1, k2, k3, and k4 are the coefficients of each quantification parameter.

[0033] Further, step S7 specifically includes: first, assigning values to the TCHI result parameter in the calculation model formula according to the relative Euclidean distance as the true value denoted as TCHI tv , that is, the average value of the relative Euclidean distance between the samples of other grades and the reference sample, to ensure the objectivity of the calculation model;

[0034] Then, regarding the respective sample parameters in the four clusters as the independent variable space and TCHI tv as the dependent variable space, the coefficient values k1 = 1.64, k2 = 0.1, k3 = 0.5, and k4 = 0.1 are obtained by fitting through the nonlinear least squares method. Finally, the following formula can be used to calculate the predicted value denoted as TCHI pv :

[0035] TCHI = 1.64(w m × l) 0.1 × e 0.5D × (1 + 0.1K)

[0036] Finally, draw the true value TCHI of each sample tv and the predicted value TCHI pv between the control relationship diagram, by observing the control relationship diagram, the true value TCHI of each sample can be obtained tv and the predicted value TCHI pv The difference between them is small, indicating that the calculation model has high precision;

[0037] Through residual analysis, it is found that the average value of the residuals between the predicted value and the true value is close to 0. From the residual distribution diagram, it can be seen that the residuals are distributed within a horizontal band, proving that the calculation model has high applicability and accurate prediction ability.

[0038] Furthermore, the step S8 specifically includes:

[0039] Conduct a grade assessment on each crack in the target tunnel section respectively. According to its width, length, type and fractal dimension and other parameters, use the calculation model established in the S6, and use the coefficients of each quantization parameter of k1, k2, k3, k4 in the S7 to calculate the health index of each crack in the tunnel section;

[0040] Statistically analyze the predicted values TCHI of all cracks in the target tunnel section pv and take the maximum value as the overall health index of the tunnel section. According to the interval where the predicted value TCHI pv is located, find the maximum TCHI pv The corresponding safety level is the overall health level of the cracks in the target tunnel section.

[0041] The beneficial effects of the above technical solutions are as follows:

[0042] The present invention proposes a method for accurately calculating key quantization parameters such as crack length, maximum width, fractal dimension and type based on principal component analysis and depth-first search for the crack image detection results, realizing the accurate quantization of crack characteristic parameters;

[0043] Using these parameter values, the classification grades and standards of crack health are established in turn through methods such as hierarchical clustering and regression analysis, and a comprehensive, objective and easy-to-calculate tunnel-lining crack health index (TCHI) is established, so as to realize the systematic evaluation of the overall safety level of the tunnel;

[0044] This method has a wide range of applications and strong adaptability, providing technical support for the future complete automation of tunnel disease identification and detection using computer vision technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic diagram of the overall method flow of the present invention;

[0046] Figure 2 It is a schematic diagram of the process when the present invention works;

[0047] Figure 3 It is a schematic diagram of obtaining the crack skeleton in the present invention;

[0048] Figure 4 It is a schematic diagram of obtaining the crack width in the present invention

[0049] Figure 5 It is a schematic diagram of the process of obtaining the crack length in the present invention;

[0050] Figure 6 It is a schematic diagram of determining the classification type of cracks in the present invention;

[0051] Figure 7 It is a tree diagram of crack data in the present invention;

[0052] Figure 8 It is a relationship diagram of the true value and predicted value of TCHI in the present invention;

[0053] Figure 9 It is a schematic diagram of the TCHI residual in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] The following further illustrates the technical solution of the present invention in conjunction with the attached Figures 1 to 9 figures shown and embodiments. The following specific examples are helpful for those skilled in the art to further understand the present invention, but do not limit the present invention in any form. Additionally, it should be noted that for ease of description, only parts related to the present invention are shown in the drawings rather than all of them.

[0055] Example 1: Obtain digital images of the tunnel lining surface;

[0056] In this example, the method for obtaining digital images of the tunnel lining surface is an unmanned aerial vehicle image acquisition system.

[0057] Example 2: On the basis of Example 1, use the improved YOLO tunnel lining crack segmentation model to identify and instance-segment these images to extract crack mask information, and finally generate a binary image of the cracks.

[0058] In this example, the YOLO network model adopts the lightweight YOLOv8s-seg network structure.

[0059] Example 3. Based on Example 2, four key crack quantification parameters, namely the length, maximum width, fractal dimension, and crack type of the crack, are identified and extracted.

[0060] The steps for identifying and extracting the four key crack quantification parameters, namely the length, maximum width, fractal dimension, and crack type of the crack, in this example are as follows:

[0061] By using the combined Median Axis Transform (MAT), calculate the midpoint of each point in the image to the nearest shape boundary, reconstruct the central path of the crack, and measure the crack tip branches and locate the local boundary points. And the Discrete Skeleton Evolution Pruning Algorithm (DSE), first construct a distance transform matrix for each point on the skeleton, and then merge the largest inscribed circles through the rasterization algorithm to reconstruct the original shape of the crack. Accurately obtain the crack skeleton, as Figure 3 shown;

[0062] Adopt the proposed intersection removal strategy. By identifying nodes with a degree greater than 2, merge adjacent nodes, and fit a circle at each intersection point through the midpoint circle algorithm to ensure covering the entire area, so as to remove the intersections in the complex structure of the crack skeleton and simplify it into linear segments;

[0063] Determine the extension direction of the crack to accurately measure its width, using the Principal Component Analysis method (PCA). First, apply a 5×5 pixel kernel in the local area near the crack, collect the coordinates of the pixel points in this area, and calculate the covariance matrix to capture the variation relationship between the data. The eigenvector corresponding to the largest eigenvalue represents the main variation direction in the data, that is, the main extension direction of the crack. Generate a vector orthogonal to the main direction at each crack skeleton point as the reference line for width measurement. Use the Canny edge detector to determine the intersection points of the orthogonal vector and the crack boundary, and calculate the distance between the two intersection points through the Euclidean distance formula, which is the crack width at this point, as Figure 4 shown;

[0064] Use the Depth-First Search algorithm (DFS) to accurately measure the crack length. Adopt a specific 3×3 convolution kernel technology to analyze the connection relationship between each pixel and its surrounding neighboring points and automatically identify the endpoints. Apply the 3×3 convolution kernel to the crack skeleton image, and the endpoint pixels obtain unique weighted values in the convolution process, so as to be accurately identified. After determining the endpoints, the DFS algorithm starts from any endpoint and gradually traverses all connected paths, preferentially selecting the branches that can extend the longest path. The measurement of the crack length is based on the total number of pixel points on the longest path determined by DFS, as Figure 5 shown.

[0065] Adopt the rotation angle r cAs a quantitative index, the classification of cracks is determined by analyzing the angular difference between the crack extension direction and the tunnel axis. The cracks are divided into circumferential cracks, longitudinal cracks, and inclined cracks, as Figure 6 shown;

[0066] The box dimension is used to quantify the density and distribution of tunnel lining cracks. First, the segmented crack image is converted into a binary image, where the crack area is marked as 1 and the non-crack area is marked as 0. Then, the entire image is covered by a series of square boxes with side lengths that are powers of 2, and the number of boxes covering the cracks is calculated. By analyzing the logarithmic relationship between the box size s and the number of boxes N(s) covering the cracks, and using the least squares method to fit a straight line, the slope of the resulting straight line is the required box dimension F d , which is used to quantify the fractal characteristics of the cracks. The box dimension of the crack image is calculated according to the following formula:

[0067]

[0068] In the formula, F d is the box dimension; N(s) represents the number of boxes covering the crack part of the image when the box size is s; s is the side length of the square box.

[0069] Example 4, based on Example 3, the method of hierarchical clustering, an unsupervised learning method, is introduced to automatically classify the crack data samples, and 4 clusters are selected as the classification levels of the crack health grades.

[0070] The specific steps to determine the classification levels of the crack health grades in this example are as follows:

[0071] Establish a data sample space based on crack quantification parameters, which contains 47 unlabeled crack data samples, and calculate four quantification parameters of the cracks, namely the maximum width, length, fractal dimension, and type, according to their binary images. Among them, sample No. 41 is set as the reference sample, representing the absence of cracks.

[0072] Introduce the method of hierarchical clustering to automatically classify the 47 unlabeled crack data samples;

[0073] Through the dendrogram analysis of the crack data, as Figure 7 shown, it is observed that when the number of clusters reaches four main clusters, the inter-class distance increases significantly, indicating that adding more clusters will not significantly improve the intra-class homogeneity or the inter-class discrimination. Select 4 clusters as the classification levels of the crack health grades.

[0074] Example 5, based on Example 4, calculate the relative Euclidean distance between the reference sample and other samples of each grade to determine the health degree of the cracks.

[0075] The specific steps for determining the health degree of cracks in this embodiment are as follows:

[0076] Calculate the relative Euclidean distance between the reference sample (Sample No. 41, belonging to Class 2) and samples of other grades. Specifically, first form a standard sample matrix with the four key crack quantification parameters of the reference sample, then form a sample matrix for each sample of other grade samples with their four key crack quantification parameters, and finally calculate the relative Euclidean distance between the standard sample matrix and the sample matrices of other grade samples.

[0077] Precisely define the health grade of the tunnel lining cracks based on the calculation results. The calculation results are shown in Table 1, including the maximum value, minimum value, average value, and standard deviation from the Class 2 reference sample. Class 4 is rated as Grade IV because of the largest average value, indicating that the severity of the cracks has reached the limit and immediate countermeasures need to be taken. Class 3 is Grade III, indicating that the crack disease is relatively serious and has potential threats and should be paid attention to; Class 1 is Grade II, the degree of the crack disease is general and may cause safety problems in the future; while Class 2 is Grade I, the crack degree is slight or there are no cracks and can be left untreated for the time being.

[0078] Table 1

[0079]

[0080] Example 6, based on Example 5, establish a calculation model for the Tunnel Crack Health Index (TCHI).

[0081] Since the two parameters of the maximum width w m and the length l have a non-linear and equally important influence form on the health degree of the cracks; the influence of the crack type K is based on the correction method for judging the safety grade of tunnel cracks. When the crack safety grade is divided into four grades, the score also needs to be adjusted accordingly according to the harm degree of the crack type; and then based on the correlation study between the influence form of the fractal dimension D and the health degree of the cracks, the formula for the calculation model of the Tunnel Crack Health Index (TCHI) is finally designed as:

[0082] TCHI = k1 × (w m × l) k2 × e k3×D × (1 + k4 × K)

[0083] where w m is the maximum width of the crack, l is the length of the crack, D is the fractal dimension of the crack, K is the type of the crack, taking 0 for circumferential cracks, 0.5 for inclined cracks, and 1 for longitudinal cracks. k1, k2, k3, and k4 are the coefficients of each quantification parameter.

[0084] Example 7. Based on Example 6, the accuracy and applicability of the model are verified through residual analysis and comparison between the true values and the predicted values.

[0085] The specific steps for verifying the accuracy and applicability of the model in this example are as follows:

[0086] Considering that the known model is non - linear, this patent estimates the parameters in the model based on the non - linear least - squares method. The established sample space is equivalent to the independent variable space, and the tunnel crack health diagnosis index (TCHI) is equivalent to the dependent variable space. The data in the dependent variable space is called the true value. The specific method is as follows: First, use the average value of the relative Euclidean distance between samples of other grades and the reference sample to assign a value to the TCHI result parameter in the calculation model formula as the true value, denoted as TCHI tv Regarded as the dependent variable space, that is, the assumed result; then regard the sample parameters in the four clusters as the independent variable space, that is, the operation parameters w m , l, D, K in the calculation model formula. Through non - linear least - squares fitting, the corresponding coefficient values for each parameter are obtained as k1 = 1.64, k2 = 0.1, k3 = 0.5, k4 = 0.1. Finally, the predicted value can be calculated through the following formula, denoted as TCHI pv ;

[0087] TCHI = 1.64(w m ×l) 0.1 ×e 0.5D ×(1 + 0.1K)

[0088] Finally, draw a comparison diagram between the true value TCHI tv of each sample and the predicted value TCHI pv . By observing the comparison diagram, it can be seen that the difference between the true value TCHI tv of each sample and the predicted value TCHI pv is small, indicating that the calculation model has high precision;

[0089] Through residual analysis, as Figure 9 shown, we find that the average residual is close to 0, and the residuals are distributed within a horizontal band, proving the applicability and predictive ability of the established regression model.

[0090] Example 8. Based on Example 7, the evaluation of a single crack is extended to the entire tunnel section to establish a tunnel crack health grade standard.

[0091] The specific steps for establishing the tunnel crack health grade standard in this example are as follows:

[0092] Perform a grade assessment on each crack in the target tunnel section respectively, and calculate the health index for each crack in the tunnel section using the calculation model established above according to parameters such as its width, length, type, and fractal dimension;

[0093] Statistically analyze the predicted values TCHI of all cracks in the tunnel section, and take its maximum value as the overall health index of the tunnel section. Take the maximum predicted value TCHI pv And use the corresponding safety level corresponding to it as the overall health level of the tunnel section. The specific tunnel crack health level standard is shown in Table 2: pv

[0094] Table 2

[0095]

[0096] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Those of ordinary skill in the art should understand that it is not necessary and impossible to enumerate all implementation manners here. It should be pointed out that without departing from the concept of the present invention, several deformations or modifications can be made to the device. These all belong to the protection scope of the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the claims of the present invention.​

Claims

1. A method for quantifying tunnel lining crack parameters and safety assessment based on digital images, characterized in that, It includes the following steps: S1: Obtain the digital image of the tunnel lining surface; S2: Use the YOLO network model to identify and instance-segment the digital image in S1, extract the crack mask information, and generate the binary image of the crack; S3: Identify and extract four key crack quantification parameters of the crack, namely the length, maximum width, fractal dimension, and crack type, from the binary image in S2; S4: Introduce the method of hierarchical clustering, an unsupervised learning method, to automatically classify multiple crack data samples according to the crack quantification parameters in S3, and confirm that the number of classification levels of the crack health level is 4 clusters; S5: Select a reference sample from the cluster without cracks among the multiple crack data samples in the 4 clusters in S4, and calculate the relative Euclidean distance between the reference sample and the other 3 cluster samples to evaluate the health of the cracks in the other 3 clusters; S6: Establish a calculation model for the Tunnel Crack Health Index (TCHI); S7: Verify the accuracy and applicability of the calculation model in S6 through residual analysis and comparison of the true value and the predicted value; S8: Expand the evaluation of a single crack to the entire tunnel section and establish the tunnel crack health level standard.

2. The method for quantifying crack parameters and safety assessment of tunnel lining based on digital images according to claim 1, wherein The step S3 includes the following steps: S3-1: Adopt a strategy combining the Median Axis Transform (MAT) and the Discrete Skeleton Evolution Pruning Algorithm (DSE) to accurately obtain the crack skeleton; S3-2: Adopt a cross-point removal strategy to remove the cross-points in the complex structure of the crack skeleton and simplify it into a linear segment; S3-3: Adopt the Principal Component Analysis method (PCA) to determine the extension direction of the crack to accurately measure its width; S3-4: Adopt the Depth-First Search algorithm (DFS) to accurately measure the crack length; S3-5: Use the rotation angle (r c ) as a quantification index, and determine the classification of cracks by analyzing the angular difference between the crack extension direction and the tunnel axis. The cracks are divided into three types: circumferential cracks, longitudinal cracks, and inclined cracks; S3-6: Use the box dimension to quantify the density and distribution of the tunnel lining cracks, that is, the fractal dimension of the crack, and calculate the box dimension of the crack image according to the following formula: Where F d is the box dimension; N(s) represents the number of boxes covering the crack part in the image when the box size is s, and s is the side length of the square box.

3. The method for quantifying crack parameters and safety assessment of tunnel lining based on digital images according to claim 2, wherein, The step S4 includes: Collect the digital images of multiple cracks and generate binary images through the method in step S2 to form a crack data sample space. For each binary image of the crack samples in the crack data sample space, obtain the four key crack quantification parameters of the maximum width, length, fractal dimension, and type of the crack for each sample according to the methods in steps S3-1 to S3-6, and use one sample without cracks as the reference sample; Introduce the hierarchical clustering method to automatically classify the remaining crack data samples according to the four key crack quantification parameters, and reveal the internal structure and classification rules of the data; Through the dendrogram analysis of the crack data, when the number of clusters reaches four main clusters, the inter-class distance increases significantly, and adding more clusters will not significantly improve the intra-class homogeneity or improve the inter-class discrimination. Therefore, 4 clusters are selected as the number of classification levels of the crack health level.

4. The method for quantifying tunnel lining crack parameters and safety assessment based on digital images according to claim 3, wherein The specific steps of S5 include: among the four clusters, assuming the cluster where the reference sample is located as Class 2, calculating the relative Euclidean distance between the reference sample and the samples in the other three clusters assumed to be Class 1, Class 3, and Class 4. The calculation results include the maximum value, minimum value, average value, and standard deviation of the four key crack quantification parameters between the samples in Class 1, Class 3, and Class 4 and the reference sample. According to the average value difference of the relative Euclidean distance between the samples in Class 1, Class 3, and Class 4 and the reference sample, they are divided from large to small into: Grade Ⅳ, which means that the crack severity has reached the limit and immediate countermeasures need to be taken; Grade Ⅲ, indicating that the crack disease is relatively serious and has potential threats, and attention should be paid; Grade Ⅱ, the crack disease level is general and may cause safety problems in the future; Grade Ⅰ, that is, Class 2 where the reference sample is located, the crack level is slight or there is no crack, and it can be temporarily not processed; thus, the health levels of the cracks corresponding to the four grades of the four clusters are determined.

5. The method for quantifying crack parameters and safety assessment of tunnel lining based on digital images according to claim 4, wherein Comprehensive maximum width w m Regarding the influence forms of length l, crack type K, fractal dimension D and crack health degree, the calculation model formula of the tunnel crack health diagnosis index (TCHI) is proposed as follows: where w m is the maximum crack width, l is the crack length, D is the fractal dimension of the crack, K is the crack type. When the crack is a circumferential crack, K takes 0; when it is an inclined crack, K takes 0.5; when it is a longitudinal crack, K takes 1. k1, k2, k3, and k4 are the coefficients of each quantization parameter.

6. The method for quantifying crack parameters and safety assessment of tunnel lining based on digital images according to claim 5, wherein The specific steps of step S7 include first assigning the TCHI result parameter in the calculation model formula according to the relative Euclidean distance as the true value, denoted as TCHI tv , that is, the average value of the relative Euclidean distances between samples of other levels and the reference sample, to ensure the objectivity of the calculation model; Then, consider the sample parameters in each of the four clusters as the independent variable space, and consider TCHI tv as the dependent variable space. By fitting using the nonlinear least squares method, the coefficient values are obtained as k1 = 1.64, k2 = 0.1, k3 = 0.5, and k4 = 0.

1. Finally, the predicted value, denoted as TCHI, can be calculated using the following formula pv : TCHI = 1.64 (w m × l) 0.1 × e 0.5D × (1 + 0.1K) Finally, draw the true value TCHI of each sample tv and the predicted value TCHI pv to obtain a comparison relationship diagram between them. By observing the comparison relationship diagram, the true value TCHI of each sample tv and the predicted value TCHI pv have a small difference, indicating that the calculation model has high precision; Through residual analysis, it is found that the average value of the residuals between the predicted values and the true values is close to 0. From the residual distribution diagram, it can be seen that the residuals are distributed within a horizontal band, proving that the calculation model has high applicability and accurate prediction ability.

7. The method for quantifying tunnel lining crack parameters and safety assessment based on digital images according to claim 6, characterized in that, The specific steps of S8 include: Conducting grade evaluation on each crack in the target tunnel section respectively, using the calculation model established in S6 according to parameters such as its width, length, type, and fractal dimension, and calculating the health index of each crack in the tunnel section by using the coefficients of each quantification parameter k1, k2, k3, k4 in S7. Statistically analyze the predicted values TCHI of all cracks within the target tunnel section pv and take the maximum value among them as the overall health index of this tunnel section. Based on the maximum TCHI pv corresponding safety level as the overall health level of the cracks within the target tunnel section.

Images