Drilling optical image crack intelligent quantification method for diaphragm wall quality evaluation

Abstract

This invention discloses an intelligent quantification method for borehole optical images for cutoff wall quality assessment, belonging to the field of cutoff wall quality technology. It aims to solve the problems of difficulty in identifying micro-cracks, strong subjectivity in quantitative judgment of geometric features, and low detection efficiency in manual inspection. The method includes the following steps: acquiring high-resolution images of the borehole wall; eliminating specular reflection through adaptive homomorphic filtering to generate an illumination-invariant image; inputting a U-Net++ network (crack segmentation model) with a fused directional gradient histogram enhancement module to identify crack pixels with a width <0.1mm; calculating the length, bifurcation angle, and dip through crack topological connectivity analysis; predicting potential penetration paths using a soil stress field model; calculating an index characterizing the integrity of the cutoff wall based on the segmentation results, geometric features, and potential paths; and generating a crack risk assessment report based on this index.

Description

Technical Field

[0001] This invention relates to the field of seepage barrier wall quality inspection technology, specifically a method for intelligent quantification of cracks in borehole optical images for seepage barrier wall quality assessment. Background Technology

[0002] With the accelerating pace of urbanization and the gradual expansion of urban infrastructure, cutoff walls are increasingly widely used in hydraulic engineering, environmental protection, and civil engineering. Especially in river embankment projects, cutoff walls serve as a core barrier to prevent flooding, control piping, and ensure the structural safety of the embankment. Their construction quality and long-term performance directly affect the safety of people's lives and property within the embankment protection zone and the reliability of the flood control system. Cutoff walls are primarily used to prevent the seepage of groundwater, sewage, or other liquids, ensuring the structural and environmental safety of the project. Their construction quality directly impacts the service life and safety of the entire project.

[0003] In recent years, with the development of non-destructive testing technologies such as endoscopy, inspecting cutoff walls using borehole optical images has become a common quality assessment method. This inspection method is particularly important for dike projects. Dike cutoff walls are typically buried at great depths, extend over long distances, and operate in complex environments (enduring long-term hydraulic seepage pressure, soil stress changes, etc.). Defects such as cracks within the wall structure are key factors affecting its seepage prevention effect and structural integrity. However, current methods for detecting borehole cracks in cutoff walls primarily rely on manual interpretation and analysis of borehole images. Operators need to observe each frame of the acquired high-definition images, identify potential crack information, assess the geometric characteristics of the cracks (such as crack length, bifurcation angle, and dip), and determine whether there is a potential risk of penetration.

[0004] In the practice of quality assessment of dikes and seepage barriers, this inspection method, which relies on manual experience, faces more severe challenges, mainly in the following aspects:

[0005] The accuracy of identifying micro-cracks is low: Since the cracks in the seepage barrier are often extremely fine, especially micro-cracks with a width of less than 0.1 mm, traditional manual observation is difficult to identify accurately. They are easily overlooked or misjudged due to the limited ability of the human eye to distinguish them, thus affecting the overall assessment results.

[0006] The judgment of crack geometry and potential penetration path is subjective: manual interpretation is not only limited by the experience level of the inspectors, but also has a strong subjectivity in the quantitative statistics of geometric features such as crack length, bifurcation angle and dip when facing complex crack networks, making it difficult to meet the requirements of standardization and objective quantification.

[0007] Low detection efficiency: Because it is necessary to rely on manual frame-by-frame image analysis, processing a large number of borehole wall images is labor-intensive and time-consuming, and is easily affected by fatigue and attention, resulting in low detection efficiency and unstable evaluation results.

[0008] The above background information is provided only to aid in understanding the concept and technical solution of this invention. It does not necessarily belong to the prior art of this patent application. In the absence of clear evidence that the above information was disclosed on the filing date of this patent application, the above background information should not be used to evaluate the novelty and inventiveness of this application. Summary of the Invention

[0009] This application provides an intelligent quantification method for cracks in borehole optical images for the quality assessment of cutoff walls, which solves the problems of difficulty in identifying microcracks, strong subjectivity in quantitative judgment of geometric features, and low detection efficiency in manual inspection.

[0010] To achieve the above objectives, the embodiments of this application disclose the following technical solutions:

[0011] A method for intelligent quantification of cracks in borehole optical images for quality assessment of cut-off walls includes the following steps:

[0012] Acquire high-resolution images of the borehole wall captured by an endoscopic camera inserted into the borehole of the impermeable wall;

[0013] Adaptive homomorphic filtering is applied to high-resolution images of borehole walls to eliminate specular reflection interference and generate illumination-invariant images.

[0014] An illumination-invariant image is input into a preset crack segmentation model to generate crack segmentation results. The crack segmentation model is a U-Net++ network with a fused directional gradient histogram enhancement module, used to identify crack pixels, including cracks with a width of less than 0.1 mm.

[0015] The crack segmentation results are analyzed for crack topological connectivity to calculate the length, bifurcation angle and dip of the identified cracks.

[0016] Based on the crack length, bifurcation angle and inclination data, combined with the pre-set soil stress field model, the potential penetration path of the crack in the seepage barrier wall is predicted.

[0017] Based on the crack segmentation results, crack length, bifurcation angle, inclination, and predicted potential penetration paths, calculate and output the integrity index characterizing the integrity of the cutoff wall.

[0018] A risk assessment report on cracks in the anti-seepage wall is generated based on the integrity index.

[0019] In this embodiment, through overall image acquisition, preprocessing, segmentation, topology analysis, and risk assessment, the crack information of the seepage barrier wall in the high-definition image of the borehole wall is automatically extracted and quantified, achieving an objective assessment of the structural state of the seepage barrier wall step by step. The high-definition image acquired by the endoscopic camera is processed by adaptive homomorphic filtering to generate an illumination-invariant image, ensuring that the details of the borehole wall structure are realistically presented and effectively eliminating interference caused by uneven illumination and specular reflection, thus guaranteeing that the acquired image only reflects the physical and structural characteristics of the borehole wall itself. After the illumination-invariant image enters the fusion directional gradient histogram enhancement module and the U-Net++ network, crack pixels are automatically identified, enabling extremely high precision in the extraction of minute cracks. This allows for accurate capture of cracks with extremely small widths during quantitative segmentation, thereby achieving crack geometric feature calculation based on quantitative data. By segmenting the crack skeleton in the image, the length, bifurcation angle, and dip of the crack are obtained, providing a reliable basis for subsequent quantitative derivation of the crack propagation process using a physical model. This method, employing a pre-defined soil stress field model to predict potential crack penetration paths, combines physical and mechanical parameters with crack geometry information to a certain extent, enabling a physical interpretation and quantitative estimation of crack propagation trends and penetration risks. The integrity index, derived by weighted fusion of crack density, geometric risk, and propagation risk, objectively reflects the overall structural condition and engineering risk level of the cutoff wall, thus achieving a comprehensive assessment of the cutoff wall's construction quality and structural safety. The entire technical solution, while ensuring scientific, standardized, and objective data processing and quantitative analysis, avoids errors and inconsistencies caused by subjective judgment in manual inspection, effectively solving technical problems such as the difficulty in manually detecting micro-cracks, the strong subjectivity of quantitative judgment of geometric features, and low inspection efficiency.

[0020] In some possible implementations, the step of performing adaptive homomorphic filtering on the high-resolution image of the borehole wall to eliminate specular reflection interference and generate an illumination-invariant image includes:

[0021] The high-resolution image of the borehole wall is converted into a logarithmic domain image, and the specific formula for generating the logarithmic domain image is as follows:

[0022]

[0023] In the formula, This indicates the original high-resolution image of the borehole wall in coordinates. Pixel value at; A positive constant set to prevent the logarithmic function input from being zero; Represents the transformed logarithmic field image;

[0024] The frequency domain image is generated by performing a Fast Fourier Transform on the logarithmic domain image, using the following formula:

[0025]

[0026] In the formula, and These are the width and height of the image, respectively; and These are the horizontal and vertical frequency coordinates of the frequency domain image, respectively; The imaginary unit; This represents the obtained frequency domain image; Represents the natural exponential function;

[0027] An adaptive Gaussian high-pass filter is applied to the frequency domain image to suppress specular reflection components, generating a filtered frequency domain image. The specific formula is as follows:

[0028]

[0029] In the formula, Represented in frequency coordinates The filter response at point is given by the following formula:

[0030]

[0031] in, and These represent the center coordinates of the dominant specular reflection frequencies identified from the frequency distribution characteristics; The cutoff characteristics of the filter are controlled by the standard deviation parameter adaptively calculated based on the specular reflection component.

[0032] Perform an inverse fast Fourier transform on the filtered frequency domain image to generate the inverse transform image. The specific formula is as follows: ;

[0033] The inverse transform image is converted to the linear domain to generate an illumination-invariant image, as shown in the following formula: The above technical solution utilizes logarithmic domain transformation and Fourier transform combined with an adaptive Gaussian high-pass filter to process the image, directly suppressing and eliminating components related to illumination variations and specular reflections in the high-definition borehole wall image from the frequency domain. After mathematical mapping, interference caused by uneven brightness in the original image is balanced. After obtaining the image frequency information using Fourier transform, specific frequency components are suppressed by designing filters, achieving separation of illumination and specular reflection components. After reconstructing the image spatial information using inverse Fourier transform, it is restored to the linear domain through exponential function mapping, ultimately obtaining an illumination-invariant image reflecting the actual borehole wall structural characteristics. In this way, factors related to ambient lighting and reflection interference can be seamlessly removed from the image, ensuring that the image data input to subsequent processing modules remains highly stable and reliable. This image preprocessing result provides a solid foundation for subsequent automatic segmentation and crack identification, ensuring that the detailed information obtained during the detection process truly reflects the structural characteristics of the seepage barrier wall.

[0034] In some possible implementations, the step of inputting an illumination-invariant image into a preset crack segmentation model to generate crack segmentation results includes:

[0035] The encoder part of the U-Net network is used to extract multi-scale feature maps of the illumination-invariant image to generate feature maps;

[0036] The feature map is processed by the Histogram of Oriented Gradients (HGP) enhancement module to enhance crack edge information and generate an enhanced feature map.

[0037] The decoder part of the U-Net network is used to upsample and skip connections the enhanced feature map to generate an initial segmentation map;

[0038] Apply the Sigmoid activation function to the initial segmentation map to generate a probability map;

[0039] Binarization of the probability map based on a preset threshold generates crack segmentation results. Using the above technical solution, illumination-invariant images are directly fed into a deep neural network. Multi-scale feature extraction and histogram of oriented gradients (HOR) techniques are used to calculate and enhance the gradient information of crack contours in the image, quantifying crack edge features mathematically. This allows the deep network to accurately identify pixel-level boundaries of minute cracks in the image during automatic segmentation. The real structural information of the borehole wall contained in the illumination-invariant image is extracted and upsampled back through the segmentation network, ensuring that the true morphology and coherent structure of the cracks are reflected in the segmentation output. The HOR technique, combined with a cosine operator, quantitatively analyzes the edge direction, ensuring the stability and accuracy of crack edge information after mathematical processing. This guarantees that the crack segmentation results output by the deep network objectively reflect the actual situation of the seepage barrier wall cracks. In this way, the image segmentation process can rely entirely on automated data processing, achieving high-precision capture and consistent quantification of crack edges, further providing scientific evidence to support subsequent calculations of crack geometric features and engineering quality assessment.

[0040] In some possible implementations, the steps of performing crack topological connectivity analysis on the crack segmentation results and calculating the length, bifurcation angle, and dip of the identified cracks include:

[0041] Morphological skeletonization processing is performed on the crack segmentation results to generate crack skeleton images;

[0042] Based on the crack skeleton image, crack branch points and endpoints are identified, and topological node data is generated.

[0043] The Euclidean distance between each crack pixel in the crack skeleton image is calculated to obtain the crack length, generating crack length data. The specific formula for calculating the crack length is as follows:

[0044] In the formula, Represents the first in the skeleton image The Euclidean distance between adjacent pixels; Indicates the total length of the identified cracks;

[0045] Analyze the vector angle between adjacent branches in the topology node data, calculate the bifurcation angle, and generate bifurcation angle data; the specific formula for calculating the bifurcation angle is as follows:

[0046]

[0047] In the formula, and The direction vectors of the two adjacent branches at the bifurcation point; " denotes the vector dot product; and Representing vectors respectively and The modulus length; To calculate the bifurcation angle

[0048] Based on the pixel gradient direction of the crack skeleton image, the crack inclination is calculated to generate inclination data; the specific formula for calculating the crack inclination is as follows:

[0049]

[0050] In the formula, This represents the set of all pixels in the crack skeleton image; For at pixel Gradient magnitude at point, formula ,in It is an illumination-invariant image generated through homomorphic filtering; For pixels The gradient direction at that location, i.e., the direction of local change in that pixel; This represents the overall crack tendency or average angle obtained from the weighted average gradient direction of all skeleton pixels.

[0051] By integrating crack length, bifurcation angle, and dip data, the crack length, bifurcation angle, and dip are output. Using the aforementioned technical solution, morphological skeletonization simplifies the crack region into a single-pixel-width connected structure, fully preserving the overall crack appearance and clearly presenting it spatially. Simultaneously, topological connectivity analysis detects and processes nodes and branches within the skeleton, enabling the quantitative calculation of continuous geometric parameters. Stable and clear skeleton information is extracted from the image data, and based on this skeleton information, a rigorous quantitative relationship is established for the geometric features of the seepage barrier cracks. This achieves standardized and efficient expression of the crack structure, thus providing a scientific data basis for subsequent engineering inspections.

[0052] In some possible implementations, the steps of predicting the potential penetration path of cracks in the cutoff wall, based on crack length, bifurcation angle, and inclination data, combined with a pre-defined soil stress field model, include:

[0053] A three-dimensional stress field mesh model is constructed based on the soil physical parameters of the seepage barrier wall to generate an initial stress field model;

[0054] The crack length, bifurcation angle, and dip direction data are mapped to the corresponding positions in the initial stress field model to generate a stress field model containing crack parameters.

[0055] Based on the elastoplastic constitutive equation, the crack propagation behavior of a stress field model containing crack parameters under gravity load and hydraulic seepage pressure is simulated, and crack propagation simulation data is generated.

[0056] The continuous propagation trajectory along the direction of maximum principal stress is extracted from the crack propagation simulation data to generate potential penetration paths. Using the above technical solution, a three-dimensional stress field mesh model is constructed to numerically map the physical stress state inside the cutoff wall with the crack geometry data. Crack length, bifurcation angle, and dip direction data are directly quantified and embedded into the mesh model, enabling the elastoplastic constitutive equations to accurately simulate crack propagation behavior under gravity and hydraulic seepage pressure. The propagation trajectory along the direction of maximum principal stress is solved using a continuous integral risk density function, allowing the intrinsic relationship between the internal structural stress state of the cutoff wall and crack progression to be characterized by a physical model, thus forming a quantitative prediction of potential penetration risk in the cutoff wall. The construction process tightly couples the mechanical field model with the crack geometry parameters, thereby reflecting the evolution trend of crack propagation within a physical and mechanical interpretation framework, promoting the objective, standardized, and scientific basis of the engineering structure safety evaluation process.

[0057] In some possible implementations, the step of calculating and outputting an integrity index characterizing the integrity of the cutoff wall, based on crack segmentation results, crack length, bifurcation angle, inclination, and predicted potential penetration paths, includes:

[0058] Calculate the ratio of the crack pixel area to the total borehole wall area in the crack segmentation result to generate the crack density coefficient;

[0059] Based on the spatial relationship between the length and direction of the potential penetration path and the boundary of the cutoff wall, the risk level of path penetration is calculated.

[0060] The bifurcation angle and inclination of the cracks are weighted and fused to generate a crack morphology risk coefficient;

[0061] The following formula integrates the crack density coefficient, path penetration risk level, and crack morphology risk coefficient: Integrity Index = ;

[0062] In the formula, This represents the crack density coefficient, which is the ratio of the crack pixel area to the total borehole wall area. This indicates the path penetration risk level obtained by analyzing the path length and its spatial relationship with the cutoff wall boundary; This indicates the risk coefficient of crack morphology, which comprehensively reflects the risk of bifurcation angle and crack tendency. , and The weighting factor is a preset value, and satisfies the following conditions: By employing the aforementioned technical solution, a mathematical model is used to quantify the crack pixel area, crack geometric features, and risk information introduced by potential penetration path prediction obtained from crack segmentation into a unified integrity index through weighted calculations. This achieves a reasonable allocation of various indicators in quantitative evaluation, thereby reflecting the overall structural state of the cutoff wall in a single numerical form. This ensures the scientific and objective nature of the overall quality evaluation of the cutoff wall, while overcoming the uncertainty caused by human judgment. The weighted fusion method enables the cutoff wall integrity index to become an important basis for a unified quantitative expression of the structural safety of the cutoff wall, further realizing the standardization and systematization of the cutoff wall condition evaluation process.

[0063] In some possible implementations, the steps for generating a risk assessment report on cracks in a cutoff wall based on an integrity index include:

[0064] The system calls a pre-defined engineering measures database, which stores the mapping relationship between integrity index ranges and engineering measures.

[0065] The integrity index is compared with a preset threshold to determine the corresponding range;

[0066] Generate a risk assessment report containing conclusions and recommended measures based on the mapping relationship;

[0067] The risk assessment report is output to the display terminal. Using the above technical solution, a pre-set engineering measures database is used to map the numerical correspondence between the integrity index and preset thresholds, thereby achieving automatic conversion between the integrity index and risk level. The integrity index is directly mapped to the risk level after threshold comparison, and corresponding engineering measures are determined based on the mapping relationship to generate the risk assessment report. The risk assessment report generation process relies entirely on objective indicators and systematic algorithms, avoiding subjective bias caused by manual judgment and ensuring consistent risk level expression and scientific selection of engineering measures.

[0068] In some possible implementations, the step of applying an adaptive Gaussian high-pass filter to the frequency domain image to suppress specular reflection components and generating the filtered frequency domain image includes:

[0069] Extract frequency distribution features from the frequency domain image to generate frequency feature data;

[0070] The dominant frequency range of the specular reflection component is calculated based on the frequency characteristic data to generate the reflection frequency range;

[0071] The cutoff frequency and attenuation coefficient of the Gaussian high-pass filter are dynamically configured according to the reflection frequency range to generate adaptive filter parameters;

[0072] An adaptive filter parameter is applied to the frequency domain image, and a frequency domain filtering operation is performed to generate a filtered frequency domain image. This technical solution utilizes the frequency distribution characteristics of the frequency domain image to quantitatively suppress specular reflections. The dominant frequency range is determined by extracting frequency feature data, and the cutoff frequency and attenuation coefficient of the Gaussian high-pass filter are dynamically adjusted to achieve the filtering operation. The frequency range corresponding to the reflective components is accurately identified using spectral characteristics, and the filter parameters are automatically adjusted to ensure that only the necessary attenuation is applied to the reflective components during frequency domain filtering, while maintaining the complete transmission of borehole wall structural information. In this way, the frequency domain filtering effect has high robustness and adaptability, fully optimizing the image preprocessing quality and providing a well-lit and highly detail-preserving image input for subsequent accurate crack detection.

[0073] In some possible implementations, the feature map is processed using an oriented gradient histogram enhancement module to enhance crack edge information. The steps for generating the enhanced feature map include:

[0074] Calculate the gradient magnitude and gradient direction of the feature map at each pixel to generate gradient magnitude data and gradient direction data;

[0075] Construct an oriented gradient histogram based on gradient direction data, and generate a histogram distribution;

[0076] The dominant direction of the crack edge is identified based on the histogram distribution, and the gradient magnitude data in this direction is enhanced to generate enhanced gradient data.

[0077] Enhanced gradient data is fused into the feature map to generate an enhanced feature map. Using the above technique, the gradient magnitude and direction of each pixel in the image are extracted, and an oriented gradient histogram is constructed to represent crack edge information, thus achieving a full representation of the local structural features of the image. Specifically, the gradient magnitude and direction are mathematically calculated to establish a precise quantitative description, and the histogram method comprehensively represents local edge features, enabling the deep model to obtain fine and stable image edge information during automatic segmentation.

[0078] In some possible implementations, the step of performing morphological skeletonization on the crack segmentation results to generate a crack skeleton image includes:

[0079] Apply structuring elements to perform iterative erosion operations on the crack segmentation results to generate eroded images;

[0080] Subtract the eroded image from the crack segmentation result to generate the initial skeleton image;

[0081] A thinning algorithm is applied to the initial skeleton image to remove redundant pixels, generating an intermediate skeleton image;

[0082] Breakpoints in the intermediate skeleton image are detected and connected to generate a crack skeleton image. Using the above technical solution, structural elements combined with erosion and morphological processing are employed to continuously process the crack segmentation results, obtaining a skeleton image that reflects the true geometric characteristics of cracks in the cutoff wall. The entire processing effectively removes redundant noise during continuous updates, ensuring that reinforcing bars or other stray pixels do not interfere with the true crack structure, thus retaining only the fine line structures that effectively represent the crack information of the cutoff wall. The stability and consistency of continuous processing ensure that the generated skeleton image can completely and objectively reflect the topological characteristics of the cracks, providing a reliable basis for subsequent geometric parameter calculations. Attached Figure Description

[0083] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0084] Figure 1 A flowchart illustrating a method for intelligent quantification of cracks in borehole optical images for quality assessment of cut-off walls, provided for some embodiments of this application. Figure 1 ;

[0085] Figure 2 A flowchart illustrating a method for intelligent quantification of cracks in borehole optical images for quality assessment of cut-off walls, provided for some embodiments of this application. Figure 2 ;

[0086] Figure 3 A flowchart illustrating a method for intelligent quantification of cracks in borehole optical images for quality assessment of cut-off walls, provided for some embodiments of this application. Figure 3 ;

[0087] Figure 4 A flowchart illustrating a method for intelligent quantification of cracks in borehole optical images for quality assessment of cut-off walls, provided for some embodiments of this application. Figure 4 ;

[0088] Figure 5 A flowchart illustrating a method for intelligent quantification of cracks in borehole optical images for quality assessment of cut-off walls, provided for some embodiments of this application. Figure 5 ;

[0089] Figure 6 A flowchart illustrating a method for intelligent quantification of cracks in borehole optical images for quality assessment of cut-off walls, provided for some embodiments of this application. Figure 6 ;

[0090] Figure 7A flowchart illustrating a method for intelligent quantification of cracks in borehole optical images for quality assessment of cut-off walls, provided for some embodiments of this application. Figure 7 ;

[0091] Figure 8 A flowchart illustrating a method for intelligent quantification of cracks in borehole optical images for quality assessment of cut-off walls, provided for some embodiments of this application. Figure 8 ;

[0092] Figure 9 A flowchart illustrating a method for intelligent quantification of cracks in borehole optical images for quality assessment of cut-off walls, provided for some embodiments of this application. Figure 9 ;

[0093] Figure 10 A flowchart illustrating a method for intelligent quantification of cracks in borehole optical images for quality assessment of cut-off walls, provided for some embodiments of this application. Figure 10 . Detailed Implementation

[0094] Specific embodiments of the invention will now be described in detail. Although the invention is described in conjunction with these specific embodiments, it should be understood that it is not intended to limit the invention to these specific embodiments. Rather, these embodiments are intended to cover alternative, modified, or equivalent embodiments that may be included within the spirit and scope of the invention as defined by the claims. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. The invention may be practiced without some or all of these specific details.

[0095] When used in conjunction with the terms "comprising," "method comprising," or similar language in this specification and appended claims, the singular forms "a," "some," and "the" include plural references unless the context clearly indicates otherwise. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0096] Please see Figure 1 This application provides a method for intelligent quantification of cracks in borehole optical images for assessing the quality of cut-off walls, comprising the following steps:

[0097] S1: Acquire high-resolution images of the borehole wall captured by an endoscopic camera inserted into the borehole of the impermeable wall;

[0098] S2: Perform adaptive homomorphic filtering on the high-definition image of the borehole wall to eliminate specular reflection interference in the image and generate an illumination-invariant image;

[0099] S3: Input the illumination-invariant image into the preset crack segmentation model to generate crack segmentation results; the crack segmentation model is a U-Net++ network with fused directional gradient histogram enhancement module, used to identify crack pixels including cracks with a width of less than 0.1mm.

[0100] S4: Perform crack topological connectivity analysis on the crack segmentation results, and calculate the length, bifurcation angle and dip of the identified cracks;

[0101] S5: Based on the crack length, bifurcation angle and inclination data, combined with the preset soil stress field model, predict the potential penetration path of the crack in the seepage barrier wall.

[0102] S6: Calculate and output the integrity index characterizing the integrity of the cutoff wall based on the crack segmentation results, crack length, bifurcation angle, inclination, and predicted potential penetration path;

[0103] S7: Generate a risk assessment report on cracks in the anti-seepage wall based on the integrity index.

[0104] This method automatically extracts and quantifies crack information from high-resolution borehole wall images through overall image acquisition, preprocessing, segmentation, topology analysis, and risk assessment, achieving an objective evaluation of the borehole wall's structural condition step by step. High-resolution images acquired by the endoscopic camera are processed using adaptive homomorphic filtering to generate illumination-invariant images, ensuring the realistic representation of borehole wall structural details and effectively eliminating interference caused by uneven illumination and specular reflection. This guarantees that the acquired images only reflect the physical and structural features of the borehole wall itself. After the illumination-invariant images are processed by the fusion directional gradient histogram enhancement module and the U-Net++ network, crack pixels are automatically identified, achieving extremely high precision in the extraction of minute cracks. This allows for accurate capture of cracks with extremely small widths during quantitative segmentation, enabling the calculation of crack geometric features based on quantitative data. By segmenting the crack skeleton in the image, the length, bifurcation angle, and dip of the cracks are obtained, providing a reliable basis for subsequent quantitative derivation of the crack propagation process using physical models. This method, employing a pre-defined soil stress field model to predict potential crack penetration paths, combines physical and mechanical parameters with crack geometry information to a certain extent, enabling a physical interpretation and quantitative estimation of crack propagation trends and penetration risks. The integrity index, derived by weighted fusion of crack density, geometric risk, and propagation risk, objectively reflects the overall structural condition and engineering risk level of the cutoff wall, thus achieving a comprehensive assessment of the cutoff wall's construction quality and structural safety. The entire technical solution, while ensuring scientific, standardized, and objective data processing and quantitative analysis, avoids errors and inconsistencies caused by subjective judgment in manual inspection, effectively solving technical problems such as the difficulty in manually detecting micro-cracks, the strong subjectivity of quantitative judgment of geometric features, and low inspection efficiency.

[0105] Please see Figure 2 Preferably, in some embodiments, S2: the step of performing adaptive homomorphic filtering on the high-definition image of the borehole wall to eliminate specular reflection interference in the image and generate an illumination-invariant image includes:

[0106] S21: Convert the high-resolution image of the borehole wall into a logarithmic domain image to generate the logarithmic domain image. The specific formula is as follows:

[0107]

[0108] In the formula, This indicates the original high-resolution image of the borehole wall in coordinates. Pixel value at; A positive constant set to prevent the logarithmic function input from being zero; Represents the transformed logarithmic field image;

[0109] S22: Perform a Fast Fourier Transform on the logarithmic domain image to generate a frequency domain image. The specific formula is as follows:

[0110]

[0111] In the formula, and These are the width and height of the image, respectively; and These are the horizontal and vertical frequency coordinates of the frequency domain image, respectively; The imaginary unit; This represents the obtained frequency domain image; Represents the natural exponential function;

[0112] S23: Apply an adaptive Gaussian high-pass filter to the frequency domain image to suppress specular reflection components and generate a filtered frequency domain image. The specific formula is as follows:

[0113]

[0114] In the formula, Represented in frequency coordinates The filter response at point is given by the following formula:

[0115]

[0116] in, and These represent the center coordinates of the dominant specular reflection frequencies identified from the frequency distribution characteristics; The cutoff characteristics of the filter are controlled by the standard deviation parameter adaptively calculated based on the specular reflection component.

[0117] S24: Perform an inverse fast Fourier transform on the filtered frequency domain image to generate the inverse transform image. The specific formula is as follows: ;

[0118] S25: Convert the inverse transform image to the linear domain to generate an illumination-invariant image. The specific formula is as follows: The above technical solution utilizes logarithmic domain transformation and Fourier transform combined with an adaptive Gaussian high-pass filter to process the image, directly suppressing and eliminating components related to illumination variations and specular reflections in the high-definition borehole wall image from the frequency domain. After mathematical mapping, interference caused by uneven brightness in the original image is balanced. After obtaining the image frequency information using Fourier transform, specific frequency components are suppressed by designing filters, achieving separation of illumination and specular reflection components. After reconstructing the image spatial information using inverse Fourier transform, it is restored to the linear domain through exponential function mapping, ultimately obtaining an illumination-invariant image reflecting the actual borehole wall structural characteristics. In this way, factors related to ambient lighting and reflection interference can be seamlessly removed from the image, ensuring that the image data input to subsequent processing modules remains highly stable and reliable. This image preprocessing result provides a solid foundation for subsequent automatic segmentation and crack identification, ensuring that the detailed information obtained during the detection process truly reflects the structural characteristics of the seepage barrier wall.

[0119] Please see Figure 3 Preferably, in some embodiments, S3: the step of inputting the illumination-invariant image into a preset crack segmentation model to generate crack segmentation results includes:

[0120] S31: Use the encoder part of the U-Net network to extract multi-scale feature maps of the illumination-invariant image and generate feature maps;

[0121] S32: Apply the Histogram of Oriented Gradients (HGP) enhancement module to process the feature map, thereby enhancing crack edge information and generating an enhanced feature map;

[0122] S33: The decoder part of the U-Net network is used to upsample and skip connections the enhanced feature map to generate an initial segmentation map;

[0123] S34: Apply the Sigmoid activation function to the initial segmentation map to generate a probability map;

[0124] S35: Binarize the probability map based on a preset threshold to generate crack segmentation results. Using the above technical solution, the illumination-invariant image is directly fed into a deep neural network. Multi-scale feature extraction and histogram of oriented gradients (HOR) techniques are used to calculate and enhance the gradient information of the crack contour in the image, quantifying the crack edge features mathematically. This allows the deep network to accurately identify the pixel-level boundaries of tiny cracks in the image during automatic segmentation. The real structural information of the borehole wall contained in the illumination-invariant image is extracted and upsampled back through the segmentation network, ensuring that the true morphology and coherent structure of the crack are reflected in the segmentation output. The HOR technique combined with the cosine operator is used to quantitatively analyze the edge direction, ensuring the stability and accuracy of the crack edge information after mathematical processing. This guarantees that the crack segmentation results output by the deep network can objectively reflect the actual situation of the seepage barrier wall cracks. In this way, the image segmentation process can rely entirely on automated data processing, achieving high-precision capture and consistent quantification of crack edges, further providing scientific evidence to support subsequent calculations of crack geometric features and engineering quality assessment.

[0125] Please see Figure 4 Preferably, in some embodiments, S4: the step of performing crack topological connectivity analysis on the crack segmentation results and calculating the length, bifurcation angle, and inclination of the identified cracks includes:

[0126] S41: Perform morphological skeletonization on the crack segmentation results to generate a crack skeleton image;

[0127] S42: Based on the crack skeleton image, identify crack branch points and endpoints, and generate topological node data;

[0128] S43: Calculate the Euclidean distance between each crack pixel in the crack skeleton image to obtain the crack length and generate crack length data; the specific formula for calculating the crack length is as follows:

[0129] In the formula, Represents the first in the skeleton image The Euclidean distance between adjacent pixels; Indicates the total length of the identified cracks;

[0130] S44: Analyze the vector angle between adjacent branches in the topology node data, calculate the bifurcation angle, and generate bifurcation angle data; the specific formula for calculating the bifurcation angle is as follows:

[0131]

[0132] In the formula, and The direction vectors of the two adjacent branches at the bifurcation point; " denotes the vector dot product; and Representing vectors respectively and The modulus length; To calculate the bifurcation angle

[0133] S45: Based on the pixel gradient direction of the crack skeleton image, calculate the crack inclination and generate inclination data; the specific calculation formula for crack inclination is as follows:

[0134]

[0135] In the formula, This represents the set of all pixels in the crack skeleton image; For at pixel Gradient magnitude at point, formula ,in It is an illumination-invariant image generated through homomorphic filtering; For pixels The gradient direction at that location, i.e., the direction of local change in that pixel; This represents the overall crack tendency or average angle obtained from the weighted average gradient direction of all skeleton pixels.

[0136] S46: Integrate crack length data, bifurcation angle data, and dip data to output the crack length, bifurcation angle, and dip. Using the above technical solution, morphological skeletonization simplifies the crack region into a single-pixel-width connected structure, fully preserving the overall crack appearance and clearly presenting it spatially. Simultaneously, topological connectivity analysis detects and processes nodes and branches in the skeleton, achieving quantitative calculation of continuous geometric parameters. Stable and clear skeleton information is extracted from the image data, and based on this skeleton information, a rigorous quantitative relationship is established for the geometric features of the seepage barrier cracks, achieving standardized and efficient expression of the crack structure, thus providing a scientific data basis for subsequent engineering inspection.

[0137] Please see Figure 5 Preferably, in some embodiments, S5: the step of predicting the potential penetration path of cracks in the anti-seepage wall based on crack length, bifurcation angle, and inclination data, combined with a preset soil stress field model, includes:

[0138] S51: Construct a three-dimensional stress field mesh model based on the soil physical parameters of the seepage barrier wall, and generate an initial stress field model;

[0139] The initial stress field model can be expressed as a solution to the equilibrium equations. When the no-volume loading condition is satisfied within the region, its mathematical expression is:

[0140]

[0141] In the formula, In the initial stress field model, the spatial coordinates are represented as follows: Stress Quantity; The first in the space The volume force in the direction can include gravity and other forces; solving this equation requires satisfying the corresponding boundary conditions to ensure that the equilibrium state is satisfied within the structural domain.

[0142] S52: Map the crack length, bifurcation angle, and dip direction data to the corresponding positions in the initial stress field model to generate a stress field model containing crack parameters;

[0143] S53: The stress field model containing crack parameters is based on the initial stress field model, introducing the incremental stress field generated by the crack parameters, and expressed using the superposition principle as follows:

[0144]

[0145] In the formula, This represents the overall stress field after considering the effects of crack parameters. Quantity; This represents the stress increment introduced by the crack parameters, which can be described by an integral expression.

[0146]

[0147] in, This represents the integral domain that includes the area affected by the crack; This represents the stress response kernel obtained based on the Green's function or influence function, which describes the contribution of crack parameters per unit volume to the stress field at a point in space. This represents the crack parameter distribution function, which reflects the mapping relationship between the crack's geometric characteristics (such as crack length, bifurcation angle, and dip) and its location. Represents the integral volume element;

[0148] S54: Based on the elastoplastic constitutive equation, simulate the crack propagation behavior of the stress field model containing crack parameters under gravity load and hydraulic seepage pressure, and generate crack propagation simulation data.

[0149] S55: Extract the continuous propagation trajectory along the direction of maximum principal stress from the crack propagation simulation data to generate potential penetration paths. Using the above technical solution, a three-dimensional stress field mesh model is constructed to numerically map the physical stress state inside the cutoff wall with the crack geometry data. Crack length, bifurcation angle, and dip direction data are directly quantified and embedded within the mesh model, enabling the elastoplastic constitutive equations to accurately simulate crack propagation behavior under gravity and hydraulic seepage pressure. The propagation trajectory along the direction of maximum principal stress is solved using a continuous integral risk density function, allowing the intrinsic relationship between the internal structural stress state of the cutoff wall and crack progression to be characterized by a physical model, thus forming a quantitative prediction of potential penetration risk in the cutoff wall. The construction process tightly couples the mechanical field model with the crack geometry parameters, thereby reflecting the evolution trend of crack propagation within a physical and mechanical interpretation framework, promoting the objective, standardized, and scientific basis of the engineering structure safety evaluation process.

[0150] Please see Figure 6 Preferably, in some embodiments, S6: the step of calculating and outputting an integrity index characterizing the integrity of the cutoff wall based on the crack segmentation results, crack length, bifurcation angle, inclination, and predicted potential penetration paths includes:

[0151] S61: Calculate the ratio of the crack pixel area to the total borehole wall area in the crack segmentation result, and generate the crack density coefficient.

[0152] S62: Calculate the path penetration risk level based on the spatial relationship between the length and direction of the potential penetration path and the boundary of the cutoff wall;

[0153] S63: Weighted fusion of crack bifurcation angle and inclination to generate crack morphology risk coefficient;

[0154] S64: Integrate crack density coefficient, path penetration risk level, and crack morphology risk coefficient based on the following formula: Integrity Index = ;

[0155] In the formula, This represents the crack density coefficient, which is the ratio of the crack pixel area to the total borehole wall area. This indicates the path penetration risk level obtained by analyzing the path length and its spatial relationship with the cutoff wall boundary; This indicates the risk coefficient of crack morphology, which comprehensively reflects the risk of bifurcation angle and crack tendency. , and The weighting factor is a preset value, and satisfies the following conditions: By employing the aforementioned technical solution, a mathematical model is used to quantify the crack pixel area, crack geometric features, and risk information introduced by potential penetration path prediction obtained from crack segmentation into a unified integrity index through weighted calculations. This achieves a reasonable allocation of various indicators in quantitative evaluation, thereby reflecting the overall structural state of the cutoff wall in a single numerical form. This ensures the scientific and objective nature of the overall quality evaluation of the cutoff wall, while overcoming the uncertainty caused by human judgment. The weighted fusion method enables the cutoff wall integrity index to become an important basis for a unified quantitative expression of the structural safety of the cutoff wall, further realizing the standardization and systematization of the cutoff wall condition evaluation process.

[0156] Please see Figure 7 Preferably, in some embodiments, step S7: generating a risk assessment report on cracks in the anti-seepage wall based on the integrity index includes:

[0157] S71: Call the preset engineering measures database, which stores the mapping relationship between integrity index range and engineering measures;

[0158] S72: Compare the integrity index with the preset threshold to determine the corresponding interval;

[0159] S73: Generate a risk assessment report containing conclusions and recommended measures based on the mapping relationship;

[0160] S74: Output the risk assessment report to the display terminal. Using the above technical solution, a pre-set engineering measures database is used to map the numerical correspondence between the integrity index and preset thresholds, thereby achieving automatic conversion between the integrity index and risk level. The integrity index is directly mapped to the risk level after threshold comparison, and corresponding engineering measures are determined based on the mapping relationship to form a risk assessment report. The risk assessment report generation process relies entirely on objective indicators and systematic algorithms, avoiding subjective bias caused by manual judgment and ensuring consistent risk level expression and scientific selection of engineering measures.

[0161] Please see Figure 8 Preferably, in some embodiments, S23: applying an adaptive Gaussian high-pass filter to the frequency domain image to suppress specular reflection components and generating a filtered frequency domain image includes:

[0162] S231: Extract the frequency distribution features of the frequency domain image and generate frequency feature data;

[0163] S232: Calculate the dominant frequency range of the specular reflection component based on frequency characteristic data, and generate the reflection frequency range;

[0164] S233: Dynamically configure the cutoff frequency and attenuation coefficient of the Gaussian high-pass filter according to the reflection frequency range to generate adaptive filter parameters;

[0165] S234: Apply adaptive filter parameters to the frequency domain image, perform frequency domain filtering, and generate a filtered frequency domain image. This technical solution utilizes the frequency distribution characteristics of the frequency domain image to quantitatively suppress specular reflections. The dominant frequency range is determined by extracting frequency feature data, and the cutoff frequency and attenuation coefficient of the Gaussian high-pass filter are dynamically adjusted to achieve the filtering operation. The frequency range corresponding to the reflective components is accurately identified using spectral characteristics, and the filter parameters are automatically adjusted to ensure that only necessary attenuation is applied to the reflective components during frequency domain filtering, while maintaining the complete transmission of borehole wall structural information. In this way, the frequency domain filtering effect has high robustness and adaptability, fully optimizing the image preprocessing quality and providing a balanced and highly detail-preserving image input for subsequent accurate crack detection.

[0166] Please see Figure 9 Preferably, in some embodiments, S32: applying the directional gradient histogram enhancement module to process the feature map to enhance crack edge information, the step of generating the enhanced feature map includes:

[0167] S321: Calculate the gradient magnitude and gradient direction of the feature map at each pixel, and generate gradient magnitude data and gradient direction data;

[0168] S322: Construct an oriented gradient histogram based on gradient direction data and generate a histogram distribution;

[0169] S323: Identify the dominant direction of the crack edge based on the histogram distribution, and enhance the gradient magnitude data in this direction to generate enhanced gradient data;

[0170] S324: Enhanced gradient data is fused into the feature map to generate an enhanced feature map. Using the above technical solution, the gradient magnitude and direction of each pixel in the image are extracted, and an oriented gradient histogram is constructed to represent crack edge information, thus achieving a full expression of the local structural features of the image. Specifically, the gradient magnitude and direction are mathematically calculated to establish a precise quantitative description, and the histogram method comprehensively represents local edge features, enabling the deep model to obtain fine and stable image edge information during automatic segmentation.

[0171] Please see Figure 10 Preferably, in some embodiments, S41: the step of performing morphological skeletonization processing on the crack segmentation results to generate a crack skeleton image includes:

[0172] S411: Apply structuring elements to perform iterative erosion on the crack segmentation results to generate an eroded image;

[0173] S412: Subtract the eroded image from the crack segmentation result to generate the initial skeleton image;

[0174] S413: Apply a thinning algorithm to the initial skeleton image to remove redundant pixels and generate an intermediate skeleton image;

[0175] S414: Detect and connect breakpoints in the intermediate skeleton image to generate a crack skeleton image. Using the above technical solution, structural elements combined with erosion and morphological processing are employed to continuously process the crack segmentation results, obtaining a skeleton image that reflects the true geometric characteristics of cracks in the anti-seepage wall. The entire processing effectively removes redundant noise during continuous updates, ensuring that reinforcing bars or other stray pixels do not interfere with the true crack structure, thus retaining only the fine line structures that effectively represent the crack information of the anti-seepage wall. The stability and consistency of continuous processing ensure that the generated skeleton image can completely and objectively reflect the topological characteristics of the cracks, providing a reliable basis for subsequent geometric parameter calculations.

[0176] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for intelligent quantification of cracks in borehole optical images for diaphragm wall quality assessment, characterized in that, Includes the following steps: Acquire high-resolution images of the borehole wall captured by an endoscopic camera inserted into the borehole of the impermeable wall; Adaptive homomorphic filtering is applied to the high-definition image of the borehole wall to eliminate specular reflection interference in the image and generate an illumination-invariant image; The illumination-invariant image is input into a preset crack segmentation model to generate crack segmentation results; the crack segmentation model is a U-Net++ network with a fused directional gradient histogram enhancement module, used to identify crack pixels including cracks with a width of less than 0.1 mm. The crack segmentation results are subjected to crack topological connectivity analysis to calculate the length, bifurcation angle, and dip of the identified cracks, including the following steps: The crack segmentation results are subjected to morphological skeletonization processing to generate a crack skeleton image; Based on the crack skeleton image, crack branch points and endpoints are identified, and topological node data is generated. Euclidean distances of each crack pixel in the crack skeleton image are calculated to obtain crack length data; wherein, a specific calculation formula of the crack length is as follows: In the formula, Represents the first in the skeleton image The Euclidean distance between adjacent pixels; Indicates the total length of the identified cracks; Analyze the vector angle between adjacent branches in the topological node data, calculate the bifurcation angle, and generate bifurcation angle data; the specific calculation formula for the bifurcation angle is as follows: In the formula, and The direction vectors of the two adjacent branches at the bifurcation point; " denotes the vector dot product; and Representing vectors respectively and The modulus length; To calculate the bifurcation angle Based on the pixel gradient direction of the crack skeleton image, the crack inclination is calculated to generate inclination data; the specific calculation formula for crack inclination is as follows: In the formula, This represents the set of all pixels in the crack skeleton image; For at pixel Gradient magnitude at point, formula ,in It is an illumination-invariant image generated through homomorphic filtering; For pixels The gradient direction at that location, i.e., the direction of local change in that pixel; This represents the overall crack tendency or average angle obtained from the weighted average gradient direction of all skeleton pixels. By integrating the crack length data, bifurcation angle data, and dip data, the crack length, bifurcation angle, and dip are output. Based on the length, bifurcation angle, and inclination data of the crack, and combined with a pre-set soil stress field model, the potential penetration path of the crack in the cutoff wall is predicted, including the following steps: A three-dimensional stress field mesh model is constructed based on the soil physical parameters of the seepage barrier wall to generate an initial stress field model; The length, bifurcation angle, and dip direction data of the crack are mapped to the corresponding positions in the initial stress field model to generate a stress field model containing crack parameters. Based on the elastoplastic constitutive equation, the crack propagation behavior of the stress field model containing crack parameters under gravity load and hydraulic seepage pressure is simulated to generate crack propagation simulation data. Extract the continuous propagation trajectory along the direction of maximum principal stress from the crack propagation simulation data to generate the potential penetration path; Based on the crack segmentation results, crack length, bifurcation angle, inclination, and predicted potential penetration paths, calculate and output an integrity index characterizing the integrity of the cutoff wall. A risk assessment report on cracks in the anti-seepage wall is generated based on the integrity index.

2. The method of claim 1, wherein, The steps of performing adaptive homomorphic filtering on the high-resolution image of the borehole wall to eliminate specular reflection interference and generate an illumination-invariant image include: The high-resolution image of the borehole wall is converted into a logarithmic domain image, and the specific formula for generating the logarithmic domain image is as follows: In the formula, This indicates the original high-resolution image of the borehole wall in coordinates. Pixel value at; A positive constant set to prevent the logarithmic function input from being zero; Represents the transformed logarithmic field image; Perform a Fast Fourier Transform on the logarithmic domain image to generate a frequency domain image, using the following formula: In the formula, and These are the width and height of the image, respectively; and These are the horizontal and vertical frequency coordinates of the frequency domain image, respectively; The imaginary unit; This represents the obtained frequency domain image; Represents the natural exponential function; An adaptive Gaussian high-pass filter is applied to the frequency domain image to suppress specular reflection components, generating a filtered frequency domain image. The specific formula is as follows: wherein represents the filter response at the frequency coordinate ω, specifically given by the formula: wherein, with respectively represent the center coordinates of the dominant frequency of the specular reflection identified from the frequency distribution characteristics; is a standard deviation parameter adaptively calculated from the specular reflection component, controlling the cutoff characteristics of the filter; Perform an inverse fast Fourier transform on the filtered frequency domain image to generate an inverse transform image, as shown in the following formula: ; Converting the inverse-transformed image into a linear domain to generate the illumination invariance image, specifically as follows: .

3. The method of claim 1, wherein, The steps of inputting the illumination-invariant image into a preset crack segmentation model to generate crack segmentation results include: The encoder portion of the U-Net network is used to extract multi-scale feature maps of the illumination-invariant image to generate feature maps; The feature map is processed by the Histogram of Oriented Gradients (HGP) enhancement module to enhance crack edge information and generate an enhanced feature map. The decoder portion of the U-Net network is used to upsample and skip connections on the enhanced feature map to generate an initial segmentation map; Apply the Sigmoid activation function to the initial segmentation map to generate a probability map; The probability map is binarized based on a preset threshold to generate the crack segmentation result.

4. The method of claim 1, wherein, The steps for calculating and outputting an integrity index characterizing the integrity of the cutoff wall based on the crack segmentation results, crack length, bifurcation angle, inclination, and predicted potential penetration paths include: Calculate the ratio of the crack pixel area to the total borehole wall area in the crack segmentation result to generate the crack density coefficient; Based on the spatial relationship between the length and direction of the potential penetration path and the boundary of the cutoff wall, the penetration risk level of the path is calculated. The bifurcation angle and inclination of the cracks are weighted and fused to generate a crack morphology risk coefficient; The fracture density factor, the path crossing risk level, and the fracture morphology risk factor are integrated based on the following equation: integrity index = 1 - (fracture density factor) x (path crossing risk level) x (fracture morphology risk factor) ; In the formula, This represents the crack density coefficient, which is the ratio of the crack pixel area to the total borehole wall area. This indicates the path penetration risk level obtained by analyzing the path length and its spatial relationship with the cutoff wall boundary; This indicates the risk coefficient of crack morphology, which comprehensively reflects the risk of bifurcation angle and crack tendency. , and The weighting factor is a preset value, and satisfies the following conditions: .

5. The method of claim 1, wherein, The steps for generating a risk assessment report on cracks in the anti-seepage wall based on the integrity index include: A preset engineering measures database is invoked, which stores the mapping relationship between integrity index ranges and engineering measures; The integrity index is compared with a preset threshold to determine the corresponding interval; A risk assessment report containing conclusions and recommended measures is generated based on the mapping relationship; Output the risk assessment report to the display terminal.

6. The method of claim 2, wherein the method is a method of borehole optical image crack intelligent quantification for diaphragm wall quality evaluation. The step of applying an adaptive Gaussian high-pass filter to the frequency domain image to suppress specular reflection components and generating a filtered frequency domain image includes: Extract the frequency distribution features of the frequency domain image to generate frequency feature data; Based on the frequency characteristic data, the dominant frequency range of the specular reflection component is calculated to generate the reflection frequency range; Based on the reflected frequency range, the cutoff frequency and attenuation coefficient of the Gaussian high-pass filter are dynamically configured to generate adaptive filter parameters. The adaptive filter parameters are applied to the frequency domain image to perform a frequency domain filtering operation, generating a filtered frequency domain image.

7. The method of claim 3, wherein the method is a method of borehole optical image crack intelligent quantification for diaphragm wall quality evaluation. The step of applying the directional gradient histogram enhancement module to process the feature map to enhance crack edge information and generate the enhanced feature map includes: Calculate the gradient magnitude and gradient direction of the feature map at each pixel to generate gradient magnitude data and gradient direction data; Based on the gradient direction data, an oriented gradient histogram is constructed, and a histogram distribution is generated; The dominant direction of the crack edge is identified based on the histogram distribution, and the gradient magnitude data in this direction is enhanced to generate enhanced gradient data. The enhanced gradient data is fused into the feature map to generate an enhanced feature map.

8. The method of claim 1, wherein, The steps of performing morphological skeletonization processing on the crack segmentation results to generate a crack skeleton image include: An iterative erosion operation is performed on the crack segmentation results using structuring elements to generate an eroded image. Subtract the eroded image from the crack segmentation result to generate an initial skeleton image; A thinning algorithm is applied to the initial skeleton image to remove redundant pixels, generating an intermediate skeleton image; Breakpoints in the intermediate skeleton image are detected and connected to generate a crack skeleton image.

Images