A method for judging severity of crack damage based on skeleton topology and gray coupling

Abstract

The application discloses a crack damage severity discrimination method based on skeleton topology and gray coupling and belongs to the technical field of computer vision and road structure health monitoring, and comprises the following steps: obtaining a crack gray image of an infrastructure structure surface and extracting a crack binary segmentation mask; performing skeleton thinning on the crack binary segmentation mask to obtain a skeleton graph with a single-pixel width, and mapping the skeleton graph into a crack undirected topology graph; according to the gray attenuation information of a crack region in the crack gray image, edge weights are given to each edge of the crack undirected topology graph; according to the edge weights, the weighted ring closure degree and the weighted directional dissipation entropy of the crack undirected topology graph are calculated; and according to the weighted ring closure degree, the weighted directional dissipation entropy and the maximum value of the edge weights of each edge in the crack undirected topology graph, a three-dimensional damage state vector is constructed to determine the crack damage severity. The application realizes objective quantitative discrimination of the crack severity, which is completely independent of artificial experience and does not depend on illumination color conditions.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and road structure health monitoring technology, and particularly relates to a method for judging the severity of crack damage based on skeleton topology and grayscale coupling. Background Technology

[0002] In the field of automated inspection of infrastructure such as highways and bridges, visual detection and damage assessment of pavement cracks are core components of structural health monitoring. Current mainstream automated detection processes typically involve two steps: first, using deep learning-based semantic segmentation or object detection models to extract crack pixel regions from overhead images; then, based on the segmentation results, performing geometric statistics, such as calculating the total crack pixel area, the length of the maximum bounding rectangle, or the average crack width, and accordingly classifying the damage into empirical levels such as mild, moderate, or severe.

[0003] The above-mentioned technical approach has been applied to some extent in engineering practice, but it still has inherent limitations on two levels.

[0004] First, the robustness of crack connectivity assessment is insufficient. Real road surface images are affected by uneven lighting, asphalt aggregate particle occlusion, stone reflection, and oil contamination, causing continuous real cracks to often appear as discrete, fragmented regions after imaging. To obtain the complete crack morphology, traditional algorithms often require designing numerous artificially heuristic rules (such as direction prediction, endpoint distance thresholds, and local curvature constraints) to stitch together breakpoints. However, these rules are highly sensitive to image quality, involve complex parameter adjustments, and struggle to accommodate the diversity of different road materials and lighting conditions. They are prone to misjudging background texture noise as crack extensions or missing real crack breakpoints, resulting in severe distortion of the segmented crack topology.

[0005] Secondly, the assessment of damage severity relies on a single dimension and is disconnected from the structural mechanics failure mechanism. Existing grading methods mainly depend on the pixel area or maximum circumscribed geometric dimension of cracks. This statistical approach confuses two fundamentally different damage modes: a long but unidirectional linear crack and a network of interwoven cracks forming a fractured area. While they may have similar pixel areas, their impact on the pavement structure's load-bearing capacity is drastically different. The former typically corresponds to localized tensile stress release, with relatively limited harm; the latter indicates multidirectional cracking of the surface material, loss of aggregate interlocking, and a serious threat to the structural integrity. However, traditional area statistics cannot distinguish between these two topological forms and lack mathematical tools to quantify the degree of structural degradation from dimensions such as the connectivity, directional distribution, and number of closed loops in the crack network. The final grading determination still largely depends on manual visual interpretation, which is highly subjective, has poor repeatability, and fails to meet the objective, quantitative, and traceable assessment requirements of large-scale automated inspections.

[0006] In summary, existing methods for automatic detection and severity assessment of pavement cracks have significant shortcomings in handling interference-resistant connectivity and mapping deep mechanical states. There is an urgent need for a more robust discrimination method that does not rely on manual rule splicing and can reflect material failure modes from the spatial structure level of cracks. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a method for determining the severity of crack damage based on skeleton topology and grayscale coupling, comprising the following steps: Obtain grayscale images of cracks on the surface of infrastructure structures; Based on the grayscale image of the crack, extract the binary segmentation mask of the crack; The binary segmentation mask of the crack is skeletonized to obtain a skeleton map with a single pixel width, and the skeleton map is mapped to an undirected topological graph of the crack containing nodes and edges. Based on the grayscale attenuation information of the crack region in the crack grayscale image, each edge of the crack undirected topological graph is assigned an edge weight. Based on the edge weights, calculate the weighted circumferential closure and weighted directional dissipation entropy of the crack undirected topology graph; A three-dimensional damage state vector is constructed based on the weighted circumferential closure, the weighted directional dissipation entropy, and the maximum value of the edge weights of each edge in the crack undirected topology graph. The severity of crack damage is determined based on the three-dimensional damage state vector.

[0008] Optionally, based on the crack grayscale image, a crack binary segmentation mask is extracted, including: A combination of top-hat transform and contrast-limited adaptive histogram equalization is performed on the crack grayscale image to obtain an enhanced crack grayscale image. The enhanced crack grayscale image is input into a pre-trained semantic segmentation network to extract a crack probability map; After edge refinement of the crack probability map, it is binarized to output a crack binary segmentation mask.

[0009] Optionally, the binary segmentation mask of the crack is skeletonized to obtain a skeleton map with a single pixel width, and the skeleton map is mapped to an undirected topological graph of the crack containing nodes and edges, including: The morphological skeletonization algorithm is used to refine the binary segmentation mask of the crack to obtain a skeleton map with a single pixel width, and the original gray value of each pixel on the skeleton map and the gray gradient change rate of each pixel in the direction perpendicular to the skeleton are recorded. Using the crack intersections in the skeleton graph as nodes and the crack branches as edges, an undirected crack topology graph is constructed. The crack intersections are pixels with three or more connecting branches in their neighborhood, and the crack endpoints are pixels with one connecting branch in their neighborhood.

[0010] Optionally, based on the grayscale attenuation information of the crack region in the crack grayscale image, each edge of the undirected topological graph of the crack is assigned an edge weight, including: For any pixel on the skeleton map, calculate the local gradient direction at the pixel, and extend a predetermined pixel distance to both sides along the normal direction of the local gradient direction to perform local background sampling. Remove sampling points that are less than a predetermined pixel threshold from the edge of the crack, and obtain the average gray value and standard deviation of the background gray value from the remaining sampling points. Based on the difference between the average grayscale value of the background and the original grayscale value of any pixel on the skeleton map, adaptive scaling is performed in conjunction with the standard deviation of the background grayscale to obtain the adaptive grayscale loss ratio of any pixel. The adaptive grayscale loss ratio of all pixels contained in each edge of the undirected topology graph of the crack is accumulated along the path to obtain the equivalent grayscale loss modulus of each edge. The equivalent grayscale loss modulus is normalized and mapped to obtain the edge weight of each edge.

[0011] Optionally, before obtaining the adaptive grayscale loss ratio of any pixel by adaptive scaling based on the difference between the average grayscale value of the background and the original grayscale value of any pixel on the skeleton map, combined with the standard deviation of the background grayscale, the method further includes: Perform a large-scale morphological closing operation on the crack grayscale image to obtain a virtual background image; The virtual background image is fitted with a two-dimensional polynomial surface, and the residual field is extracted to remove global illumination variations. Based on the local grayscale standard deviation of the crack grayscale image and the pre-calibrated material standard texture base, the denominator of the adaptive grayscale loss ratio is normalized and compensated. The material standard texture base is pre-calibrated in the following way: a standard image of the undamaged road surface is acquired under a lighting environment with standard light source and fixed geometric conditions, the standard deviation of the neighborhood gray level is calculated pixel by pixel, the standard deviation histogram of the whole image is statistically analyzed, and the peak value is taken as the material standard texture base.

[0012] Optionally, based on the edge weights, the weighted cyclic closure of the undirected topological graph of the crack is calculated, including: Calculate the cyclospace dimension of the undirected topological graph of the crack based on the number of edges and nodes. When the dimension of the circle space is greater than or equal to one, all basic cycles in the undirected topology graph of the crack are traversed. For each basic cycle, the sum of the edge weights of all edges in each basic cycle is divided by the number of edges contained in each basic cycle to obtain the weighted average weight of each basic cycle. The weighted average weights of all basic loops are summed to obtain the weighted loop closure. At the same time, traverse all basic cycles and take the maximum value among the sums of the edge weights of each basic cycle as the total weight of the largest cycle.

[0013] Optionally, based on the edge weights, the weighted directional dissipation entropy of the undirected topological graph of the crack is calculated, including: Divide the angle interval into a predetermined number of histogram intervals; For each edge in the undirected topology graph of the crack, calculate the geometric tilt angle of each edge, and add the edge weight of each edge as a voting weight to the histogram interval corresponding to the geometric tilt angle to obtain the weighted cumulative value of each histogram interval. Calculate the probability distribution based on the proportion of the weighted cumulative value of each histogram interval to the total weighted cumulative value of all histogram intervals; The information entropy is calculated based on the probability distribution to obtain the weighted directional dissipation entropy.

[0014] Optionally, the three-dimensional damage state vector includes the weighted cyclic closure, the weighted directional dissipation entropy, and the maximum value of the edge weights of each edge in the crack undirected topological graph; based on the three-dimensional damage state vector, the severity of crack damage is determined, including: The three components of the three-dimensional damage state vector are coupled by a weighted power exponent to obtain a scalar damage criterion; The scalar damage criterion is compared with a pre-calibrated first threshold, second threshold, and third threshold, respectively. Based on the comparison results, determine the degradation phase to which the severity of crack damage belongs.

[0015] Optionally, the pre-calibrated first, second, and third thresholds are obtained in the following ways: Fracture tests were conducted on standard specimens, and images of the entire process from microcrack initiation to macroscopic fracture were simultaneously acquired. The scalar damage criterion value obtained after processing the image corresponding to the end point of the initial linear segment during the experiment is calibrated as the first threshold. The scalar damage criterion value obtained by processing the image corresponding to the peak load time during the test is calibrated as the second threshold. The scalar damage criterion value obtained by processing the image corresponding to the moment when the residual strength drops to a predetermined percentage threshold of the peak load during the test is calibrated as the third threshold.

[0016] Optionally, based on the comparison results, determine the degradation phase to which the severity of crack damage belongs, including: When the weighted circumferential closure is equal to zero and the scalar damage criterion is less than the first threshold, it is determined to be an intact micro-phase; When the weighted circumferential closure is equal to zero and the scalar damage criterion is greater than or equal to the second threshold, it is determined to be a linear open fracture phase; When the weighted circumferential closure is greater than or equal to one and the scalar damage criterion is greater than or equal to the third threshold, it is determined to be a network dissipative structure phase; When the weighted circumferential closure is greater than or equal to one, the weighted directional dissipation entropy approaches its maximum value, and the maximum edge weight increases significantly, it is determined to be a chaotic structure collapse phase.

[0017] Compared with the prior art, the present invention has the following advantages and technical effects: This invention combines crack image processing with topological graph theory to construct a complete and objective structural damage quantitative assessment mechanism.

[0018] This method abandons the traditional coarse statistical approach that relies on pixel area or circumscribed size. Instead, it transforms the visual segmentation results into an undirected topological graph that represents the connectivity of cracks, thus fundamentally preserving the spatial structure information of the crack network and avoiding fragmentation recognition errors caused by uneven lighting or material occlusion.

[0019] Building upon this foundation, this invention assigns a weight to each edge in the topology graph using local grayscale attenuation information from the crack region. This weight is not a simple measure of length or area, but effectively reflects the integral effect of the actual opening displacement of crack branches, enabling subsequent assessments to distinguish between true structural cracks and shallow scratches. Furthermore, this invention calculates weighted circumferential closure and weighted directional dissipation entropy based on the weighted topology graph. These two indicators characterize the degradation process of pavement structure from single crack expansion to network collapse from two independent dimensions: the degree to which the crack network forms closed cutting blocks and the degree of disorder in the distribution of crack directions. This overcomes the deficiency of traditional methods in being unable to distinguish the essential differences between linear cracks and network cracks of the same area.

[0020] Finally, by combining the above topological indices with the maximum edge weight to construct a three-dimensional damage state vector and determining the damage level accordingly, this invention achieves an objective quantitative assessment of crack severity that is completely independent of human experience and does not depend on lighting or color conditions, providing a reliable and reproducible structural health assessment method for automated infrastructure inspection. Attached Figure Description

[0021] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is an overall flowchart of the crack damage severity determination method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the mapping process from crack skeleton to weighted topology map according to an embodiment of the present invention; wherein, (a) is an overlay image of the original image and a binary mask; (b) is a single-pixel skeleton and node detection map; (c) is an unweighted topology map; (d) is a weighted topology map; (e) is a schematic diagram of the calculation of weighted circumferential closure degree and weighted directional dissipation entropy; (f) is a schematic diagram of the final classification judgment result based on the three-dimensional damage state vector, wherein, (f-1) is a schematic diagram of the calculation of weighted circumferential closure degree, and (f-2) is a schematic diagram of the calculation of weighted directional dissipation entropy; Figure 3 This is a schematic diagram of the adaptive local grayscale loss sampling principle in an embodiment of the present invention; wherein, (a) is a magnified view of a local part of the original image; (b) is a view of the skeleton point normal sampling window; and (c) is a diagram illustrating the loss depth formula. Figure 4 This is a schematic diagram comparing the weighted topology method of this invention with the traditional method based on equivalent area statistics. Detailed Implementation

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0024] Example 1 This embodiment provides a method for determining the severity of crack damage based on skeleton topology and grayscale coupling, including the following steps: Obtain grayscale images of cracks on the surface of infrastructure structures; Based on the grayscale image of the crack, extract the binary segmentation mask of the crack; The binary segmentation mask of the crack is skeletonized to obtain a skeleton map with a single pixel width, and the skeleton map is mapped to an undirected topological graph of the crack containing nodes and edges. Based on the grayscale attenuation information of the crack region in the crack grayscale image, each edge of the crack undirected topological graph is assigned an edge weight. Based on the edge weights, calculate the weighted circumferential closure and weighted directional dissipation entropy of the crack undirected topology graph; A three-dimensional damage state vector is constructed based on the weighted circumferential closure, the weighted directional dissipation entropy, and the maximum value of the edge weights of each edge in the crack undirected topology graph. The severity of crack damage is determined based on the three-dimensional damage state vector.

[0025] As a possible implementation method, such as Figure 1 As shown, the specific steps include: (1) Obtain parallel overhead grayscale images of cracks in asphalt pavement or bridge structures collected by existing automated inspection equipment; (2) Perform morphological preprocessing and crack binary mask extraction; (3) Perform skeleton refinement and establish crack topology map; (4) Calculate local adaptive grayscale loss modulus along the skeleton normal direction and assign weights to each edge; (5) Calculate weighted circumferential closure degree and weighted directional dissipation entropy; (6) Construct structural damage driving force potential and output crack severity level.

[0026] Step 1, as follows Figure 2 As shown, image preprocessing and crack binary mask extraction: Before performing crack entity segmentation, the original image I is preprocessed with a combination of top cap transformation and contrast-limited adaptive histogram equalization (CLAHE) to extract dark morphological residuals, suppress asphalt aggregate particle reflection and pavement texture noise, and prevent non-crack dark areas from being mistakenly segmented as damaged areas.

[0027] ; in, This represents the morphological opening operation, where SE is a linear structuring element. After this processing, the contrast between the dark areas of the crack and the background is significantly enhanced.

[0028] Subsequently, a U-Net++ semantic segmentation network with an encoder-decoder structure (the encoder backbone is ResNet-34, fine-tuned on at least 5000 road surface crack images after loading ImageNet pre-trained weights) is used to extract crack probability maps. The probability maps are then thinned at the edges using a Conditional Random Field (CRF) and binarized with a confidence threshold of 0.5 to output a binary crack segmentation mask. This network architecture and training parameters are fixed to ensure the reproducibility of the segmentation results.

[0029] Step 2, Crack skeleton refinement and topology mapping: Morphological skeletonization (Zhang-Suen thinning algorithm) is performed on the mask to obtain a skeleton map with a single pixel width. The skeleton coordinate set is obtained. Simultaneously, the original grayscale value corresponding to each pixel on the skeleton map is recorded. And the rate of change of grayscale gradient at that point perpendicular to the skeleton direction (i.e., the steepness of the crack edge, used to remove shallow scratches).

[0030] skeleton Transform into an undirected topological graph Where V is the crack intersection point ( ) and endpoints ( E is the crack branch edge.

[0031] Step 3, Calculation of equivalent grayscale loss modulus – Adaptive weighting of crack edge weights: This step utilizes the grayscale attenuation information of the crack region in the original grayscale image I to apply grayscale values ​​to each edge of the topological graph G. A physical weight is assigned, which quantifies the equivalent opening displacement of the crack branch, i.e., the integral effect of the crack's depth and width in the direction perpendicular to the road surface.

[0032] (1) Local background sampling (e.g.) Figure 3 As shown): For any pixel on the skeleton Calculate the local gradient direction at that point. (That is, using the Sobel operator to find the tangent direction of the skeleton point); along the normal direction ( Extend L pixels to each side (e.g., L=15 if resolution is 1mm / pixel), but do not directly take the endpoints. Remove points less than 3 pixels from the crack edge (i.e., the binary mask boundary), and take the average grayscale value in the remaining clean background area. and standard deviation .

[0033] (2) Calculate the depth loss ratio: in, This represents the original grayscale value at the center of the crack. It is a minimal constant to prevent division by zero. This ratio uses local standard deviation for adaptive scaling, making the loss depth of high-texture pavements (such as rough asphalt surfaces) comparable to that of low-texture pavements (such as smooth cement surfaces), fundamentally counteracting changes in light intensity and differences in pavement materials.

[0034] (3) Anti-interference processing: Perform large-scale morphological closing operation on the original grayscale image I (structural element radius R = 30 pixels) to obtain a virtual background image. Closure operations can effectively fill all tiny cracks, simulating a non-destructive road surface. To address potential issues caused by uneven lighting, two-dimensional polynomial surface fitting is used. Spatial trend (removing global brightness variations), extracting residual field .at this time, Only localized crack shadows are included, completely eliminating vignetting in the image corners or the effects of oblique sunlight. Finally, adaptive adjustments are made to account for potential errors caused by material textures, reducing the local standard deviation of the original image. The standard texture base of this material established offline. Normalize: ; For severely cracked areas, the closing operation smooths out some of the crack shadows at the intersections of dense cracks, resulting in a virtual background. Slightly below the true lossless background. This invention introduces a local standard deviation. Adaptive scaling compensation: When dense cracks cause the background area to darken. Synchronous reduction, making The denominator automatically shifts to By converging, the distortion of the loss ratio caused by background underestimation is suppressed. This compensation strategy does not require modification of the closing operation itself, and can statistically ensure that the gray loss modulus of the severely cracked region and the sparsely cracked region remain comparable.

[0035] in, The offline calibration procedure is as follows: Under a fixed lighting box with a D65 standard light source and 45° geometric conditions, acquire undamaged, clean, and dry road surface images. For the same type of road surface mixture, calculate the 3×3 neighborhood grayscale standard deviation for each pixel of the image, statistically analyze the standard deviation histogram of the entire image, and take the peak value (mode) of the histogram as the standard texture base of the material.

[0036] (4) Edge weight aggregation: The loss ratio of all skeleton points traversed by edge e is accumulated along the path, and the equivalent gray-level loss modulus of the edge is defined. ,in The physical step size for a single pixel (determined by camera calibration). The wider and deeper the crack, the heavier the shadow and the lower the grayscale value. The weighted weights of each edge (skeleton branch) are then assigned to the final edge weights. .

[0037] It not only reflects the visual length of the crack in the image, but also directly represents the actual volume opening of the crack branch. A wide and deep crack has a very high weight, while a shallow scratch has a weight close to 1, realizing a quantitative mapping between grayscale perception and mechanical damage depth.

[0038] Step 4, as follows Figure 4 As shown, the weighted topological dissipative structure is calculated as follows: In the topology diagram Based on this, edge weights are introduced. Calculate two topological invariants that have clear solid mechanics significance.

[0039] (1) Weighted Loop Closure Index (WLC): Define the cyclic space dimension (i.e., Betti number) of graph G. ; when When the crack network is in a tree structure (without a closed mesh); when At this time, the existence of a closed loop indicates that the crack tip has a meandering intersection, forming an independent block.

[0040] Introduce physical weights and traverse all basic cycles. Define the weighted cyclic closure: ,in For ring The number of edges contained within. To compensate for the loss of large-cycle scale information due to averaging, the total weight of the largest cycle is also calculated: ,Will Used as auxiliary features in subsequent grading. If WLCs are similar, then take... Larger ones are judged as more serious structural subsidence (larger ring-cut blocks are larger and pose a greater risk).

[0041] when The structure is in a state of "unconstrained fracture", with cracks propagating freely in a single direction. The corresponding mechanical model is Type I opening fracture in linear elastic fracture mechanics. when Gradually increasing: The number of annular closed areas is constantly increasing, indicating that the road surface layer has been cut into several independent blocks, and the interlocking friction between aggregates is gradually lost.

[0042] (2) Weighted directional dissipation entropy: The angle interval [0°, 180°) is uniformly divided into K = 18 histogram intervals (each interval being 10°). For each edge e ∈ E, its geometric tilt angle is calculated. [0°, 180°), and set the edge weights. As a voting weight, it is accumulated to the corresponding interval k: , ; Define weighted directional dissipative entropy ; Low-entropy state ( ): Cracks with consistent orientation (such as transverse / longitudinal through cracks) indicate that the structure is under stable uniaxial tensile stress (Mode I). Although there is obvious cracking, the stress mechanism is clear and the structural stiffness matrix is ​​stable. High-entropy state ( The uniform dispersion of cracks at all angles (such as network cracking) indicates that the structure is subjected to complex multidirectional mixed shear stress (Mode II + Mode III), with topological disorder competition and scattering interference occurring between microcracks. This entropy value essentially reveals the non-equilibrium dissipative structure inside the material and is a precursor signal of fatigue aging and structural damage.

[0043] Step 5: Construction and grading of structural damage driving forces: Construct a three-dimensional damage state vector Φ to fully characterize the structural degradation state of the current pavement block: ; in, The maximum equivalent grayscale loss modulus in the entire graph represents the ultimate opening displacement of the weakest link in the crack network.

[0044] By coupling the three sub-indices through a weighted power exponent, the final scalar damage criterion is obtained: .

[0045] Feasible . The physical meaning is defined as the generalized topological energy dissipation rate per unit area of ​​the asphalt pavement surface layer.

[0046] The hierarchical decision-making criteria in this embodiment are as follows: Calibration threshold , , It was obtained through pre-calibration of indoor fracture tests on standard specimens.

[0047] Pre-calibrated first threshold Second threshold and the third threshold Obtained through the following methods: Fracture tests were conducted on standard specimens, and images of the entire process from microcrack initiation to macroscopic fracture were acquired simultaneously. The scalar damage criterion value obtained by processing the image corresponding to the end point of the initial linear segment during the test was calibrated as the first threshold. The scalar damage criterion value obtained by processing the image corresponding to the peak load moment during the test was calibrated as the second threshold. The scalar damage criterion value obtained by processing the image corresponding to the moment when the residual strength drops to a predetermined percentage threshold of the peak load during the test was calibrated as the third threshold.

[0048] Specifically, firstly, a three-point bending fracture test is conducted on an MTS electro-hydraulic servo testing machine, simultaneously using an industrial camera to continuously acquire images of the entire process from microcrack initiation, stable propagation, peak load to macroscopic fracture. The critical point images are then input into steps S1-S4 of this invention to calculate the corresponding index values. Thus, the basic threshold anchor point is obtained. , , The end point of the initial linear segment in the load-displacement curve of the specimen ( The crack network image corresponding to the peak load moment of the specimen is defined as the critical point of structural bearing capacity loss. The point at which the residual strength drops to 20% of its peak value is defined as the critical point of structural collapse. ).

[0049] The actual threshold value on site is calculated by multiplying the benchmark value by the on-site environmental adjustment factor. get: , , ,in, The standard deviation is determined by the ratio of the local texture standard deviation of the image to the standard deviation of the laboratory standard specimen. , This represents the average local standard deviation of the entire image of the scene. If the road surface itself is extremely rough, Significantly higher than , Therefore, the tolerance for grayscale loss is allowed to be increased proportionally to prevent coarse structures from being misreported as deep cracks.

[0050] The grading criteria are shown in Table 1 below: Table 1 In practical implementation, this method is adaptive to the resolution of the original image. For asphalt pavement, the acquired top-view grayscale detection image preferably uses an image block resolution of 1 mm / pixel; for bridge concrete pavement, a higher resolution of 0.1 mm to 0.2 mm / pixel is used to capture fine crack widths, in which case the original image is usually a color image. Under different resolution conditions, the pixel physical step size in this method is determined accordingly by the camera calibration results to ensure that the physical scale conversion involved in each step remains consistent.

[0051] Regarding the determination of the grading threshold, this embodiment uses a standard three-point bending fatigue specimen and conducts graded loading tests on an MTS electro-hydraulic servo testing machine, simultaneously and continuously acquiring surface images of the specimen. Using the topological parameters of the crack network at actual specimen failure as anchor points, the discrimination thresholds corresponding to each severity level are determined in reverse, thereby ensuring a traceable correspondence between the grading criteria and the actual mechanical failure behavior of the material.

[0052] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for determining the severity of crack damage based on skeleton topology and grayscale coupling, characterized in that, Includes the following steps: Obtain grayscale images of cracks on the surface of infrastructure structures; Based on the grayscale image of the crack, extract the binary segmentation mask of the crack; The binary segmentation mask of the crack is skeletonized to obtain a skeleton map with a single pixel width, and the skeleton map is mapped to an undirected topological graph of the crack containing nodes and edges. Based on the grayscale attenuation information of the crack region in the crack grayscale image, each edge of the crack undirected topological graph is assigned an edge weight. Based on the edge weights, calculate the weighted circumferential closure and weighted directional dissipation entropy of the crack undirected topology graph; A three-dimensional damage state vector is constructed based on the weighted circumferential closure, the weighted directional dissipation entropy, and the maximum value of the edge weights of each edge in the crack undirected topology graph. The severity of crack damage is determined based on the three-dimensional damage state vector.

2. The method according to claim 1, characterized in that, Based on the grayscale image of the crack, extract the binary segmentation mask of the crack, including: A combination of top-hat transform and contrast-limited adaptive histogram equalization is performed on the crack grayscale image to obtain an enhanced crack grayscale image. The enhanced crack grayscale image is input into a pre-trained semantic segmentation network to extract a crack probability map; After edge refinement of the crack probability map, it is binarized to output a crack binary segmentation mask.

3. The method according to claim 1, characterized in that, The binary segmentation mask of the crack is skeletonized to obtain a skeleton map with a single pixel width, and the skeleton map is mapped to an undirected topological graph of the crack containing nodes and edges, including: The morphological skeletonization algorithm is used to refine the binary segmentation mask of the crack to obtain a skeleton map with a single pixel width, and the original gray value of each pixel on the skeleton map and the gray gradient change rate of each pixel in the direction perpendicular to the skeleton are recorded. Using the crack intersections in the skeleton graph as nodes and the crack branches as edges, an undirected crack topology graph is constructed. The crack intersections are pixels with three or more connecting branches in their neighborhood, and the crack endpoints are pixels with one connecting branch in their neighborhood.

4. The method according to claim 1, characterized in that, Based on the grayscale attenuation information of the crack region in the crack grayscale image, each edge of the undirected topological graph of the crack is assigned an edge weight, including: For any pixel on the skeleton map, calculate the local gradient direction at the pixel, and extend a predetermined pixel distance to both sides along the normal direction of the local gradient direction to perform local background sampling. Remove sampling points that are less than a predetermined pixel threshold from the edge of the crack, and obtain the average gray value and standard deviation of the background gray value from the remaining sampling points. Based on the difference between the average grayscale value of the background and the original grayscale value of any pixel on the skeleton map, adaptive scaling is performed in conjunction with the standard deviation of the background grayscale to obtain the adaptive grayscale loss ratio of any pixel. The adaptive grayscale loss ratio of all pixels contained in each edge of the undirected topology graph of the crack is accumulated along the path to obtain the equivalent grayscale loss modulus of each edge. The equivalent grayscale loss modulus is normalized and mapped to obtain the edge weight of each edge.

5. The method according to claim 4, characterized in that, Before obtaining the adaptive grayscale loss ratio of any pixel based on the difference between the average grayscale value of the background and the original grayscale value of any pixel on the skeleton map, combined with the standard deviation of the background grayscale, adaptive scaling is performed, the process further includes: Perform a large-scale morphological closing operation on the crack grayscale image to obtain a virtual background image; The virtual background image is fitted with a two-dimensional polynomial surface, and the residual field is extracted to remove global illumination variations. Based on the local grayscale standard deviation of the crack grayscale image and the pre-calibrated material standard texture base, the denominator of the adaptive grayscale loss ratio is normalized and compensated. The material standard texture base is pre-calibrated in the following way: a standard image of the undamaged road surface is acquired under a lighting environment with standard light source and fixed geometric conditions, the standard deviation of the neighborhood gray level is calculated pixel by pixel, the standard deviation histogram of the whole image is statistically analyzed, and the peak value is taken as the material standard texture base.

6. The method according to claim 1, characterized in that, Based on the edge weights, the weighted cyclic closure of the undirected topological graph of the crack is calculated, including: Calculate the cyclospace dimension of the undirected topological graph of the crack based on the number of edges and nodes. When the dimension of the circle space is greater than or equal to one, all basic cycles in the undirected topology graph of the crack are traversed. For each basic cycle, the sum of the edge weights of all edges in each basic cycle is divided by the number of edges contained in each basic cycle to obtain the weighted average weight of each basic cycle. The weighted average weights of all basic loops are summed to obtain the weighted loop closure. At the same time, traverse all basic cycles and take the maximum value among the sums of the edge weights of each basic cycle as the total weight of the largest cycle.

7. The method according to claim 1, characterized in that, Based on the edge weights, the weighted directional dissipation entropy of the undirected topological graph of the crack is calculated, including: Divide the angle interval into a predetermined number of histogram intervals; For each edge in the undirected topology graph of the crack, calculate the geometric tilt angle of each edge, and add the edge weight of each edge as a voting weight to the histogram interval corresponding to the geometric tilt angle to obtain the weighted cumulative value of each histogram interval. Calculate the probability distribution based on the proportion of the weighted cumulative value of each histogram interval to the total weighted cumulative value of all histogram intervals; The information entropy is calculated based on the probability distribution to obtain the weighted directional dissipation entropy.

8. The method according to claim 1, characterized in that, The three-dimensional damage state vector includes the weighted circumferential closure, the weighted directional dissipation entropy, and the maximum value of the edge weights of each edge in the crack undirected topology graph; Based on the three-dimensional damage state vector, the severity of crack damage is determined, including: The three components of the three-dimensional damage state vector are coupled by a weighted power exponent to obtain a scalar damage criterion; The scalar damage criterion is compared with a pre-calibrated first threshold, second threshold, and third threshold, respectively. Based on the comparison results, determine the degradation phase to which the severity of crack damage belongs.

9. The method according to claim 8, characterized in that, The pre-defined first, second, and third thresholds are obtained in the following way: Fracture tests were conducted on standard specimens, and images of the entire process from microcrack initiation to macroscopic fracture were simultaneously acquired. The scalar damage criterion value obtained after processing the image corresponding to the end point of the initial linear segment during the experiment is calibrated as the first threshold. The scalar damage criterion value obtained by processing the image corresponding to the peak load time during the test is calibrated as the second threshold. The scalar damage criterion value obtained by processing the image corresponding to the moment when the residual strength drops to a predetermined percentage threshold of the peak load during the test is calibrated as the third threshold.

10. The method according to claim 8, characterized in that, Based on the comparison results, the degradation phase to which the severity of crack damage belongs is determined, including: When the weighted circumferential closure is equal to zero and the scalar damage criterion is less than the first threshold, it is determined to be an intact micro-phase; When the weighted circumferential closure is equal to zero and the scalar damage criterion is greater than or equal to the second threshold, it is determined to be a linear open fracture phase; When the weighted circumferential closure is greater than or equal to one and the scalar damage criterion is greater than or equal to the third threshold, it is determined to be a network dissipative structure phase; When the weighted circumferential closure is greater than or equal to one, the weighted directional dissipation entropy approaches its maximum value, and the maximum edge weight increases significantly, it is determined to be a chaotic structure collapse phase.

Images