Road crack change detection method and system based on graph structure

Abstract

The application provides a kind of road crack change detection method and system based on graph structure, including preliminary extraction obtains only crack and background label binary graph;Crack content in binary graph is expressed using graph structure, including according to crack skeleton network connection feature point, the feature point set of crack and the curve segment set of crack are obtained, and the connected adjacency matrix is used to record the connected relationship of feature point;Based on graph structure, using Monte Carlo tree search method, the graph structure curve segment matching of different periods is completed;Based on the end point of matched curve segment, the rotation matrix is calculated, and the splicing alignment of different period images is realized;Based on the graph structure curve segment matching result and image splicing alignment result, road crack change detection analysis is carried out and output. The detection scheme has strong implementability in engineering project, and has high stability and sufficient precision.

Description

Technical Field

[0001] This invention relates to the field of road crack detection technology, and specifically to a road crack change detection technology based on graph structure. Background Technology

[0002] Roads, bridges, tunnels, and other transportation infrastructure are crucial components connecting different regions and promoting social development. However, in actual use, the surfaces of these infrastructures are affected by various factors such as structural aging, traffic loads, temperature changes, and extreme weather, leading to cracks. The presence of cracks reduces the performance and safety of the infrastructure. If effective countermeasures are not taken in a timely manner, cracks will continue to worsen, even leading to serious defects such as potholes and collapses, increasing maintenance difficulties and costs, and jeopardizing driving safety and efficiency. Therefore, monitoring the changes and development of cracks at different stages and promptly identifying potential crack deterioration trends is of great significance for the scientific maintenance of roads.

[0003] Due to variations in factors such as the age of road sections, geographical environment, climate conditions, pavement materials, traffic volume, and maintenance history, the trends and rates of crack deterioration also differ. Employing appropriate maintenance methods for cracks with different development trends, conducting inspections at varying frequencies for road sections with different deterioration rates, and carrying out timely repairs can significantly reduce maintenance costs. Therefore, an effective method for detecting crack changes is needed to provide a scientific basis for road maintenance.

[0004] Crack change detection is a challenging task, requiring the resolution of two main issues: geometric aberration and topological change. Geometric aberration refers to changes in scale, rotation, and translation between image data from different periods due to factors such as shooting angle, distance, and lighting. Topological change refers to changes in cracks between image data from different periods due to crack development, such as the increase, decrease, bifurcation, and merging of cracks. Because image data is inexpensive, readily available, and visually intuitive, change detection tasks typically use image data. However, unlike change detection tasks using general remote sensing imagery, cracks lack obvious feature points such as corners, making it difficult to correct for geometric and topological aberrations through feature matching. Furthermore, the presence of geometric and topological aberrations makes pixel-level or object-level matching impossible, rendering pixel-based and object-based change detection ineffective for crack change detection tasks.

[0005] To address the problem of crack change detection, this invention proposes a road crack change detection method based on graph structures. A graph structure is a data structure composed of nodes and edges, which can describe the topological relationships and attribute information between objects. The attributes of the edges in a graph structure can describe the relative positional relationships between feature points. The nodes in a graph structure, unlike geometric feature points such as corner points, are feature points in the crack structure that describe the topological adjacency information between crack curve segments. They are easy to find, and the adjacency information they describe is unaffected by geometric and topological deviations, exhibiting invariance. A crack structure can be viewed as a network structure composed of multiple crack curves, with multiple branching points, endpoints, and inflection points, which shares certain similarities with graph structures. This invention utilizes graph structures to abstractly describe cracks and solves the problem of road crack change detection based on graph structures. Summary of the Invention

[0006] To address the shortcomings of the existing technologies, this invention proposes a road crack change detection method based on graph structures.

[0007] To achieve the above objectives, this invention proposes a road crack change detection method based on graph structure, comprising the following processing steps.

[0008] The initial extraction yielded a binarized image containing only cracks and background labels;

[0009] The graph structure is used to express the crack content in the binary graph, including connecting feature points according to the crack skeleton network to obtain the set of feature points and the set of crack segments, and using a connected adjacency matrix to record the connectivity of feature points.

[0010] Based on the graph structure, the Monte Carlo tree search method is used to complete the matching of graph structure segments at different periods;

[0011] Based on the endpoints of the matching curve segments, the rotation matrix is ​​calculated to achieve the stitching and alignment of images from different periods;

[0012] Based on the graph structure curve segment matching results and image stitching alignment results, road crack change detection and analysis are performed and output.

[0013] Furthermore, the FPCN deep learning network framework was used to extract cracks in road images, resulting in a binarized image containing only crack and background labels.

[0014] Furthermore, the method for expressing the crack content within a binary graph using a graph structure is as follows:

[0015] First, a morphological thinning algorithm is used to iteratively remove boundary pixels in the crack region that do not affect the continuity of the crack until a connected skeleton with a single pixel width is obtained. The original crack skeleton network is decomposed into non-intersecting skeleton curves. Then, the inflection points of the crack skeleton curves are extracted based on the curvature scale space.

[0016] Then, based on the crack skeleton network, the feature points are connected to obtain the set of crack feature points and the set of crack segments, and the connectivity relationship of the feature points is recorded using a connected adjacency matrix. Furthermore, based on the crack skeleton segments, the crack length and direction are calculated in units of segments to obtain the geometric properties of the edges of the crack graph structure.

[0017] Moreover, the extraction of inflection points of the crack skeleton curve based on curvature scale space is carried out in two stages. The first stage is based on curvature scale space. At a large scale, the crack curve is convolved with a Gaussian convolution kernel to smooth the curve and then the crack curvature is calculated. Points exceeding a set threshold are taken as significant inflection points found at the large scale. The second stage, based on the first stage, uses the geometric-Douglaspus algorithm to find inflection points in detail as a supplement to the inflection points at the large scale.

[0018] Furthermore, the matching of graph structure segments at different times is implemented as follows:

[0019] For existing crack segments, iteratively merge them with adjacent segments that have the same extension trend;

[0020] Calculate the relative length difference of curve segments and select the curve segments with smaller relative length differences as the initial pre-matching curve segments;

[0021] Based on three principles—that a new matching curve pair must be adjacent to at least one matching curve pair in the set of matching curve pairs, that a newly selected matching curve pair cannot be repeated with an already matched curve or its sub-edges, and that the relative positional relationship between the matching curve pair and the matching curve pairs in the set of matching curve pairs is consistent—a Monte Carlo tree search strategy is used to incrementally construct a Monte Carlo search tree. Each tree node corresponds to a matching method, and a heuristic search method is used to extract the globally optimal matching result.

[0022] Furthermore, the road crack change detection and analysis is implemented as follows:

[0023] Based on the matched curve segment, the coordinates of both ends are obtained. A buffer of a certain width is constructed based on the line connecting the two ends. Let the binarized image of the pre-crack phase have the following... Each pixel falls into the buffer. For the binarized image after the crack, it has... Each pixel falls into the buffer, and the crack widening rate is obtained. When R N Exceeding the corresponding threshold ε R At that time, it is considered that the crack has widened;

[0024] Suppose we obtain the connectivity matrix A in the later stage of the crack. B Feature point set VB Calculate the feature point set V B Number of segments connected by each point in the middle Each segment in the set of unmatched segments is analyzed to determine whether its two ends coincide with the endpoints of existing crack segments. If coincidence occurs, and To determine if this curve segment is an extended curve segment, we use the connectivity matrix A. B Mark connected curve segments; if they overlap, and Determine if this segment is a bifurcated segment, and mark the connected segments as bifurcated segments.

[0025] On the other hand, the present invention provides a road crack change detection system based on graph structure, for implementing the road crack change detection method based on graph structure as described above.

[0026] Moreover, it includes the following modules,

[0027] The first module is used to initially extract a binarized image containing only cracks and background labels;

[0028] The second module is used to express the crack content in the binary graph using a graph structure, including connecting feature points according to the crack skeleton network to obtain the set of feature points and the set of crack segments, and using a connected adjacency matrix to record the connectivity relationship of feature points.

[0029] The third module is used to match graph structure segments at different times using the Monte Carlo tree search method based on the graph structure.

[0030] The fourth module is used to calculate the rotation matrix based on the endpoints of the matching curve segments, so as to realize the stitching and alignment of images from different periods.

[0031] The fifth module is used to perform road crack change detection and analysis based on the graph structure curve segment matching results and image stitching alignment results, and then output the results.

[0032] Alternatively, it may include a processor and a memory, with the memory used to store program instructions and the processor used to call the stored instructions in the memory to execute a graph-based road crack change detection method as described above.

[0033] Alternatively, it may include a readable storage medium storing a computer program that, when executed, implements a graph-based road crack change detection method as described above.

[0034] Compared with the prior art, the above-described technical solutions conceived in this invention can achieve the following beneficial effects:

[0035] This paper presents an in-depth study and analysis of the challenges in detecting road crack changes, and proposes a graph-based road crack change detection scheme. This scheme focuses on the continuity characteristics of cracks, using a deep convolutional neural network for semantic segmentation of crack images. It then extracts crack feature points and quantifies curve segment attributes to achieve information conversion from image to graph representation. A Monte Carlo tree search method is used to match crack graph structures from different periods from a global perspective. Based on the matching results, the type and location of crack changes are determined through graph structure alignment and fusion. This detection scheme demonstrates strong feasibility in engineering projects, along with high stability and sufficient accuracy.

[0036] The present invention is simple and convenient to implement, highly practical, and solves the problems of low practicality and inconvenience in actual application of related technologies. It can improve user experience and has significant market value. Attached Figure Description

[0037] Figure 1 This is a flowchart of a method according to an embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of the FPCN deep learning network framework according to an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the crack skeleton and feature points in an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of the Monte Carlo tree search strategy according to an embodiment of the present invention.

[0041] Figure 5 The diagram shows an example of the application effect of this invention. Detailed Implementation

[0042] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0043] like Figure 1 As shown in the figure, an embodiment of the present invention proposes a road crack change detection method based on graph structure, which includes the following steps:

[0044] (1) A binarized image containing only cracks and background labels is initially extracted. In the preferred embodiment, the FPCN deep learning network framework is used to extract cracks in the road image to obtain a binarized image containing only cracks and background labels.

[0045] This invention matches road images from different periods. In the embodiment, road images from period A (early stage of cracking) and period B (late stage of cracking) are matched. The road images from these two periods are preprocessed (steps 1 and 2) before matching.

[0046] Preferably, step (1) of the embodiment is implemented in the following ways:

[0047] (1.1) Using the publicly available road crack dataset GAPs, a trained crack semantic segmentation network model is obtained by training on the FPCN deep learning network framework; the FPCN deep learning network framework can adopt existing technologies, but preferably adopts the crack segmentation network proposed by the inventors' team, whose encoder and decoder architecture is as follows: Figure 2 As shown: Specifically, the encoder contains four modules combining convolutions and pooling. Each module first performs two 3×3 convolutions and ReLU non-linear activations (Convs), then 2×2 max pooling downsampling (indicated by red arrows), and finally uses a spatial attention module (SAM) at the top layer of feature extraction. The network's decoder uses an upsampling unit (SEU) with a stride of 2 to progressively restore the image resolution. One of its inputs is the feature extraction data obtained through multiple dilated convolutions (MDF). Simultaneously, the encoder and decoder layers are connected using multiple convolutional feature extraction (MCF), which, along with the MDF, serves as input to the SEU for processing.

[0048] (1.2) Use the trained model to complete the semantic segmentation task of the road crack image of the task target, and obtain a binarized image containing only crack and background labels.

[0049] (2) Use graph structures to represent the crack content within a binary graph;

[0050] Preferably, step (2) of the embodiment is implemented in the following ways:

[0051] (2.1) A morphological thinning algorithm is used to iteratively remove boundary pixels in the crack region that do not affect the continuity of the crack until a connected skeleton with a width of one pixel is obtained; the specific implementation of the morphological thinning algorithm is the prior art, and will not be described in detail in this invention.

[0052] (2.2) The original crack skeleton network is decomposed into non-intersecting skeleton curves. Then, based on the curvature scale space, the inflection points of the crack skeleton curves are extracted at a large scale. See [link to relevant documentation] Figure 3 Feature points include inflection points, endpoints, and intersections. Endpoints are the two ends of a crack, intersections are the points where multiple cracks intersect, and inflection points are intermediate points added to provide a detailed description of the cracks.

[0053] In specific implementation, the present invention further proposes a preferred method for extracting inflection points as follows:

[0054] Because the crack skeleton curve is noisy, directly extracting inflection points at a small scale is easily affected by noise, resulting in an excessive number of inflection points. However, the features of inflection points gradually disappear as the scale increases. Therefore, extracting inflection points only at a large scale leads to a lack of detail and inaccurate mapping. To address this, the embodiment divides crack inflection point extraction into two stages. The first stage, based on curvature scale space, uses a Gaussian convolution kernel to smooth the crack curve at a large scale, then calculates the crack curvature. Points exceeding a set threshold are identified as significant inflection points found at the large scale. The second stage, building on the subdivision in the first stage, uses the geometric-Douglaspochon algorithm to find inflection points in detail, supplementing the large-scale inflection points.

[0055] (2.3) Based on the crack skeleton network, the feature point set V and the crack segment set S are obtained, and the connectivity relationship of the feature points is recorded using the connected adjacency matrix A. Based on the crack skeleton segments, the crack length and direction are calculated on a segment-by-segment basis to obtain the geometric properties of the edges of the crack graph structure. The segment length L is represented by the geodesic distance between the two endpoints of the segment. In actual calculations, this is discretized into an operation based on the adjacency relationship of the segment pixels, as shown in Formula 1. Here, n represents the number of pixels in the segment, and d(i,i+1) represents the Euclidean distance between the i-th pixel and the (i+1)-th pixel. If the (i+1)-th pixel is within the four neighbors of i, the distance is 1; otherwise, the distance is 0. The direction θ of the curve segment is represented by the angle between the line connecting the two ends of the curve segment and the horizontal straight line, as shown in Formula 2, where Δx and Δy represent the interval distance between the two ends of the curve segment in the x and y directions, respectively.

[0056]

[0057]

[0058] (3) Based on the graph structure, the Monte Carlo tree search method is used to complete the graph structure segment matching in periods A and B;

[0059] Preferably, step (3) of the embodiment is implemented in the following ways:

[0060] (3.1) For an existing crack segment, iteratively merge it with adjacent segments that have the same extension trend, adding one segment each time, and grouping all currently merged segments into a new crack segment. For two segments that need to be regrouped, it is assumed that the difference in their directions should not exceed a constant ε. T If the condition is met, the segments are grouped into new crack segments, and the next adjacent segment is judged; otherwise, the grouping in that direction ends. Experiments have shown that the constant ε... TThe recommended value range is [10°, 30°], and the specific value can be determined according to the actual situation.

[0061] (3.2) Calculate the relative length difference of the curve segments, and select the curve segments with smaller relative length differences as the initial pre-matching curve segments. The preferred selection strategy is based on Formula 3, where l(r) is the length of the curve segment before the crack change, l(s) is the length of the curve segment after the crack change, and ε G As a proportionality constant, the constant ε is determined experimentally. G The recommended value range is [0.1, 0.2], and the specific value can be determined according to the actual situation;

[0062]

[0063] (3.3) This invention proposes that, based on three principles—that a new matching curve pair must be adjacent to at least one set of matching curve pairs in the set of matching curve pairs, that a newly selected matching curve pair cannot repeat an already matched curve or its sub-edges, and that the relative positional relationship between the matching curve pair and the matching curve pairs in the set of matching curve pairs is consistent—a Monte Carlo tree search strategy can be used to incrementally construct a Monte Carlo search tree. Each tree node corresponds to a matching method, and a heuristic search method is used to extract the globally optimal matching result, thus obtaining a set of matching points. and Matching edge set and

[0064] The Monte Carlo tree search strategy is implemented using existing technology. For ease of application of this invention, please refer to [link to existing technology]. Figure 4 The following instructions are provided: First, select the steps, starting from the root node, based on the benefits. The highest-efficiency child node is selected until an unexplored leaf node is reached. The efficiency calculation formula is as follows, where lm is the length of the matched edges and ls is the set of matched edges. The total length, k is the number of search iterations, n is the number of times the current node is visited, and γ is an adjustable parameter for balancing the two terms, usually set to 1, but the specific value can be determined according to the actual situation.

[0065]

[0066] The next step is the expansion step, which adds one or more new child nodes; followed by the simulation step, which performs a fast random simulation starting from the new nodes until the simulation ends; finally, the backtracking step, which updates the node information based on the simulation results. This process is repeated 1000 times, and then the leaf node with the most visits is selected as the optimal matching scheme.

[0067] (4) Based on the endpoints of the matching curve segments, the rotation matrix is ​​calculated using the ICP algorithm to achieve the stitching and alignment of images from periods A and B.

[0068] In practice, this step can be implemented using existing technologies. This invention preferably recommends using the ICP algorithm, which is an iterative nearest-point algorithm for point cloud matching.

[0069] Preferably, step (4) of the embodiment is proposed to be implemented in the following ways:

[0070] (4.1) For Point set and For each point set, calculate the average coordinates. By determining the distance between two points, the translation matrix between the point sets can be obtained.

[0071] (4.2) Through multiple iterations, the nearest point is continuously searched and the least squares transformation is calculated to obtain the optimal image registration transformation matrix;

[0072] (4.3) The coordinates of each pixel representing the crack in the pre-change crack binarized image are iteratively multiplied by the image registration change matrix to obtain the pre-change crack image aligned with the later crack coordinates.

[0073] (5) Based on the graph structure curve segment matching results and image stitching alignment results, extract and output the road crack change detection and analysis conclusions.

[0074] Based on the matching results, this invention determines the type and location of crack changes through the alignment and fusion of graph structures.

[0075] Preferably, step (5) of the embodiment is implemented in the following ways:

[0076] (5.1) Based on the matched segments Obtain the coordinates of its two ends, and construct a buffer of a certain width based on the line connecting the two ends. Assume that for the binarized image of the pre-crack phase, it has... Each pixel falls into the buffer. For the binarized image after the crack, it has... Each pixel falls into the buffer, and the crack widening rate R can be obtained based on Formula 1. N When R N Exceeding a certain threshold ε R At this point, the crack can be considered to have widened. ε R The recommended value range is [0.3, 1], and the specific value can be determined according to the actual situation;

[0077]

[0078] (5.2) The connectivity matrix A of the cracks in period B obtained according to step (2.3)B Feature point set V B Calculate the feature point set V B Number of segments connected by each point in the middle Each segment in the set of unmatched segments is analyzed to determine whether its two ends coincide with the endpoints of existing crack segments. If coincidence occurs, and This allows us to determine that the curve segment is an extended curve segment, based on the connectivity matrix A. B Mark connected curve segments; if they overlap, and This allows us to determine that the segment is a bifurcated segment and mark the connected segments as bifurcated segments.

[0079] See Figure 5 This invention provides a set of crack detection examples applying the method provided in this embodiment. For crack images from different periods, corresponding graph structures are constructed based on their crack skeletons, where blue represents the crack skeleton of the previous period and red represents the crack skeleton of the subsequent period. This method effectively achieves graph structure matching, yielding the final crack change detection conclusion.

[0080] In specific implementation, the method proposed in the technical solution of this invention can be automatically executed by those skilled in the art using computer software technology. System devices for implementing the method, such as computer-readable storage media storing the corresponding computer program of the technical solution of this invention and computer equipment including the computer program running the corresponding computer program, should also be within the protection scope of this invention.

[0081] In some possible embodiments, a road crack change detection system based on graph structure is provided, including the following modules: a first module for initially extracting a binary map containing only cracks and background labels;

[0082] The second module is used to express the crack content in the binary graph using a graph structure, including connecting feature points according to the crack skeleton network to obtain the set of feature points and the set of crack segments, and using a connected adjacency matrix to record the connectivity relationship of feature points.

[0083] The third module is used to match graph structure segments from different periods using the Monte Carlo tree search method based on the graph structure; the fourth module is used to calculate the rotation matrix based on the endpoints of the matched segments to achieve image stitching and alignment from different periods.

[0084] The fifth module is used to perform road crack change detection and analysis based on the graph structure curve segment matching results and image stitching alignment results, and then output the results.

[0085] In some possible embodiments, a graph-based road crack change detection system is provided, including a processor and a memory. The memory is used to store program instructions, and the processor is used to call the stored instructions in the memory to execute a graph-based road crack change detection method as described above.

[0086] In some possible embodiments, a graph-based road crack change detection system is provided, including a readable storage medium storing a computer program that, when executed, implements the graph-based road crack change detection method described above.

[0087] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A method for detecting road crack changes based on graph structure, characterized in that: The following processing steps are included: The initial extraction yielded a binarized image containing only cracks and background labels; The graph structure is used to express the crack content in the binary graph, including connecting feature points according to the crack skeleton network to obtain the set of feature points and the set of crack segments, and using a connected adjacency matrix to record the connectivity of feature points. Based on the graph structure, the Monte Carlo tree search method is used to complete the matching of graph structure segments at different periods; Based on the endpoints of the matching curve segments, the rotation matrix is ​​calculated to achieve the stitching and alignment of images from different periods; Based on the graph structure curve segment matching results and image stitching alignment results, road crack change detection and analysis are performed and output; The FPCN deep learning network framework is used to extract cracks in road images, resulting in a binarized image containing only cracks and background labels. The FPCN deep learning network framework adopts an encoder and decoder architecture. The encoder contains four modules that combine convolution and pooling. Each module first performs convolution and ReLU nonlinear activation, then performs max pooling downsampling, and finally uses the spatial attention module SAM at the top layer of feature extraction. The decoder uses the upsampling module SEU to gradually restore the resolution of the image. One of its inputs is the feature extraction data obtained by processing the MDF through multiple dilated convolutions. Meanwhile, the encoding layer and the decoding layer are connected by the multiple convolutional feature extraction method MCF, which, together with MDF, is used as input to SEU for processing; The method for expressing the crack content within a binary graph using a graph structure is as follows. First, a morphological thinning algorithm is used to iteratively remove boundary pixels in the crack region that do not affect the continuity of the crack until a connected skeleton with a single pixel width is obtained. The original crack skeleton network is decomposed into non-intersecting skeleton curves. Then, the inflection points of the crack skeleton curves are extracted based on the curvature scale space. Then, based on the crack skeleton network, the feature points are connected to obtain the set of crack feature points and the set of crack segments, and the connectivity relationship of the feature points is recorded using a connected adjacency matrix; furthermore, based on the crack skeleton segments, the crack length and direction are calculated in units of segments to obtain the geometric properties of the edges of the crack graph structure. The extraction of inflection points from the crack skeleton curve based on curvature scale space is divided into two stages. The first stage uses Gaussian convolution kernels to perform convolution operations on the crack curve at a large scale based on curvature scale space to smooth the curve and then calculate the crack curvature. Points exceeding a set threshold are identified as significant inflection points found at the large scale. The second stage, based on the first stage, uses the geometric-Douglaspus algorithm to find inflection points in detail as a supplement to the inflection points at the large scale. The matching of graph structure segments at different periods is implemented as follows. For existing crack segments, iteratively merge them with adjacent segments that have the same extension trend; Calculate the relative length difference of curve segments and select curve segments with small relative length differences as preliminary pre-matching curve segments; Based on three principles—that a new matching curve pair must be adjacent to at least one set of matching curve pairs in the set of matching curve pairs, that a newly selected matching curve pair cannot be repeated with a already matched curve or its sub-edges, and that the relative positional relationship between the matching curve pair and the matching curve pairs in the set of matching curve pairs is consistent—a Monte Carlo tree search strategy is used to incrementally construct a Monte Carlo search tree, with each tree node corresponding to a matching method, and a heuristic search method is used to extract the globally optimal matching result. The method for detecting and analyzing changes in road cracks is as follows. Based on the matched curve segment, the coordinates of both ends are obtained. A buffer of a certain width is constructed based on the line connecting the two ends. Let the binarized image of the pre-crack phase have the following... Each pixel falls into the buffer. For the binarized image after the crack, it has... Each pixel falls into the buffer, and the crack widening rate is obtained. ,when Exceeding the corresponding threshold At that time, it is considered that the crack has widened; Suppose we obtain the connectivity matrix in the later stage of the crack. Feature point set Calculate the set of feature points Number of segments connected by each point in the middle Each segment in the set of unmatched segments is analyzed to determine whether the two ends of the segment coincide with the endpoints of the existing crack segment. If coincidence occurs, and To determine if a curve segment is an extended curve segment, the connectivity matrix is ​​used. Mark connected curve segments; if they overlap, and Determine if this segment is a bifurcated segment, and mark connected segments as bifurcated segments.

2. A road crack change detection system based on graph structure, characterized in that: This method is used to implement the road crack change detection method based on graph structure as described in claim 1.

3. The road crack change detection system based on graph structure according to claim 2, characterized in that: Includes the following modules, The first module is used to initially extract a binarized image containing only cracks and background labels; The second module is used to express the crack content in the binary graph using a graph structure, including connecting feature points according to the crack skeleton network to obtain the set of feature points and the set of crack segments, and using a connected adjacency matrix to record the connectivity relationship of feature points. The third module is used to match graph structure segments at different times using the Monte Carlo tree search method based on the graph structure. The fourth module is used to calculate the rotation matrix based on the endpoints of the matching curve segments, so as to realize the stitching and alignment of images from different periods. The fifth module is used to perform road crack change detection and analysis based on the graph structure curve segment matching results and image stitching alignment results, and then output the results.

4. The road crack change detection system based on graph structure according to claim 2, characterized in that: It includes a processor and a memory, the memory being used to store program instructions, and the processor being used to call the stored instructions in the memory to execute the road crack change detection method based on graph structure as described in claim 1.

5. The road crack change detection system based on graph structure according to claim 2, characterized in that: The method includes a readable storage medium on which a computer program is stored, which, when executed, implements the road crack change detection method based on graph structure as described in claim 1.

Images