An asphalt pavement crack volume estimation method, device and medium

Abstract

The application discloses an asphalt pavement crack volume estimation method, device and medium, and relates to the technical field of road detection. The method comprises the following steps: collecting a high-resolution image of a road surface, segmenting the high-resolution image by using an image segmentation model to obtain a crack binary image; performing skeleton extraction on the crack binary image to obtain a structured crack skeleton graph, and obtaining a crack type and a crack area; based on the crack skeleton graph, surface geometric parameters of the crack are extracted through direction correction, orthogonal projection and edge detection, wherein the surface geometric parameters comprise a crack length, a crack width and a shape factor; and the surface geometric parameters, the crack type and the crack area are input into a pre-trained crack volume prediction model to obtain a crack volume prediction value. The application realizes the transformation from destructive local measurement to non-destructive intelligent prediction by using the above method.

Description

Technical Field

[0001] This application relates to the field of road inspection technology, and in particular to a method, equipment and medium for estimating the volume of cracks in asphalt pavement. Background Technology

[0002] Currently, the quantitative assessment of crack volume in asphalt pavement mainly relies on two technical approaches: direct measurement and indirect estimation, both of which have significant limitations. Direct measurement methods, represented by destructive core drilling, can obtain local true values, but are inefficient, damage the pavement structure, and cannot be implemented on a large scale. Non-destructive testing technologies, represented by ground-penetrating radar, are limited by complex data interpretation and low signal-to-noise ratio, making it difficult to automatically and accurately reconstruct the three-dimensional morphology of cracks and calculate their volume. Overall, direct measurement methods have bottlenecks in efficiency and scalability, while indirect non-destructive methods are severely lacking in accuracy and reliability.

[0003] To further improve detection efficiency, existing technologies attempt to acquire three-dimensional features of crack surfaces through methods such as laser scanning, and then estimate the volume based on simplified geometric assumptions or shallow statistical models. However, the mapping relationship between surface features and internal volume established by these methods heavily relies on empirical parameters and fails to incorporate the physical and mechanical mechanisms of crack development. Since the actual depth propagation of cracks is controlled by multiple complex factors such as material heterogeneity, interlayer bonding state, and environmental loads, there is a highly nonlinear relationship between its morphology and surface parameters. This results in large prediction errors and weak generalization ability in such models, making it difficult to adapt to the accurate assessment needs under different pavement structures and environmental conditions.

[0004] Based on the above analysis, the problems and shortcomings of the existing technology are as follows: The quantitative assessment of crack volume in asphalt pavement using existing technologies cannot overcome the shortcomings of low efficiency in existing direct measurement methods and insufficient accuracy and generalization of indirect extrapolation methods without damaging the pavement. Summary of the Invention

[0005] This application provides a method, device, and medium for estimating the volume of cracks in asphalt pavement, which can solve the problems of low efficiency of existing direct measurement methods and insufficient accuracy and generalization of indirect inference methods in the quantitative assessment of crack volume in asphalt pavement without damaging the pavement.

[0006] In a first aspect, embodiments of this application provide a method for estimating the volume of cracks in asphalt pavement. The method comprises: acquiring a high-resolution image of the pavement; segmenting the high-resolution image using an image segmentation model to obtain a binary image of the crack; extracting the skeleton from the binary image of the crack to obtain a structured crack skeleton map, and obtaining the crack type and crack area; based on the crack skeleton map, extracting the surface geometric parameters of the crack through orientation correction, orthogonal projection, and edge detection, the surface geometric parameters including crack length, crack width, and shape factor; and inputting the surface geometric parameters, crack type, and crack area into a pre-trained crack volume prediction model to obtain a predicted crack volume value.

[0007] In one implementation of this application, a skeleton extraction is performed on a binary image of a crack to obtain a structured crack skeleton map, and the crack type and crack area are obtained. Specifically, this includes: generating an initial crack centerline in the binary image of the crack using a morphological skeleton extraction algorithm; identifying the nodes, intersections, and continuous pixels of the initial crack centerline to construct a weighted directed graph representing the crack topology; using a branch pruning strategy to remove noise branches and repair fractures in the weighted directed graph to obtain a structured crack skeleton map; statistically analyzing the pixels representing cracks in the binary image of the crack and converting them according to pixel resolution to obtain the crack area; and identifying the crack type based on the direction and intersection features of the crack skeleton map.

[0008] In one implementation of this application, based on a crack skeleton map, surface geometric parameters of the crack are extracted through direction correction, orthogonal projection, and edge detection. Specifically, this includes: calculating the main direction of crack pixels in the crack skeleton map and correcting the direction of the crack skeleton; laying vertical search lines at preset intervals along the corrected direction of the crack skeleton; searching for edge pixels on both sides along each vertical search line and determining the edge pixels on both sides of the crack based on the gray-level abrupt change value of the crack pixels; calculating the Euclidean distance between each pair of edge pixels to obtain the crack width, and calculating the average width and width distribution based on the crack width of all search lines; performing smooth fitting on continuous pixels, calculating the length of the fitted curve, and obtaining the total length of the crack.

[0009] In one implementation of this application, edge pixel searches are performed along each vertical search line to both sides, and edge pixels on both sides of the crack are determined based on pixel grayscale abrupt change values. Specifically, this includes: scanning crack pixels from the crack skeleton to both sides along the vertical search line; on each side, the first position point where the pixel grayscale abrupt change value exceeds a preset threshold is recorded as a candidate edge pixel; calculating the grayscale gradient value of the candidate edge pixel to obtain the edge confidence; and selecting the point with the highest edge confidence among all candidate edge pixels on each side as the valid edge pixel.

[0010] In one implementation of this application, based on the crack skeleton map, the surface geometric parameters of the crack are extracted through orientation correction, orthogonal projection, and edge detection. Specifically, this includes: calculating the crack profile based on shape factors, including fractal dimension and circularity, using box counting to obtain the fractal dimension; obtaining the crack perimeter through a boundary tracking algorithm; and calculating the crack area and crack perimeter to obtain the circularity.

[0011] In one implementation of this application, the box counting method is used to calculate the crack profile to obtain the fractal dimension. Specifically, this includes: setting the grid side length covering the binary image of the crack to multiple successively decreasing scales; for each scale, counting the number of grids covering the crack profile; establishing a data point set between the logarithm of the scale and the logarithm of the corresponding number of grids; performing linear fitting on the data point set, and determining the slope of the straight line obtained by the linear fitting as the fractal dimension of the crack profile.

[0012] In one implementation of this application, before inputting the surface geometric parameters, crack type, and crack area into a pre-trained crack volume prediction model to obtain the crack volume prediction value, the method further includes: acquiring a training sample set, which includes the surface geometric parameters extracted from the sample cracks and the true volume value of the sample cracks obtained through destructive measurement, wherein the destructive measurement adopts controlled fluid injection and displacement; introducing environmental covariates corresponding to the training sample set as input features, wherein the environmental covariates include the road surface material type, the temperature at the time of collection, and regional climate information; and using the training sample set and input features to train and validate candidate machine learning models in order to select the optimal model.

[0013] In one implementation of this application, after training and validating candidate machine learning models using a training sample set to select the optimal model, the method further includes: calculating the SHAP value corresponding to each input feature for the crack volume prediction value; and generating a global feature importance ranking list based on the SHAP values.

[0014] Secondly, embodiments of this application also provide an apparatus for estimating the volume of cracks in asphalt pavement. The apparatus includes at least one processor and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform any step of an asphalt pavement crack volume estimation method.

[0015] Thirdly, embodiments of this application also provide a non-volatile computer storage medium for estimating the volume of cracks in asphalt pavement, storing computer-executable instructions configured to perform any one of the steps of an asphalt pavement crack volume estimation method.

[0016] This application provides a method, device, and medium for estimating the volume of cracks in asphalt pavement. By constructing a machine learning model that uses high-precision volume data as the training objective and integrates surface geometric parameters and environmental covariates, it can automatically and accurately predict crack volume simply by collecting pavement images. This eliminates the reliance on repetitive core sampling and achieves a leap in detection efficiency while maintaining accuracy. It overcomes the core challenge of missing mechanisms in crack volume estimation. Instead of simple data fitting, this application introduces high-fidelity volume data obtained through a destructive precision measurement fluid displacement method, as well as physical shape factors such as fractal dimension and circularity that characterize the complex morphology and expansion patterns of cracks. This enables the machine learning model to learn the deep, mechanically significant mapping relationship between surface features and internal volume, significantly improving the model's prediction accuracy and cross-scenario generalization ability. This method not only outputs predicted volume values ​​but also allows for risk classification based on the prediction results and the generation of differentiated monitoring and maintenance strategies. It achieves full-chain coverage of engineering applications from detection and diagnosis to decision support, effectively supporting the intelligent and precise maintenance of pavement. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a method for estimating the volume of cracks in asphalt pavement, provided as an embodiment of this application; Figure 2 This is a schematic diagram of the internal structure of an asphalt pavement crack volume estimation device provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] This application provides a method, device, and medium for estimating the volume of cracks in asphalt pavement, which solves the problems of low efficiency of existing direct measurement methods and insufficient accuracy and generalization of indirect estimation methods without damaging the pavement.

[0020] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0021] Figure 1 A flowchart illustrating a method for estimating the volume of cracks in asphalt pavement, provided as an embodiment of this application. Figure 1 As shown in the embodiment of this application, a method for estimating the volume of cracks in asphalt pavement specifically includes the following steps: Step 10: Acquire high-resolution images of the road surface, and use an image segmentation model to segment the high-resolution images to obtain binary images of the cracks.

[0022] In this step, a high-resolution imaging system mounted on a mobile detection platform is used to acquire digital images of the asphalt pavement to be inspected. After acquiring the high-resolution images, a pre-trained image segmentation model is used to process the images. Based on architectures such as U-Net and DeepLab, the model can accurately identify and segment all pixel regions in the image that belong to cracks. The output is a binary image, in which crack pixels are marked as foreground with a pixel value of 1 or 255, while the pavement background and other non-crack areas are marked as background with a pixel value of 0.

[0023] Step 20: Extract the skeleton from the binary image of the crack to obtain a structured crack skeleton map, and obtain the crack type and crack area.

[0024] In this step, after obtaining the binary image of the crack, it is necessary to extract the skeleton map that can characterize the morphology and topology of the crack trunk, and transform the two-dimensional crack surface region into a one-dimensional linear structure.

[0025] As an optional embodiment, the skeleton of the crack binary image is extracted to obtain a structured crack skeleton map, and the crack type and crack area are obtained. Specifically, step 201: the initial crack center line is generated in the crack binary image by means of a morphological skeleton extraction algorithm.

[0026] In this step, the Zhang-Suen thinning algorithm or the median transformation algorithm based on distance transformation can be used to process the binary image of the crack. Through iterative erosion operation, edge pixels are gradually stripped away while maintaining the connectivity of the crack, and finally a single-pixel width curve located at the center of the crack region is obtained, which is the initial crack centerline. This centerline depicts the extension path of the crack.

[0027] Step 202: Identify the nodes, intersections, and consecutive pixels of the initial crack centerline, and construct a weighted directed graph representing the crack topology.

[0028] In this step, the initial crack centerline is scanned and analyzed to identify all topological feature points, mainly including: endpoints: pixels in the centerline that are connected to only one pixel, representing the start and end points of the crack; intersections: pixels in the centerline that are connected to three or more pixels, representing the bifurcation or confluence of the crack; continuous pixels: a series of continuous pixels located between the endpoints or intersections of two feature points, representing a branch of the crack without bifurcation.

[0029] Based on these identified features, the crack centerline is abstracted as a weighted directed graph, where the vertex set consists of all endpoints, intersections, and interpolation points inserted to maintain continuity; the edge set corresponds to the continuous pixels connecting these vertices; each edge can be assigned a weight, such as the actual pixel length of the centerline segment, average width, or directional consistency.

[0030] Step 203: Use a branch pruning strategy to remove noisy branches and repair fractures in the weighted directed graph to obtain a structured crack skeleton graph.

[0031] In this step, the initially constructed graph structure often contains short, isolated spurious branches, or burrs, caused by image noise, surface texture, or minor segmentation flaws. It may also contain centerline breaks due to low local contrast. All edges can be traversed; if the length of consecutive pixels corresponding to an edge is less than a preset empirical threshold (far smaller than the typical length of a real crack branch), it is identified as a noise branch. All endpoints in the graph that are very close but not directly connected are examined. If geometric continuity conditions are met, such as near-collinear orientations, a new edge is added between them, thus repairing the skeleton breakage caused by incomplete segmentation.

[0032] Step 204: Count the pixels representing the crack in the binary image of the crack, and convert them according to the pixel resolution to obtain the crack area; In this step, pixel statistics can be directly performed on the binary image of the crack. The total number of pixels marked as cracks is counted and multiplied by the actual physical area represented by each pixel to obtain the surface area of ​​the crack.

[0033] Step 205: Identify the crack type based on the orientation and intersection features of the crack skeleton diagram.

[0034] In this step, the main direction of the crack skeleton as a whole or the dominant direction of a long continuous series of pixels is calculated and compared with the reference direction of the road, such as the direction of the lane; the number of intersections in the skeleton image is counted and the connection pattern is analyzed; a transverse crack is a crack skeleton whose main direction is roughly perpendicular to the road reference direction and has very few or no intersections; a longitudinal crack is a crack skeleton whose main direction is roughly parallel to the road reference direction and has very few or no intersections; a mesh crack is a crack in which there are multiple intersections in the skeleton image, forming a network-like or tortoise-shell-like connected structure, and the direction may not be singular.

[0035] Step 30: Based on the crack skeleton map, extract the surface geometric parameters of the crack through orientation correction, orthogonal projection and edge detection. The surface geometric parameters include crack length, crack width and shape factor.

[0036] As an optional embodiment, based on the crack skeleton map, the surface geometric parameters of the crack are extracted through orientation correction, orthogonal projection and edge detection. Specifically, it may include: Step 301: Calculate the main orientation of the crack pixels in the crack skeleton map and correct the orientation of the crack skeleton.

[0037] In this step, to eliminate the distortion of the skeleton orientation caused by the image acquisition perspective or road surface curvature, the skeleton orientation is geometrically corrected. Principal component analysis can be used to calculate the orientation of each pixel or local neighborhood on the crack skeleton map. The direction of the first principal component is the main direction of the local skeleton segment. By applying this analysis to the skeleton as a whole or in segments, a corrected direction field that better matches the physical extension trend of the crack can be obtained, replacing the original pixel connection direction.

[0038] Step 302: Along the direction of the corrected crack skeleton, lay vertical search lines at preset intervals.

[0039] In this step, along the corrected crack skeleton direction, with a fixed preset spatial interval (which can be set according to the detection accuracy requirements and must be much smaller than the crack length), measurement lines are laid on the skeleton pixels. These measurement lines are set to be strictly perpendicular to the main direction of the skeleton at the current position, thereby ensuring the ideal measurement position across the crack.

[0040] Step 303: Search for edge pixels on both sides along each vertical search line, and determine the edge pixels on both sides of the crack based on the gray-scale change value of the crack pixels.

[0041] As an optional embodiment, edge pixel searches are performed along each vertical search line to both sides, and edge pixels on both sides of the crack are determined based on pixel grayscale change values. Specifically, this may include: Step 3031: Scanning crack pixels from the crack skeleton to both sides along the vertical search line; Step 3032: On each side, the first position point where the pixel grayscale change value exceeds a preset threshold is recorded as a candidate edge pixel; Step 3033: Calculating the grayscale gradient value of the candidate edge pixel to obtain the edge confidence; Step 3034: Among all candidate edge pixels on each side, the point with the highest edge confidence is selected as the valid edge pixel.

[0042] In this step, starting from the intersection of the vertical search line and the skeleton (the preset measurement center point), a pixel-by-pixel scan is performed along the search line to both sides (the left and right sides perpendicular to the skeleton's extension direction), traversing the pixel grayscale values ​​along the path. During the scan on each side, the grayscale difference between the current pixel and its adjacent background pixels is calculated in real time. When the difference value, i.e., the intensity of the grayscale change, first exceeds a preset threshold adaptively set based on global image contrast and noise levels, it is determined that a significant transition from the crack area to the road surface background has occurred at that location, and this pixel is recorded as a candidate edge pixel for that side. To evaluate the reliability of the candidate edge pixels, the image grayscale gradient value at each candidate point is calculated. A gradient operator is used to calculate the gradient magnitude in the neighborhood of the candidate point. The resulting gradient magnitude reflects the sharpness of the edge and is defined as the edge confidence of the candidate point. The higher the confidence, the clearer and more defined the boundary. Among all the candidate edge pixels obtained from the scan on each side, sometimes multiple points may exceed the threshold due to texture interference. The pixel with the highest edge confidence is selected and finally determined as the valid edge pixel of the crack on that side.

[0043] Step 304: Calculate the Euclidean distance for each pair of edge pixels to obtain the crack width, and calculate the average width and width distribution based on the crack widths of all search lines.

[0044] In this step, the Euclidean distance between two points is calculated, representing the crack width measured at the current search line position. This width characterizes the opening size of the crack at that cross-section. All vertical search lines are traversed, and all calculated crack width values ​​are compiled into a width sequence. The arithmetic mean is calculated as the average width of the crack, used to describe the overall opening scale of the crack. Simultaneously, the statistical characteristics of this width sequence can be analyzed, such as generating a width distribution histogram, calculating the standard deviation, maximum / minimum values, etc., to describe the uniformity, variability, and extreme cases of crack opening along its length. These distribution characteristics are inputs for predicting the volume of irregular cracks.

[0045] Step 305: Fit the continuous pixels to calculate the total length of the crack, perform smooth fitting on the continuous pixels, calculate the length of the fitted curve, and obtain the total length of the crack.

[0046] In this step, based on the crack skeleton map, all edges, i.e., continuous pixels, are extracted from the image. For each continuous skeleton segment represented by an edge, curve fitting is performed using pixel coordinates, either polynomial fitting or spline fitting. Then, the length of the fitted curve is calculated. The fitted lengths of all continuous pixels are accumulated, and the units are converted using the spatial resolution of the image to obtain the total length. This reduces the error caused by pixel discretization and is more accurate than directly accumulating pixels.

[0047] As an optional embodiment, based on the crack skeleton map, the surface geometric parameters of the crack are extracted through orientation correction, orthogonal projection and edge detection, which may specifically include: Step 306: Based on shape factors including fractal dimension and circularity.

[0048] In this step, fractal dimension is used to quantify the irregularity and complexity of the crack edge profile; circularity is used to characterize the compactness of the crack region shape and distinguish between linear cracks and planar or network cracks.

[0049] Step 307: Calculate the crack profile using the box counting method to obtain the fractal dimension; As an optional embodiment, the box counting method is used to calculate the crack profile to obtain the fractal dimension, which may specifically include: Step 3071: Set the grid side length covering the binary image of the crack to multiple successively decreasing scales; Step 3072: For each scale, count the number of grids covering the crack profile; Step 3073: Establish a data point set between the logarithm of the scale and the logarithm of the corresponding number of grids; Step 3074: Perform linear fitting on the data point set, and determine the slope of the straight line obtained by linear fitting as the fractal dimension of the crack profile.

[0050] In this step, a series of square grids with progressively decreasing side lengths are defined, sufficient to cover the entire binary image of the crack, with the minimum value approaching the pixel size. These grids of different scales are used to sequentially cover the binary image containing the crack outline. For each specific grid scale, the number of grids containing at least one crack outline pixel under that scale's grid division is counted. A crack outline pixel is a pixel located at the boundary between the crack region and the background region. For all the scales used and the corresponding number of grids containing crack outline pixels, the logarithm is calculated to form a table containing... nA set of data points; the set of data points is linearly fitted in a double logarithmic coordinate system, and the slope of the fitted line is the fractal dimension of the crack profile, which is a unitless scalar between 1 and 2. 1 represents a simple smooth curve, and 2 represents an extremely complex curve that almost fills the plane. The higher the fractal dimension, the more tortuous and irregular the crack edge is, which usually means a larger specific surface area and a more complex internal spatial morphology.

[0051] Step 308: Obtain the crack perimeter using the boundary tracking algorithm, calculate the crack area and crack perimeter, and obtain the circularity.

[0052] In this step, circularity is calculated based on the relationship between the crack's area and perimeter: a boundary tracking algorithm is used to traverse the binary image of the crack, accurately extracting the entire outer contour of the crack region, and calculating the total length of the pixel paths contained within this contour. This length is then multiplied by the actual pixel size to obtain the crack's perimeter. Using the crack's surface area and perimeter, the circularity C is calculated: C = (4π * A) / P². Circularity C is a value between 0 and 1. The closer C is to 1, the closer the shape of the region is to a circle and the more compact it is; the smaller the C value, the more elongated or irregular the shape, such as a linear crack. Circularity helps the model distinguish between crack types with similar areas but drastically different morphological distributions. By introducing and calculating fractal dimension and circularity, a deep geometric descriptor characterizing the morphological complexity and spatial distribution features of cracks is provided for the crack volume prediction model.

[0053] Step 40: Input the surface geometry parameters, crack type, and crack area into the pre-trained crack volume prediction model to obtain the crack volume prediction value.

[0054] In this step, all surface geometric parameters are used as input feature vectors and fed into a pre-trained crack volume prediction model. This model is trained on a large amount of data and can learn the nonlinear mapping relationship between complex surface morphology and three-dimensional internal volume. After receiving the input, the model performs forward calculation and finally outputs a continuous value, which is the predicted volume value of the crack.

[0055] As an optional embodiment, before inputting the surface geometry parameters, crack type, and crack area into a pre-trained crack volume prediction model to obtain the crack volume prediction value, the method may further include: Step 01: Obtaining a training sample set, which includes the surface geometry parameters extracted from the sample cracks and the true volume value of the sample cracks obtained by destructive measurement, wherein the destructive measurement is performed by controlled fluid injection and displacement.

[0056] In this step, representative road sections are selected to include cracks of different types and sizes as samples. High-resolution images of these sample cracks are acquired non-destructively, and their complete set of surface geometric parameters is extracted. After the surface information is collected, destructive measurements are performed on the same sample crack using a controlled fluid injection and displacement method to obtain its true volume value. A small piece of the complete road structure containing the crack is cut off using precision cutting equipment to form a laboratory sample. After sealing the non-observation surface of the sample, a low-viscosity, high-flow-rate special fluid of known density is completely injected into the cavity of the crack under controlled conditions. By accurately measuring the mass or volume difference of the fluid before and after injection, the volume of fluid required to completely fill the crack cavity is calculated based on the fluid density. This volume is the high-precision true volume value of the crack, avoiding any geometric assumptions and directly measuring the physical space.

[0057] Furthermore, the surface geometric parameters of each sample crack are associated with the corresponding volume true value to form a training data set. A large number of such data pairs constitute the training sample set for model training.

[0058] Step 02: Introduce the environmental covariates corresponding to the training sample set as input features. The environmental covariates include road surface material type, temperature at the time of collection, and regional climate information. In this step, to enable the model to adapt to different materials and environments of the road surface and improve its generalization ability, environmental covariates are introduced as additional input features during the training phase. These features, together with surface geometric parameters, constitute an enhanced feature vector, including the type of road surface material, the gradation type of asphalt mixture, and the petrochemical composition, which are encoded using unique thermal coding or embedded coding; the temperature at the time of data collection, which reflects the thermodynamic state of the material; and regional climate information, such as climate zones, annual average humidity, and annual precipitation, which reflects long-term environmental loads.

[0059] By fusing these environmental covariates, the model can learn the adjustment mechanism of surface features and volume relationships under different contexts, thus maintaining reliable prediction accuracy when facing new scenarios not fully covered by training data.

[0060] Step 03: Use the training sample set to train and validate the candidate machine learning models in order to select the optimal model.

[0061] In this step, several representative machine learning regression models are selected for comparison, such as the multivariate linear regression benchmark model, support vector regression, random forest, gradient boosting decision tree, and multilayer perceptron. The training sample set is divided into a training set and a validation set according to a certain ratio. The training set is used to train each candidate model and learn the mapping relationship. The predictive performance of each trained candidate model is evaluated on the validation set. Evaluation metrics typically include root mean square error, mean absolute percentage error, and coefficient of determination. Techniques such as cross-validation are used to ensure the robustness of the evaluation. The model with the best overall performance on the validation set is selected as the final crack volume prediction model. As an optional embodiment, after training and validating the candidate machine learning models using the training sample set to select the optimal model, the method may further include: Step 501: Calculate the SHAP value corresponding to each input feature for the crack volume prediction value; Step 502: Generate a global feature importance ranking list based on the SHAP values.

[0062] In this step, SHAP (SHapley Additive exPlanations) value analysis is used to interpret the prediction results. First, the global importance interpretation shows the average influence of each feature on the model's prediction results. Then, the local interpretation, for the prediction of a single crack, reveals how each feature specifically drives the predicted value to change in the direction of larger or smaller, showing how an increase in the maximum width directly leads to an increase in the predicted volume, making the prediction results transparent and traceable.

[0063] As an optional embodiment, the method may further include: during the training of the crack volume prediction model, the regional climate characteristics of annual average rainfall, extreme low temperatures, and real-time pavement condition parameters such as moisture content and surface temperature, along with crack surface parameters and structural parameters, can be used as inputs to reflect the influence of crack development on depth under different climatic conditions. To eliminate data distribution differences across regions, a domain adversarial neural network can be employed. This network learns universal features across regions, such as the ratio of crack length to depth, through a feature extractor, and then trains adversarially using a domain classifier. This allows the model to ignore regional differences, improving the universality and reliability of depth prediction nationwide. The focus is on extracting common features that effectively characterize crack volume, such as the intrinsic relationship between crack surface parameter set area, fractal dimension, and the true volume value. The extracted features are then used to identify the region of the sample, such as whether it is in the south or north. Through adversarial training, the features output by the feature extraction module are unaffected by regional differences, obtaining a universal representation across regions. This eliminates data distribution bias and improves the stability and reliability of crack depth prediction in different geographical areas.

[0064] As an optional implementation, based on the predicted crack volume value combined with the structural bearing characteristics and crack propagation risk of semi-rigid asphalt pavement, a crack volume level label is generated, and a differentiated maintenance strategy is matched. Specifically, this can include: classifying crack volume levels based on the correlation threshold between crack volume and pavement structural layers; Low-risk cracks: Low-risk threshold, this threshold corresponds to cracks that are limited to the shallow layer of the surface layer, have not damaged the surface layer's load-bearing structure, and have no obvious propagation trend; Medium-risk cracks: Medium-risk threshold, this range of cracks has penetrated the surface layer and partially intruded into the top surface of the base layer, which may lead to rainwater infiltration causing the base layer to loosen, weakening the pavement's bearing capacity, and posing a risk of slow propagation; High-risk cracks: High-risk threshold, this threshold corresponds to cracks that have penetrated into the base layer, forming a surface-base layer through-crack, whose volume expansion easily leads to uneven stress on the base layer, a decrease in pavement bearing capacity, and is prone to rapid propagation into potholes under repeated vehicle loads. By comparing the predicted crack volume value with the above thresholds, the crack's level is automatically marked, providing a basis for matching subsequent maintenance strategies.

[0065] Furthermore, low-risk cracks are monitored using a combination of regular manual inspections and image re-collection. Crack images are collected periodically via a mobile detection platform, and the volume change rate is compared. If the volume change rate increases, the crack is upgraded to a medium-risk level. For medium-risk cracks, humidity sensors are deployed at intervals along the crack's direction. Simultaneously, volume data is updated weekly via image detection. The humidity sensors monitor the moisture content of the base layer in the crack area in real time. When the critical moisture content threshold of the base layer material or the weekly volume change rate exceeds a preset threshold, an early warning is triggered. For high-risk cracks, micro-strain sensor arrays and temperature sensors are deployed at the crack tip and on both sides. The sensors are connected to the road surface intelligent monitoring platform via LoRa wireless transmission to achieve 24-hour real-time monitoring. The focus is on tracking the strain value at the crack tip. Based on the crack resistance performance test of the semi-rigid base layer material, the daily volume change rate is obtained. When any indicator exceeds the limit, an emergency response is initiated.

[0066] Furthermore, for low-risk cracks, the system recommends lightweight surface sealing treatments, such as fog sealants or low-viscosity sealants, to prevent rainwater infiltration. Simultaneously, quarterly inspections should be maintained to monitor volume changes. When the crack growth rate is within a stable range, no additional structural intervention is required; only moderate traffic organization optimization suggestions are provided for heavily loaded sections. For medium-risk cracks, the system recommends a repair method combining grouting and local reinforcement to restore the interface integrity between the surface layer and the base layer. Based on humidity and volume change monitoring results, timely base layer waterproofing reinforcement treatment should be carried out. If the volume growth rate remains high, the system will increase the monitoring frequency and prompt early structural maintenance. For high-risk cracks, the system will immediately output a structural repair plan, including base layer grouting reinforcement and deep crack filling treatment, to quickly curb the expansion of through cracks. Based on real-time strain and temperature monitoring results, emergency speed limits or local closure measures will be triggered if necessary, and the emergency maintenance system will be coordinated to prioritize deep repairs to ensure the safety and stability of the pavement structure.

[0067] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide a device for estimating the volume of cracks in asphalt pavement, the structure of which is as follows: Figure 2 As shown.

[0068] Figure 2 This is a schematic diagram of the internal structure of an asphalt pavement crack volume estimation device provided in an embodiment of this application. Figure 2 As shown, the device includes: At least one processor 201; And a memory 202 that is communicatively connected to at least one processor; The memory 202 stores instructions that can be executed by at least one processor, which are executed by at least one processor 201 to enable at least one processor 201 to: perform any step of an asphalt pavement crack volume estimation method.

[0069] Some embodiments of this application provide corresponding to Figure 1 A non-volatile computer storage medium for estimating the volume of cracks in asphalt pavement, storing computer-executable instructions, wherein the computer-executable instructions are configured to execute any one of the steps of a method for estimating the volume of cracks in asphalt pavement.

[0070] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for IoT devices and media are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0071] The systems, media, and methods provided in this application are one-to-one correspondences. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0072] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0073] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0074] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0075] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0076] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0077] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0078] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0079] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0080] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for estimating the volume of cracks in asphalt pavement, characterized in that, The method includes: High-resolution images of the road surface are acquired, and the high-resolution images are segmented using an image segmentation model to obtain binary images of the cracks. The skeleton of the crack binary image is extracted to obtain a structured crack skeleton map, and the crack type and crack area are obtained. Based on the crack skeleton map, the surface geometric parameters of the crack are extracted through orientation correction, orthogonal projection and edge detection. The surface geometric parameters include crack length, crack width and shape factor. The surface geometry parameters, crack type, and crack area are input into a pre-trained crack volume prediction model to obtain the crack volume prediction value.

2. The method for estimating the volume of cracks in asphalt pavement according to claim 1, characterized in that, The binary image of the crack is subjected to skeleton extraction to obtain a structured crack skeleton map, which reveals the crack type and crack area. Specifically, this includes: An initial crack centerline is generated in the binary crack image using a morphological skeleton extraction algorithm. Identify the nodes, intersections, and consecutive pixels of the initial crack centerline to construct a weighted directed graph representing the crack topology; A branch pruning strategy is used to remove noisy branches and repair fractures in the weighted directed graph to obtain the structured crack skeleton graph. The area of ​​the crack is obtained by statistically analyzing the pixels representing the crack in the binary image of the crack and converting them according to the pixel resolution. The crack type is identified based on the orientation and intersection features of the crack skeleton diagram.

3. The method for estimating the volume of cracks in asphalt pavement according to claim 2, characterized in that, Based on the crack skeleton map, the surface geometric parameters of the crack are extracted through orientation correction, orthogonal projection, and edge detection, specifically including: Calculate the main direction of the crack pixels in the crack skeleton image, and correct the direction of the crack skeleton; Vertical search lines are laid out at preset intervals along the direction of the corrected crack skeleton; Edge point search is performed along each of the vertical search lines to both sides, and the edge pixel points on both sides of the crack are determined based on the gray-scale change value of the crack pixel; Calculate the Euclidean distance between each pair of edge pixels to obtain the crack width, and calculate the average width and width distribution based on the crack widths of all search lines; The continuous pixels are smoothly fitted, the length of the fitted curve is calculated, and the total length of the crack is obtained.

4. The method for estimating the volume of cracks in asphalt pavement according to claim 3, characterized in that, The step of searching for edge pixels along each of the vertical search lines to both sides, and determining the edge pixels on both sides of the crack based on the pixel grayscale abrupt change value, specifically includes: Crack pixels are scanned from the crack skeleton to both sides along the vertical search line; On each side, the first location point where the pixel grayscale change value exceeds a preset threshold is recorded as a candidate edge pixel point; Calculate the grayscale gradient values ​​of the candidate edge pixels to obtain the edge confidence score; Among all candidate edge pixels on each side, the point with the highest edge confidence is selected as the valid edge pixel.

5. The method for estimating the volume of cracks in asphalt pavement according to claim 1, characterized in that, Based on the crack skeleton map, the surface geometric parameters of the crack are extracted through direction correction, orthogonal projection, and edge detection, specifically including: The shape factor includes fractal dimension and circularity; The fractal dimension was obtained by calculating the crack profile using the box counting method. The crack perimeter is obtained by using a boundary tracking algorithm. The crack area and crack perimeter are then calculated to obtain the circularity.

6. The method for estimating the volume of cracks in asphalt pavement according to claim 5, characterized in that, The calculation of the crack profile using the box counting method to obtain the fractal dimension specifically includes: The grid side length covering the binary image of the crack is set to multiple successively decreasing scales; For each scale of the grid, count the number of grids that cover the crack profile; Establish a data point set between the logarithm of the scale and the logarithm of the corresponding number of grid cells; A linear fit is performed on the data point set, and the slope of the straight line obtained by the linear fit is determined as the fractal dimension of the crack profile.

7. The method for estimating the volume of cracks in asphalt pavement according to claim 1, characterized in that, Before inputting the surface geometry parameters, crack type, and crack area into a pre-trained crack volume prediction model to obtain the crack volume prediction value, the method further includes: A training sample set is obtained, which includes surface geometric parameters extracted from sample cracks and the true volume of the sample cracks obtained through destructive measurement using controlled fluid infusion and displacement. The environmental covariates corresponding to the training sample set are introduced as input features. The environmental covariates include road surface material type, temperature at the time of collection, and regional climate information. The candidate machine learning models are trained and validated using the training sample set and the input features to select the optimal model.

8. The method for estimating the volume of cracks in asphalt pavement according to claim 7, characterized in that, After training and validating candidate machine learning models using the training sample set to select the optimal model, the method further includes: Calculate the SHAP value corresponding to each input feature for the predicted crack volume; A global feature importance ranking list is generated based on the SHAP values.

9. A device for estimating the volume of cracks in asphalt pavement, characterized in that, The device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: Perform the steps of the method for estimating the volume of cracks in asphalt pavement as described in any one of claims 1-8.

10. A non-volatile computer storage medium for estimating the volume of cracks in asphalt pavement, storing computer-executable instructions, characterized in that, The computer-executable instructions are set as follows: Perform the steps of the method for estimating the volume of cracks in asphalt pavement as described in any one of claims 1-8.

Images