Highway pavement damage intelligent identification and evaluation method

Abstract

The invention relates to the technical field of intelligent identification, in particular to a highway pavement damage intelligent identification and evaluation method, which comprises the following steps: detecting a structure continuity and texture density region based on a pavement image, extracting crack, pit slot and track contour marks, grouping crack, pit slot and track texture to generate an identification graph, and evaluating the pavement damage. And extracting a crack area texture and gradient change to generate a difference layer, judging trend offset to generate an evolution label, and adjusting a classification boundary to evaluate a damage level. According to the method, the structure continuity and the texture density area are detected, the positioning precision is improved, misrecognition is reduced, direction gradient and texture changes are collected under multiple scales, the crack directions are clustered and grouped, the texture recognition stability is enhanced, the difference of the crack area and the periphery is analyzed, boundary jump is recognized, difference perception is enhanced, and the dynamic recognition depth is improved. Evaluation fuzzy errors are reduced, the overall process improves morphological representation precision and trend tracking and classification boundary stability, and the method adapts to various road condition changes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent identification, and particularly relates to a highway pavement damage intelligent identification and evaluation method. BACKGROUND

[0002] The technical field of intelligent identification contains automatic identification and classification processing of various types of information such as images, voices and signals through algorithm models. The core content of this technical field is to extract identifiable features from unstructured or semi-structured data by using image acquisition, feature extraction, pattern recognition and machine learning techniques, and to perform efficient information processing and target determination. In this technical field, systematic research includes sub-directions such as image recognition, target detection, semantic segmentation and behavior recognition, and is widely applied in many fields such as traffic monitoring, safety protection and industrial detection. In particular, in the direction of road detection, image information is collected by front-end sensors, and the state of the road is analyzed in combination with a trained model, which has become one of the key contents of current research and application.

[0003] Among them, the highway pavement damage intelligent identification and evaluation method refers to a specific method for identifying and quantitatively evaluating the type and degree of pavement damage by constructing an image recognition and classification process for the condition of highway pavement damage. This method covers technical matters such as pavement image acquisition and processing, damage area identification, damage type classification and damage degree evaluation. It uses image acquisition devices mounted on mobile platforms to obtain continuous pavement images, uses image segmentation methods to locate the position of damaged areas, identifies specific damage types such as cracks, potholes and ruts through convolutional neural networks, and estimates the damage degree level based on indicators such as area, length-width ratio and other indicators, and finally forms damage evaluation data.

[0004] In the prior art, the identification of damaged areas relies on image segmentation and basic classification model processing of feature maps, which is difficult to distinguish local structure continuity and texture density changes in depth, leading to non-damaged areas being easily misjudged as damaged points, especially under the conditions of light interference, material reflection or uneven road surface. The type distinction of cracks, potholes and ruts mainly depends on image feature extraction under a single scale, lacking direction aggregation processing under multi-scale gradient changes, so that the texture details of the same type of cracks, potholes and ruts under different scales cannot be uniformly labeled, which easily leads to type misclassification. The existing evaluation process is mostly based on static analysis of single-frame images, which cannot reflect the trend evolution of road damage over time, lacks comparison of the evolution trend of original images, and easily ignores the morphological migration caused by progressive expansion or seasonal changes. In the classification process, the joint determination mechanism of the contour range and the gray gradient change cannot be established, and the classification boundary judgment under a single feature is unstable, which easily causes boundary jumping phenomenon in the transition area, affecting the continuity and reliability of the final grade evaluation. For example, when identifying a developing road section with initial cracking, the existing technology cannot distinguish the transition process from fine lines to multiple crack intersections, which finally leads to low grade evaluation, affecting the maintenance priority ranking and resource allocation efficiency. SUMMARY

[0005] The purpose of the present application is to solve the problems existing in the prior art and to provide an intelligent identification and evaluation method for highway pavement damage.

[0006] In order to achieve the above purpose, the present application adopts the following technical scheme: an intelligent identification and evaluation method for highway pavement damage, comprising the following steps: S1: Based on the highway pavement image, the pixel region with structure continuity and texture density is detected in the image, the target area contour is extracted according to the closure degree of the pixel boundary, and the crack, pothole and rut contour identification data are obtained; S2: Based on the crack, pothole and rut contour identification data, the direction gradient distribution and texture contrast change of the crack, pothole and rut region are collected under different scales, and the grouping and labeling are carried out by using continuous pixel gradient direction clustering, and the crack, pothole and rut texture grouping identification map is generated; S3: The crack, pothole and rut texture grouping identification map is called to extract the texture fluctuation and gray gradient change of the asphalt pavement cracking area and the surrounding area, identify the gradient jump coefficient of the damage boundary, screen the difference area, and generate the damage texture difference layer; S4: Based on the damage texture difference layer, the image gradient distribution change of the same positioning area in the highway inspection image database is compared, the texture aggregation evolution direction is extracted and the trend deviation type is judged, and the damage trend evolution label set is generated.

[0007] As a further scheme of the present application, the crack, pothole and rut profile identification data comprises a profile point set density, a boundary closure degree coefficient and a texture continuity index, the crack, pothole and rut texture grouping identification map comprises a direction distribution cluster label, a texture contrast difference layer and a pixel clustering number, and the damage texture difference layer comprises a gray level jump amplitude map, a texture disturbance area boundary and a difference area mask, and the damage trend evolution label set comprises a texture aggregation trajectory mode, an evolution trend classification label and a time sequence offset direction.

[0008] As a further scheme of the present application, the crack, pothole and rut profile identification data comprises a profile point set density, a boundary closure degree coefficient and a texture continuity index, the crack, pothole and rut texture grouping identification map comprises a direction distribution cluster label, a texture contrast difference layer and a pixel clustering number, and the damage texture difference layer comprises a gray level jump amplitude map, a texture disturbance area boundary and a difference area mask, and the damage trend evolution label set comprises a texture aggregation trajectory mode, an evolution trend classification label and a time sequence offset direction. S111: based on the highway pavement image, extracting a pixel boundary set of a texture dense area and a gray level mutation area in the image, identifying a gradient direction distribution density and a texture distribution interval of the pixel boundary, identifying boundary continuity and delimiting a closed structure area, and generating a closed structure pixel distribution map; S112: calling the closed structure pixel distribution map, screening a target block with a complete boundary and an abnormal area texture density, identifying a pixel density variation and a shape deviation rate of the profile edge, and labeling a damage area according to a crack feature, to obtain the crack, pothole and rut profile identification data.

[0009] As a further scheme of the present application, the crack, pothole and rut profile identification data comprises a profile point set density, a boundary closure degree coefficient and a texture continuity index, the crack, pothole and rut texture grouping identification map comprises a direction distribution cluster label, a texture contrast difference layer and a pixel clustering number, and the damage texture difference layer comprises a gray level jump amplitude map, a texture disturbance area boundary and a difference area mask, and the damage trend evolution label set comprises a texture aggregation trajectory mode, an evolution trend classification label and a time sequence offset direction. S211: based on the crack, pothole and rut profile identification data, collecting a direction gradient distribution value and a local texture contrast value of a pixel in the area, identifying a direction variation rate and a contrast fluctuation between adjacent pixels, classifying the direction gradient, and obtaining a direction gradient contrast structure quantity; S212: calling the direction gradient contrast structure quantity, clustering to form a continuous pixel group according to a pixel direction homogeneity and an adjacent position relationship, analyzing a direction concentration degree and a contrast fluctuation in the group, calculating a continuous pixel direction grouping separation rate, delimiting a boundary interval and clustering to label a texture group, and generating the crack, pothole and rut texture grouping identification map.

[0010] As a further scheme of the present application, the crack, pothole and rut profile identification data comprises a profile point set density, a boundary closure degree coefficient and a texture continuity index, the crack, pothole and rut texture grouping identification map comprises a direction distribution cluster label, a texture contrast difference layer and a pixel clustering number, and the damage texture difference layer comprises a gray level jump amplitude map, a texture disturbance area boundary and a difference area mask, and the damage trend evolution label set comprises a texture aggregation trajectory mode, an evolution trend classification label and a time sequence offset direction. S311: calling the crack, pothole and rut texture grouping identification map, collecting a gray level value and a texture variation rate of a cracking area and a neighboring area of the asphalt pavement, and obtaining a texture gray level difference of the cracking area through a bidirectional comparison of a gray level variation gradient and a texture fluctuation difference; S312: According to the crack area texture gray difference, the gray mutation value, the texture fluctuation coefficient and the grouping center offset degree of each pixel in the spatial neighborhood are detected, the pixel boundary jump response value is calculated, the interval comparison of the pixel response value is performed in combination with the set boundary mutation reference value, the spatial position meeting the mutation characteristics is screened, and the highway surface jump response data is obtained; S313: The highway surface jump response data is called, the region pixel set with a jump response value greater than the boundary mutation reference value is screened, the spatial continuous region is aggregated and classified, the region morphology and distribution characteristics are matched to the existing pavement damage template, and a damage texture difference layer is generated.

[0011] As a further scheme of the present application, the acquisition step of the damage trend evolution label set is specifically: S411: Based on the damage texture difference layer, the gradient direction distribution graph and the texture contour structure boundary graph of the time node image are extracted, the texture boundary direction of the key region is angle counted, and the region concentration of the main direction mutation position in the adjacent image is compared, so as to obtain the damage deformation main direction trajectory; S412: The damage deformation main direction trajectory is called, the trend bias motion index is calculated in combination with the direction concentration degree ratio in the target section, the texture structure extension ratio and the disturbance amplitude of the mutation point neighborhood, the bias evolution path in the image sequence is identified according to the index direction turning distribution and the dense region aggregation characteristics, and the damage trend evolution label set is generated.

[0012] As a further scheme of the present application, the method further comprises a step S5: S5: The damage trend evolution label set is called, the contour expansion range and the gradient direction change degree in the current image region are jointly judged, the image region classification boundary is adjusted according to the matching degree of the label set, and the pavement damage evaluation level is obtained. The pavement damage evaluation level comprises a structure expansion level, a gradient change amplitude index and a matching degree judgment factor.

[0013] As a further scheme of the present application, the acquisition step of the pavement damage evaluation level is specifically: S511: The damage trend evolution label set is called, the contour gradient change comparison sequence is generated in combination based on the contour expansion range value and the gradient direction change value extracted in the image region, the matching degree is calculated in sequence with the reference combination value in the damage trend evolution label set, and the label matching difference is obtained. S512: The label matching difference is called, the contour gradient change sequence of the current boundary position is compared, the boundary point with a difference coefficient less than a threshold value is screened as an adjustment reference point according to the difference coefficient and the matching threshold value, and the image boundary coordinate adjustment value is obtained in combination with the coordinate repositioning classification boundary. S513: According to the image boundary coordinate adjustment value and the original classification boundary mapping value, the consistency degree of the contour change rate and the gradient direction in each region is extracted, the level standard in the damage trend evolution label set is matched, and a road surface damage evaluation level is obtained.

[0014] Compared with the prior art, the advantages and positive effects of the present application are that: In the present application, by detecting the regions with prominent structure continuity and texture density in the image, the crack, pit and rut contour data extracted according to the closure degree of the pixel boundary are used to effectively improve the spatial positioning accuracy of the target region, reduce the probability of misidentification of non-damaged regions, collect the directional gradient distribution and texture contrast change under the differentiated scale, and realize clear grouping of the direction structure of cracks, pits and ruts through clustering processing of continuous pixel gradient directions, thereby improving the stability of texture pattern recognition. In the process of comparing the texture fluctuation and gray gradient difference between the crack region and the surrounding region, the severe jump area of the damage boundary is screened out, and the difference perception ability between the damage region and the background is further strengthened. By comparing the gradient evolution process of the same positioning region in the original image data, the evolution direction and trend offset of the texture aggregation are extracted, and the recognition breadth and time depth of the dynamic change of the damage are enhanced. The contour expansion phenomenon in the image region and the gradient direction change degree are jointly evaluated, the continuity level between different damage forms is divided, and the recognition error caused by the fuzzy evaluation level is effectively alleviated. The overall processing procedure runs through five core links of spatial structure extraction, local texture clustering, boundary strength judgment, time sequence evolution analysis and joint classification determination, so that the representation accuracy of complex damage forms in the image, the change trend tracking ability and the classification boundary stability are comprehensively enhanced, the damage level evaluation result has stronger detail distinction and time reliability, and is suitable for various road condition change scenes. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 It is a main step schematic diagram of the present application; Figure 2 It is a flowchart of acquiring crack, pit and rut contour identification data in the present application; Figure 3 It is a flowchart of acquiring crack, pit and rut texture grouping identification map in the present application; Figure 4 It is a flowchart of acquiring damage texture difference layer in the present application; Figure 5 It is a flowchart of acquiring damage trend evolution label set in the present application; Figure 6 It is a flowchart of acquiring road surface damage evaluation level in the present application. DETAILED DESCRIPTION

[0016] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0017] In the description of the present application, it should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings, which are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application. In addition, in the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise explicitly and specifically limited. Embodiments

[0018] Please refer to Figure 1 The present application provides a technical solution: a highway pavement damage intelligent identification and evaluation method, comprising the following steps: S1: Based on the highway pavement image, the pixel region with structural continuity and texture density is detected in the image, the target region contour is extracted according to the closure degree of pixel boundary, and the crack, pit and rut contour identification data is obtained; S2: Based on the crack, pit and rut contour identification data, the directional gradient distribution and texture contrast change of the crack, pit and rut region are collected under the differentiation scale, the continuous pixel gradient direction clustering is used for grouping and labeling, and the crack, pit and rut texture grouping identification map is generated; S3: The crack, pit and rut texture grouping identification map is called to extract the texture fluctuation and gray gradient change of the asphalt pavement cracking area and the surrounding area, identify the gradient jump coefficient of the damage boundary, screen the difference area, and generate the damage texture difference layer; S4: Based on the damage texture difference layer, the image gradient distribution change of the same positioning area in the highway inspection image database is compared, the texture aggregation evolution direction is extracted and the trend deviation type is determined, and the damage trend evolution label set is generated; S5: The damage trend evolution label set is called to jointly determine the contour expansion range and gradient direction change degree in the current image area, and the image area classification boundary is adjusted according to the matching degree of the label set, and the pavement damage evaluation grade is obtained.

[0019] The data for identifying the contours of cracks, potholes, and ruts includes the density of contour points, the degree of boundary closure, and the texture continuity index. The texture grouping identification map for cracks, potholes, and ruts includes directional distribution cluster labels, texture contrast difference layers, and pixel cluster numbers. The damage texture difference layer includes grayscale jump amplitude maps, texture disturbance area boundaries, and difference area masks. The damage trend evolution label set includes texture aggregation trajectory patterns, evolution trend classification labels, and temporal offset directions. The pavement damage assessment level includes structural expansion level, gradient change amplitude index, and matching degree judgment factor.

[0020] Please see Figure 2 The specific steps for obtaining the contour identification data of cracks, potholes, and ruts are as follows: S111: Based on highway pavement images, extract the pixel boundary set of texture-dense areas and gray-level abrupt change areas in the image, identify the gradient direction distribution density and texture distribution spacing of pixel boundaries, identify boundary continuity and delineate closed structure regions, and generate a closed structure pixel distribution map.

[0021] The process extracts the pixel boundary sets of texture-dense regions and gray-level abrupt change regions in the image. Gradient calculations are used to determine the degree of pixel change. Gradient calculations are performed on the gray-level values ​​of each pixel's surrounding neighborhood to obtain the gradient magnitude and direction, defined as the gradient direction distribution density. The change between each group of neighboring pixels is calculated, and texture-dense regions in the image are marked. The texture distribution spacing is calculated using pixel spacing to obtain the pixel homogeneity of different regions, refining the interval of texture changes. The continuity of boundaries in the image is analyzed to determine whether they exhibit a closed shape. If the boundary is continuous and the directional change is small, it is considered a closed structure. Closed regions are obtained through connected component analysis. Based on this region, the distribution of pixels in the closed structure is further identified. Through pixel connections and edge recognition, the final image distribution of the closed structure region is obtained. In practice, the integrity of the boundary can be judged by defining a specific threshold. If the gray value change is less than the set threshold, the area is considered to be a smooth area with high boundary continuity. Conversely, if the change is greater than the threshold, it is considered to be an area with large gray value abrupt changes. This method can be used to extract different structural areas such as cracks, potholes, and ruts on the road surface. For example, if the gray value change gradient of a certain area in the image is greater than 50 and the gradient direction of the area is consistent and smooth, the area can be identified as a continuous boundary, and a closed structure pixel distribution map is finally generated.

[0022] S112: Call the closed structure pixel distribution map, filter target blocks with complete boundaries and abnormal regional texture density, identify pixel density changes and morphological deviation rates of contour edges, mark damaged areas according to crack features, and obtain contour identification data of cracks, potholes and ruts.

[0023] To filter target blocks with complete boundaries and abnormal texture density, the first step is to classify the target blocks in the image. Filtering is based on the texture density of the target blocks; abnormal density regions are defined by the number and distribution of pixels. Texture density is determined by calculating the variation in grayscale values ​​of pixels within the region. If the texture density of a region exceeds a set threshold, it is considered to have abnormal texture. After filtering the target blocks, the pixel density of their contour edges is analyzed. Specifically, the edges of the target blocks are detected using edge detection methods such as the Sobel operator to obtain the gradient direction and grayscale change of each edge pixel. Then, the characteristics of cracks are determined based on the shape deviation rate. The shape deviation rate is obtained by calculating the smoothness and regularity of the edges. If the deviation rate exceeds a certain threshold, it indicates damage such as cracks in the region, and the region is further labeled as a damaged area, obtaining the contour identification data of cracks, potholes, and ruts. The shape deviation rate and contour curvature are calculated based on the curve changes of edge pixels, while uniformity is statistically analyzed through the distribution of pixels within the region. For example, when the difference between the contour curvature and uniformity of a target block is greater than 0.3, the area can be considered to have crack characteristics, and cracks can be marked.

[0024] Please see Figure 3 The specific steps for obtaining the texture grouping and identification maps of cracks, potholes, and ruts are as follows: S211: Based on the contour identification data of cracks, potholes and ruts, collect the directional gradient distribution value and local texture contrast value of pixels in the region, identify the directional change rate and contrast fluctuation between adjacent pixels, classify the directional gradient, and obtain the directional gradient contrast structure quantity.

[0025] Acquiring directional gradient and texture contrast data of pixels within crack areas is a crucial step in road maintenance for detecting minute cracks and enabling early intervention. High-resolution road surface images are used to obtain detailed pixel information about the crack area. For example, on an urban road, data captured using automated image acquisition can include crack images from different angles and under different lighting conditions. For each pixel, its gradient changes in the vertical and horizontal directions, as well as its texture contrast with neighboring pixels, are calculated. From this data, the directional gradient change value and texture contrast fluctuation value for each pixel can be calculated. For example, assuming a pixel's gradient is 0.3 in the horizontal direction and 0.2 in the vertical direction, with a texture contrast of 15%, this calculation helps determine whether the crack is continuous and its width and depth. Based on this, an orientation consistency matrix is ​​constructed using the data. This matrix helps determine the overall direction and morphology of the crack, ultimately obtaining the directional gradient contrast structure, a key parameter for assessing crack severity and necessary maintenance measures. For example, by analyzing the orientation consistency matrix, it can be determined whether the road surface crack extends in a straight line or has a branching tendency.

[0026] S212: Using the directional gradient comparison structure, based on pixel orientation similarity and adjacency relationships, clusters pixels into continuous pixel groups, analyzes the orientation concentration and contrast fluctuation within each group, using the following formula: ; Calculate the group separation rate of continuous pixel directions, delineate the boundary intervals and cluster and label the texture groups to generate texture group identification maps of cracks, potholes and ruts; Where T represents the group separation rate in the direction of continuous pixels, and G i G represents the gradient value in the direction of the i-th pixel. i+1 C represents the gradient value in the direction of the (i+1)th pixel. i Let D be the local texture contrast ratio of the i-th pixel. i Let L be the Euclidean distance between the i-th pixel and its neighboring pixels, and L be the directional gradient of the i-th pixel. i Let i be the spatial location projection value of the i-th pixel. The mean of the projected values ​​for all pixel locations is given by , where n represents the total number of pixels.

[0027] Pixels are grouped based on directional similarity and spatial proximity, effectively clustering pixels with similar textures within consecutive pixel regions to facilitate accurate identification and measurement of cracks, potholes, and ruts. By analyzing previously calculated directional gradient contrast structure data, the directional gradient value and texture contrast value of each pixel in the crack, pothole, and rut images are identified. For example, considering a damaged urban road, after preprocessing the image data of crack, pothole, and rut areas, the directional gradient and texture contrast value of each pixel will be used for subsequent clustering analysis. For each pixel, its directional gradient Euclidean distance to neighboring pixels is measured, and the deviation of its position projection value from the mean of all pixel position projection values ​​is calculated. Assuming a pixel has an directional gradient of 0.32, neighboring pixels have an directional gradient of 0.28, texture contrast is 0.15, directional gradient Euclidean distance is 0.04, and position projection is 500, while the average position projection of all pixels is 450, the data needs to be normalized to eliminate the influence of dimensions and maintain data consistency. The directional gradient value and directional gradient Euclidean distance are normalized using the minimax method, while the position deviation is normalized using the standard deviation.

[0028] Substituting specific values, let n=5 (analyzing groups of five pixels), the specific parameters are as follows: G i =0.32, G i+1 =0.28, therefore |G i+1 -G i |=0.04; C i =0.15 (texture contrast value) ; D i =0.04 (Euclidean distance of directional gradient); Substitute the parameter values ​​into the formula: ; The calculation result This value represents the grouping separation rate of continuous pixels. A lower value indicates high consistency of pixel direction within a group, resulting in high accuracy of the crack, pothole, and rut texture identification map. This value can be directly used in subsequent crack, pothole, and rut assessments and maintenance decisions, such as determining the severity of cracks and maintenance priorities. In this way, pixel groups with similar texture features within crack, pothole, and rut areas are marked to generate crack, pothole, and rut texture grouping identification maps, providing accurate data support for road maintenance.

[0029] Please see Figure 4 The specific steps for obtaining the damaged texture difference layer are as follows: S311: Call the texture grouping and identification map of cracks, potholes and ruts, collect the gray value and texture change rate of the cracked area and adjacent area of ​​the asphalt pavement, and obtain the texture gray value difference of the cracked area by bidirectional comparison of gray value change gradient and texture fluctuation difference.

[0030] Image data of asphalt pavement is acquired through a high-resolution scanning device mounted on an inspection vehicle. This device automatically captures high-definition images of the road surface while in motion, such as continuously scanning urban main roads. Real-time image processing software identifies and records various road surface features. This software can process large amounts of data, enabling rapid identification and marking of potential cracked areas. Next, a specialized image processing algorithm analyzes the captured images, separating the cracked areas and their surrounding regions. It calculates the grayscale value change and texture change indices for each pixel within the region, such as the grayscale value change rate and texture non-uniformity index. The data is then statistically analyzed using specialized software to obtain the average grayscale change rate and texture change index for each pixel within the region. This allows for comparison of significant differences between cracked and non-cracked areas. This difference helps identify genuine damaged areas from false alarms, ultimately generating a comprehensive image of the cracked area's texture and grayscale differences. This provides accurate data support for road maintenance and repair. For example, it serves as a primary reference in determining maintenance priorities and scope when developing an annual road maintenance plan for a city.

[0031] S312: Based on the grayscale differences in the cracked area, detect the grayscale abrupt change, texture fluctuation coefficient, and grouping center offset of each pixel in its spatial neighborhood, using the following formula: ; Calculate the pixel boundary jump response value, combine it with the set boundary abrupt change benchmark value, perform interval comparison on the pixel response value, filter the spatial locations that meet the abrupt change characteristics, and obtain the highway surface jump response data; Where Y represents the pixel boundary transition response value, A p A represents the grayscale value of the p-th pixel. p+1 B represents the grayscale value of the (p+1)th pixel. p N represents the texture volatility of the p-th pixel. p This represents the grayscale offset of the p-th pixel from the centroid of the texture group, where m represents the total number of pixels.

[0032] The road surface image is processed pixel-level based on the grayscale difference map of the cracked area texture to obtain the boundary grayscale abrupt change, texture fluctuation coefficient, and group offset grayscale value of each pixel. Specifically, the grayscale value A of each pixel is obtained. p The image is read directly from the grayscale channels of its original pixels, with a grayscale range of 0 to 255. To maintain dimensional consistency, this value needs to be normalized. The normalization process is as follows: , where a p The original grayscale value is used. For example, if the grayscale value of a pixel is 192, then after normalization... ; Texture volatility B p This is obtained by calculating the standard deviation of the grayscale values ​​of each pixel and its neighboring pixels (within a 3x3 window). This value also needs to be normalized using the maximum normalization method. Assuming the maximum standard deviation of texture fluctuation in the sample set is 20, if the standard deviation of a certain point is 8.4, then... ; Grayscale offset N p This represents the absolute value of the difference between the gray level of this pixel and the gray level of the centroid of its texture group. Let's say the centroid of a certain group is 180, and the original gray level of the current pixel is 210. ; After normalization, all parameters are substituted into the formula for calculation. Taking m=3, the original gray values ​​of the three sets of continuous boundary pixels are 192, 184, and 200. After normalization, they are A1=0.752, A2=0.721, and A3=0.784, respectively. The texture fluctuation rates of the three points are normalized to B1=0.42, B2=0.39, and B3=0.45, respectively, and the gray-level offsets are N1=0.118, N2=0.105, and N3=0.127, respectively. Substitute the data into the formula: .

[0033] The jump response value Y is 1.590. If the baseline jump response value is set to 1.200 (calculated from the average of 95% grayscale fluctuation range from the undamaged road surface sample set), then this jump value is higher than the threshold, indicating that there is a significant jump feature at the corresponding position of the pixel sequence, which belongs to the crack boundary and should be marked as an abnormal area. By combining three factors—grayscale abrupt change, local texture fluctuation, and grayscale centroid shift—the misjudgment error caused by single feature judgment is eliminated, thereby improving the accuracy of damage boundary identification. This result shows that the response value of the pixel set has exceeded the set threshold, and its position needs to be included in the subsequent crack boundary clustering area as an important data support for constructing the highway surface jump response map.

[0034] S313: Call the highway surface abrupt response data, filter the set of pixels in the region whose abrupt response value is greater than the boundary abrupt change benchmark value, aggregate and classify the spatially continuous regions, match them to the existing pavement damage template according to the region morphology and distribution characteristics, and generate a damage texture difference layer.

[0035] Using identified and labeled crack edge data, the process involves overlaying data from the road surface abrupt response map with the original road image. GIS (Geographic Information System) software is then used to merge and analyze the data layers. The GIS software can display the specific location and extent of different crack areas in detail, distinguishing cracks of varying severity through color coding or visual marking methods. This allows road management departments to intuitively understand the damage status of the entire road network, enabling them to develop more precise and targeted maintenance strategies. For example, in the annual road maintenance plan, this damage texture difference layer will serve as one of the bases for determining which road sections should be prioritized for maintenance, especially major roads in busy traffic areas. Generating a detailed damage texture difference layer can also help assess the effectiveness of existing road maintenance strategies and provide a scientific basis for future road design and material selection.

[0036] Please see Figure 5 The specific steps for obtaining the damage trend evolution label set are as follows: S411: Based on the damaged texture difference layer, extract the gradient direction distribution map and texture contour structure boundary map of the time node image, perform angle statistics on the texture boundary direction of key areas, and compare the regional concentration of the main direction change position in adjacent images to obtain the main direction trajectory of the damaged deformation.

[0037] Images of the same location area from the highway inspection image database are retrieved, and gradient direction distribution maps of images at each time point are extracted. Image processing software is used to calculate the gradient value of each pixel in the image and statistically analyze the distribution of the main direction. For example, if the distribution of the main direction is mainly concentrated in the horizontal direction during image processing, it indicates that the texture of the area mainly exhibits horizontal arrangement characteristics. By comparing the gradient distribution maps in adjacent images, the initial trend of road damage can be identified. For example, a sudden change in gradient direction compared to the previous time point indicates the development or change of damage. Then, the texture contour structure boundary map is further retrieved, which involves more refined image analysis. By extracting image edges and texture boundaries, edge detection algorithms are used to determine the contour and boundary of the damaged area. For example, edge detection can accurately identify the edges of cracks or potholes on the road, thereby enabling more accurate location and description of the damaged area. The concentration of the main direction change locations in adjacent images is compared to provide decision support for highway repair and maintenance. The main direction trajectory of damage deformation is obtained, which intuitively shows the direction and speed of damage development, providing an intuitive basis for subsequent repairs and enabling the prediction of future damage trends through specific change trajectories.

[0038] S412: Call the main direction trajectory of the damaged deformation, and combine it with the ratio of direction concentration in the target segment, the texture structure extension ratio, and the disturbance amplitude in the neighborhood of the abrupt change point, using the following formula: ; Calculate the trend offset dynamic index, identify the offset evolution path in the image sequence based on the index direction turning distribution and dense area aggregation characteristics, and generate a damage trend evolution label set; Where Z represents the trend offset dynamic index, d represents the direction concentration ratio, x represents the texture structure extension ratio, y represents the disturbance amplitude of the neighborhood of the mutation point, k represents the value of the mutation position in the main direction, v represents the total amplitude of the gradient direction difference, and w represents the deviation distance of the structure boundary. The ratio of directional concentration in the target region, d, is extracted. This parameter is obtained by extracting the principal gradient directions of local regions in different time-series images and statistically analyzing the frequency of directional distribution. For example, in five consecutively captured images, the principal gradient directions in region A are 0°, 5°, 10°, 10°, and 15°. The ratio of the frequency of the direction with the highest proportion of principal gradients to the total frequency is calculated as d. If the most frequent 10° appears twice, then d = 2 / 5 = 0.4, indicating that the directional concentration in this region is average. Obtain the texture structure extension ratio x. This value is obtained by measuring the ratio of the maximum extension length of the texture boundary of the damaged area at different time sequences to the initial length. If the initial boundary length is 1.2 meters and the boundary length measured in the current image is 1.8 meters, then x = 1.8 / 1.2 = 1.5, which reflects that the damaged texture in this area has undergone spatial expansion. Then, the disturbance amplitude y in the neighborhood of the mutation point is calculated. The rate of change of gray value is calculated for the local area within a radius of 0.5 meters of the mutation point. If the average gray value in the neighborhood of the mutation point changes from 130 to 100, the disturbance amplitude is y=|130-100| / 255=0.1176, where 255 is the maximum gray value. Normalization is performed to avoid inconsistencies in numerical dimensions. For the value k at the position of abrupt change in the main direction, count the number of pixels in the current image where the gradient direction abrupt change angle is greater than 30°. Let the number of pixels at the abrupt change position in the image be 120. After normalization, let the total number of pixels in the entire image be 1920×1080=2,073,600. Then k=120 / 2073600≈0.0000579. The total gradient direction difference, v, is the sum of the average gradient angle differences in the region across different time-series images. Assuming it is 45°, then v = 45 / 180 = 0.25. Normalization is applied to make it fall within the 0-1 range. The structural boundary deviation distance, w, is calculated by the change in the Euclidean distance between the boundary displacement vectors in adjacent images. If the average offset is 0.35 meters and the maximum offset is 2 meters, then the normalized w = 0.35 / 2 = 0.175. Substitute into the formula to calculate: .

[0039] By introducing practically collectable or calculable indicators such as directional concentration ratio, texture extension rate, and disturbance amplitude, and by normalizing and synthesizing the abrupt change location and offset, a dynamic trend offset index map with unified dimensions and sensitive change trends is effectively constructed. This map is suitable for dynamic damage path analysis. The results show that a high Z value reflects the active damage change trend of the current target section. Combined with Z>1.2 as the trend fluctuation offset warning threshold (derived from statistical fitting of original inspection image samples), the current result has entered the warning range. Based on this, a corresponding damage trend evolution label set will be generated for intelligent identification and subsequent intervention strategy suggestions.

[0040] Please see Figure 6 The specific steps for obtaining the road surface damage assessment level are as follows: S511: Call the damage trend evolution label set, and combine the contour expansion range value and gradient direction change value extracted from the image region to generate a contour gradient change comparison sequence. Then, calculate the matching degree with the reference combination value in the damage trend evolution label set in sequence to obtain the label matching difference.

[0041] The contour expansion range and gradient direction change values ​​are extracted from the image region. These two parameters represent the degree of boundary expansion and the boundary gradient change in the image, respectively. The contour expansion range refers to the distance that the boundary points expand outward according to a specific algorithm (such as morphological dilation) during image processing. This process requires measurement of the pixel neighborhood. During dilation, circular or square structuring elements are used for expansion, and each pixel position is adjusted according to the influence of its neighboring regions. The resulting expanded boundary is the contour expansion range value. The gradient direction change value is extracted using an image edge detection algorithm (such as the Sobel operator or the Canny algorithm). This value represents the degree of gradient change at different locations in the image, reflecting the trend of image boundary changes. These two values ​​are combined to generate a contour gradient change comparison sequence. This sequence can be understood as a continuous tracking record of image boundary changes, with each data point representing a combination of boundary expansion and gradient change at a specific location. Based on the comparison sequence, each comparison sequence is matched one by one with the reference combination values ​​in the damage trend evolution label set. The matching degree between each comparison sequence and the reference value is calculated. The matching degree can be calculated using metrics such as Euclidean distance and cosine similarity. The degree of matching between the contour of the current image and the preset damage trend in the label library is calculated, thus obtaining the label matching difference. This difference value reflects the deviation between the actual damage in the image and the ideal trend in the reference label library. For example, if the contour of a certain region has a large outward expansion range and significant gradient changes, but the matching degree with the normal boundary reference value in the label library is low, it indicates that there is a large damage trend in this region. The potential degree of damage can be inferred through the matching difference.

[0042] S512: Call the label matching difference, compare the contour gradient change sequence at the current boundary position, filter the boundary points with a difference coefficient less than the threshold based on the difference coefficient and the matching threshold, use them as adjustment reference points, and combine the coordinate relocation classification boundary to obtain the image boundary coordinate adjustment value.

[0043] The label matching difference is used to compare the contour gradient change sequence at the current boundary position. The difference coefficient is used to measure the matching difference between a boundary point in the current image and the label library. This coefficient is usually calculated as the ratio of the matching degree to the deviation of the ideal value in the label library, or by using statistical methods such as standard deviation to measure the similarity between the two. By calculating the difference coefficient, it can be determined which boundary points have smaller changes, i.e., the difference coefficient is less than the preset matching threshold. The setting of the difference coefficient threshold needs to be adjusted according to the actual scenario. This threshold can be set to a standardized value, such as 0.1, indicating that only boundary points with a difference coefficient less than 0.1 will be considered as adjustment reference points. The selection of adjustment reference points is based on the case where a boundary point has a small change, indicating that the point does not need adjustment, or that the point is located in the overlapping area between the actual boundary and the ideal boundary. Based on this, through coordinate relocation, the area where the adjustment reference point is located is compared with the preset classification boundary in the image, and the image boundary coordinate adjustment value is obtained according to the actual measured deviation. For example, if the boundary points of a certain region show a small difference coefficient in the detection and meet the conditions for adjusting the reference points, then the coordinates are adjusted according to the information, and the coordinates of the boundary points are updated to ensure that the image boundary matches the damage trend better, which can be used for subsequent processing or to correct the classification of the image.

[0044] S513: Based on the image boundary coordinate adjustment value and the original classification boundary mapping value, extract the consistency degree of contour change rate and gradient direction in each region, match the damage trend evolution label set level standard, and obtain the road damage assessment level.

[0045] The adjusted boundary coordinates are compared with the original classification boundary mapping values ​​to ensure consistency between the adjusted and initial boundary coordinates. The boundary mapping values ​​are generated by the image classification algorithm and represent the classification boundary of each region in the image. After comparison, the contour change rate and gradient direction consistency within each region are extracted. The contour change rate is the ratio of the change in contour within a specific region to the initial contour of that region, calculated using an image edge detection algorithm (e.g., gradient operator), reflecting the speed of contour change in the image. Gradient direction consistency refers to the consistency of the gradient direction at various points in the image. If the gradient directions of all points in a region are approximately the same, it indicates that the boundary of that region is regular. Based on these two parameters, the current region is matched with the level criteria in the damage trend evolution label set. The level criteria are set based on the contour change rate and gradient direction consistency of the image region, defining different ranges to label the damage level. For example, if a region has a high contour change rate and low gradient direction consistency, it indicates that the region has a severe damage trend. By comparing with the standards in the label set, the final road damage assessment level is obtained, which can be divided into multiple levels (e.g., mild, moderate, and severe damage).

[0046] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for intelligent identification and assessment of highway pavement damage, characterized in that, Includes the following steps: S1: Based on highway pavement images, detect pixel regions with prominent structural continuity and texture density in the images, extract the contours of target regions according to the degree of pixel boundary closure, and obtain contour identification data of cracks, potholes and ruts. S2: Based on the outline identification data of cracks, potholes and ruts, collect the directional gradient distribution and texture contrast changes of crack, pothole and rut areas at different scales, and use continuous pixel gradient direction clustering to group and label them to generate crack, pothole and rut texture group identification map. S3: Call the texture grouping and identification map of cracks, potholes and ruts, extract the texture fluctuation and gray-scale gradient change of the cracked area of ​​the asphalt pavement and the surrounding area, identify the gradient jump coefficient of the damaged boundary, filter the difference area, and generate a damage texture difference layer. S4: Based on the damaged texture difference layer, compare the image gradient distribution changes of the same location area in the highway inspection image database, extract the texture aggregation evolution direction and determine the trend offset type, and generate a damage trend evolution label set.

2. The intelligent identification and assessment method for highway pavement damage according to claim 1, characterized in that, The crack, pothole, and rut contour identification data includes contour point set density, boundary closure coefficient, and texture continuity index. The crack, pothole, and rut texture grouping identification map includes directional distribution cluster labels, texture contrast difference layer, and pixel cluster number. The damage texture difference layer includes grayscale jump amplitude map, texture disturbance area boundary, and difference area mask. The damage trend evolution label set includes texture aggregation trajectory pattern, evolution trend classification label, and temporal offset direction.

3. The intelligent identification and assessment method for highway pavement damage according to claim 1, characterized in that, The specific steps for obtaining the contour identification data of cracks, potholes, and ruts are as follows: S111: Based on highway pavement images, extract the pixel boundary set of texture-dense areas and gray-level abrupt change areas in the image, identify the gradient direction distribution density and texture distribution spacing of pixel boundaries, identify the continuity of boundaries and delineate closed structure regions, and generate a closed structure pixel distribution map. S112: Call the closed structure pixel distribution map, filter target blocks with complete boundaries and abnormal regional texture density, identify pixel density changes and shape deviation rates of contour edges, mark damaged areas according to crack features, and obtain contour identification data of cracks, potholes and ruts.

4. The intelligent identification and assessment method for highway pavement damage according to claim 3, characterized in that, The specific steps for obtaining the texture grouping and identification map of cracks, potholes, and ruts are as follows: S211: Based on the crack, pothole, and rut contour identification data, collect the directional gradient distribution value and local texture contrast value of pixels in the area, identify the directional change rate and contrast fluctuation between adjacent pixels, classify the directional gradient, and obtain the directional gradient contrast structure quantity. S212: Call the directional gradient comparison structure quantity, cluster the pixels into continuous pixel groups based on the sameness of pixel directions and the relationship between adjacent positions, analyze the directional concentration and contrast fluctuation within the group, calculate the separation rate of continuous pixel direction groups, delineate the boundary intervals and cluster and label the texture groups, and generate crack, pothole and rut texture group identification maps.

5. The intelligent identification and assessment method for highway pavement damage according to claim 4, characterized in that, The specific steps for obtaining the damaged texture difference layer are as follows: S311: Call the texture grouping and identification map of cracks, potholes and ruts, collect the gray value and texture change rate of the cracked area and adjacent area of ​​the asphalt pavement, and obtain the texture gray value difference of the cracked area by bidirectional comparison of gray value change gradient and texture fluctuation difference. S312: Based on the grayscale difference of the texture in the cracked area, detect the grayscale abrupt change, texture fluctuation coefficient and grouping center offset of each pixel in the spatial neighborhood, calculate the pixel boundary jump response value, combine it with the set boundary abrupt change benchmark value, perform interval comparison on the pixel response value, filter the spatial positions that meet the abrupt change characteristics, and obtain the highway surface jump response data. S313: Call the highway surface jump response data, filter the set of pixels in the region whose jump response value is greater than the boundary change reference value, aggregate and classify the spatial continuous regions, match them to the existing road surface damage template according to the region morphology and distribution characteristics, and generate a damage texture difference layer.

6. The intelligent identification and assessment method for highway pavement damage according to claim 5, characterized in that, The specific steps for obtaining the damage trend evolution label set are as follows: S411: Based on the damaged texture difference layer, extract the gradient direction distribution map and texture contour structure boundary map of the time node image, perform angle statistics on the texture boundary direction of the key area, and compare the regional concentration of the main direction change position in adjacent images to obtain the main direction trajectory of the damage deformation. S412: Call the main direction trajectory of the damage deformation, combine the ratio of the direction concentration degree in the target segment, the texture structure extension ratio and the disturbance amplitude of the neighborhood of the mutation point, calculate the trend offset dynamic index, and identify the offset evolution path in the image sequence based on the index direction turning distribution and dense area aggregation characteristics, and generate a damage trend evolution label set.

7. The intelligent identification and assessment method for highway pavement damage according to claim 1, characterized in that, The method also includes step S5: S5: Call the damage trend evolution label set, jointly determine the extent of contour expansion and gradient direction change in the current image region, adjust the image region classification boundary according to the matching degree of the label set, and obtain the road damage assessment level; The road surface damage assessment level includes structural expansion level, gradient change magnitude index, and matching degree determination factor.

8. The intelligent identification and assessment method for highway pavement damage according to claim 7, characterized in that, The specific steps for obtaining the road surface damage assessment level are as follows: S511: Call the damage trend evolution label set, and based on the contour expansion range value and gradient direction change value extracted from the image region, combine them to generate a contour gradient change comparison sequence, and sequentially calculate the matching degree with the reference combination value in the damage trend evolution label set to obtain the label matching difference. S512: Call the label matching difference, compare the contour gradient change sequence of the current boundary position, filter the boundary points with a difference coefficient less than the threshold according to the difference coefficient and the matching threshold, use them as adjustment reference points, and combine the coordinate relocation classification boundary to obtain the image boundary coordinate adjustment value; S513: Based on the image boundary coordinate adjustment value and the original classification boundary mapping value, extract the consistency degree of contour change rate and gradient direction in each region, match the damage trend evolution label set level standard, and obtain the road damage assessment level.

Images