Method and device for dynamic evaluation of asphalt pavement diseases by remote sensing

Abstract

The application discloses a kind of asphalt pavement disease remote sensing dynamic evaluation method and device. Including obtaining asphalt pavement RGB optical image, and divide into multiple equal length asphalt sections;Based on color and shape features, respectively, each asphalt section is analyzed by clustering, and homogeneous map spot is obtained;Excluding the interference of non-asphalt pavement map spot in the homogeneous map spot to obtain asphalt section map spot, the non-asphalt pavement map spot includes vegetation, street lamp, building shadow, map spot generated by driving vehicle interference;The brightness value of the asphalt section map spot is calculated, and brightness value Brightness is counted;Define suspected disease map spot by threshold classification method;Based on the suspected disease map spot, the section is evaluated.The application can quantitatively detect the degree of asphalt pavement disease, and the spatial and temporal distribution of disease is accurate.

Description

Technical Field

[0001] This invention generally relates to the field of image processing technology. More specifically, this invention relates to a method and apparatus for remote sensing dynamic assessment of asphalt pavement defects. Background Technology

[0002] With the acceleration of urbanization and the continuous expansion of road networks, the problems of road surface aging and damage are becoming increasingly prominent. Long-term natural factors (temperature, moisture, weathering) and vehicle loads have led to various road surface defects such as cracks, potholes, and network cracks, which seriously affect driving safety and the service life of roads, and there is an urgent need for effective detection and assessment methods.

[0003] Currently, commonly used remote sensing survey methods for road aging and damage are categorized into multispectral / hyperspectral remote sensing, thermal infrared remote sensing, and microwave remote sensing, depending on the data used. Multispectral / hyperspectral remote sensing monitoring generally employs image processing and recognition methods, brightness variation methods, and reflectance spectral modeling methods to establish the relationship between road aging and damage and the image spectrum. Existing technologies detect road damage from a pixel perspective, utilizing only the color features of the image and failing to directly assess geometric damage, thus limiting the applicability and scope of the methods. Liu Ye (2008) explored the relationship between hyperspectral remote sensing data (AVIRIS, HST) and the Road Condition Index (PCI), finding that the relationship between reflectance and PCI is non-linear and fluctuates greatly with changes in road condition. The model is only applicable to excellent and good road surfaces and cannot be extended to medium and poor road surfaces. Pan et al. (2016) used Multi-Endomembrane Hybrid Pixel Decomposition (MESMA) to classify road aging stages based on WorldView-2 satellite imagery. Jin Xu (2017), based on the blue and near-infrared bands of Worldview-2 satellite imagery, constructed four asphalt pavement health spectral indices by plotting spectral curves of different materials, aging levels, and damaged pavements, identifying characteristic bands, and generating pavement health index maps. These indices include the Difference Health Index (DHI), Ratio Health Index (RHI), Normalized Difference Health Index (NDHI), and Logarithmic Health Index (LHI). The essence of pavement health spectral indices is to transplant water body indices to asphalt pavement detection. Although both water bodies and asphalt pavements have low reflectivity, their mechanisms are not entirely the same. Furthermore, many publicly available online datasets, such as Google Earth and Tianditu, only have the RGB multispectral bands and lack the near-infrared band, greatly limiting the application of asphalt pavement health spectral indices. Summary of the Invention

[0004] To address at least some of the limitations of existing remote sensing methods for road surface distress surveys mentioned in the background, this invention utilizes readily available RGB three-band true-color high-resolution optical imagery from the internet. Based on patch segmentation, a deep learning model is used to eliminate interference from non-asphalt pavement factors. By constructing a road surface remote sensing fragmentation index and combining color and shape features, a threshold classification method is used to dynamically and quantitatively assess the condition of a large area of ​​road surface at different levels, from pixel to patch to road segment. The spatiotemporal distribution of distress is accurate, providing a new approach for low-cost, high-efficiency macroscopic surveys of road surface conditions. Therefore, this invention provides solutions in the following aspects.

[0005] The first aspect of this invention provides a remote sensing dynamic assessment method for asphalt pavement defects, comprising: acquiring an RGB optical image of an asphalt pavement; dividing the RGB optical image of the asphalt pavement into multiple asphalt road segments of equal length; performing cluster analysis on each of the asphalt road segments based on color and shape features to obtain homogeneous patches; excluding interference from non-asphalt pavement patches in the homogeneous patches to obtain asphalt road segment patches, wherein the non-asphalt pavement patches include patches generated by vegetation, streetlights, building shadows, and interference from vehicles; calculating the brightness value of the asphalt road segment patches and performing brightness value statistics, including the mean brightness, median brightness, standard deviation brightness, and 95th quantile brightness value; using a threshold classification method, detecting road patches whose asphalt pavement brightness value exceeds the 95th quantile or is 1.5 standard deviations higher than the average brightness, defining them as suspected defect patches; and assessing the road segments based on the suspected defect patches.

[0006] In one embodiment, the evaluation of road segments based on the suspected disease patches includes: calculating the suspected disease patch rate for each road segment, finding the 95th and 90th percentiles of the suspected disease patch rate; identifying road segments with a suspected disease patch rate exceeding the 95th percentile and defining them as poor-quality road segments; and monitoring road segments with a suspected disease patch rate exceeding the 90th percentile in the remaining road segments and defining them as medium-quality road segments.

[0007] In one embodiment, the assessment of road sections based on the suspected defect patches further includes: calculating the road surface remote sensing fragmentation index characteristics, and deriving them together with the remote sensing fragmentation index characteristics of the poor and medium-grade road sections and their corresponding suspected defect patches, providing decision support for further detailed road maintenance investigation and verification; the road surface remote sensing fragmentation index characteristics include: road surface damage area, which is the sum of the areas of defect patches in the road section, intuitively reflecting the severity of road surface defects; road surface damage rate = sum of defect patch areas / total road section area; road surface damage density = number of defect patches / total road section area, intuitively reflecting whether the road surface defects are scattered small cracks or concentrated large potholes.

[0008] In one embodiment, the clustering analysis of each asphalt pavement segment to obtain homogeneous patches includes: Step 1: Initialization, treating each pixel in the image as an independent initial object; Step 2: Calculating the merging cost of adjacent objects, traversing all adjacent object pairs in the image, and for each pair, calculating the total heterogeneity increase value f that would be generated if they were merged; the total heterogeneity increase value f is composed of a weighted average of color heterogeneity and shape heterogeneity: f = w_color h_color+(1-w_color) h_shape; h_color=Σw_c (n_merge σ_c_merge-(n_obj1 σ_c_obj1+ n_obj2 σ_c_obj2));h_shape=w_compact h_compact + (1 - w_compact) h_smooth; where w_color is the color weight, (1-w_color) is the shape weight; w_c is the weight of each band, with 1 for all three RGB bands; w_compact is the compactness weight, and (1-w_compact) is the smoothness weight; Step 3: Perform optimal merging, selecting the pair with the smallest f-value from all adjacent object pairs for merging; Step 4: Repeat steps 2 and 3. In each iteration, the f-values ​​of the newly generated object and its adjacent objects are recalculated. The entire segmentation process terminates when the smallest f-value among all adjacent object pairs is greater than the square of the scale parameter.

[0009] In one embodiment, the step of eliminating interference from non-asphalt pavement patches in the homogeneous patch to obtain asphalt pavement patches includes: marking a certain number of vegetation, shadows, vehicles, and road markings in the segmented asphalt pavement patches; generating training and testing sample sets; training the object convolutional neural network model OCNN using the training samples; verifying the model accuracy using the testing samples; obtaining the target model; inputting the asphalt pavement patches of each road segment into the target model to obtain predicted result patches; and excluding the four types of predicted result patches (vegetation, shadows, vehicles, and road markings) from the asphalt pavement patches.

[0010] A second aspect of the present invention provides a remote sensing dynamic assessment system for asphalt pavement defects, comprising: an image acquisition module for acquiring RGB optical images of asphalt pavement and dividing the RGB optical images into multiple asphalt road segments of equal length; a clustering analysis module for performing clustering analysis on each asphalt road segment based on color and shape features to obtain homogeneous patches; an interference removal module for removing interference from non-asphalt pavement patches in the homogeneous patches to obtain asphalt road segment patches, wherein the non-asphalt pavement patches include patches caused by vegetation, streetlights, building shadows, and interference from vehicles; a brightness calculation module for calculating the brightness values ​​of the asphalt road segment patches and performing brightness value statistics, including the mean brightness, median brightness, standard deviation brightness, and 95th percentile brightness value; a threshold classification module for detecting road patches whose asphalt pavement brightness values ​​exceed the 95th percentile or are 1.5 standard deviations higher than the average brightness using a threshold classification method, defining them as suspected defect patches; and an assessment module for assessing the road segments based on the suspected defect patches.

[0011] In one embodiment, the evaluation of road segments based on the suspected disease patches includes: calculating the suspected disease patch rate for each road segment, finding the 95th and 90th percentiles of the suspected disease patch rate; identifying road segments with a suspected disease patch rate exceeding the 95th percentile and defining them as poor-quality road segments; and monitoring road segments with a suspected disease patch rate exceeding the 90th percentile in the remaining road segments and defining them as medium-quality road segments.

[0012] In one embodiment, the assessment of road sections based on the suspected defect patches further includes: calculating the road surface remote sensing fragmentation index characteristics, and deriving them together with the remote sensing fragmentation index characteristics of the poor and medium-grade road sections and their corresponding suspected defect patches, providing decision support for further detailed road maintenance investigation and verification; the road surface remote sensing fragmentation index characteristics include: road surface damage area, which is the sum of the areas of defect patches in the road section, intuitively reflecting the severity of road surface defects; road surface damage rate = sum of defect patch areas / total road section area; road surface damage density = number of defect patches / total road section area, intuitively reflecting whether the road surface defects are scattered small cracks or concentrated large potholes.

[0013] In one embodiment, the clustering analysis of each asphalt pavement segment to obtain homogeneous patches includes: Step 1: Initialization, treating each pixel in the image as an independent initial object; Step 2: Calculating the merging cost of adjacent objects, traversing all adjacent object pairs in the image, and for each pair, calculating the total heterogeneity increase value f that would be generated if they were merged; the total heterogeneity increase value f is composed of a weighted average of color heterogeneity and shape heterogeneity: f = w_color h_color+(1-w_color) h_shape; h_color=Σw_c (n_merge σ_c_merge-(n_obj1 σ_c_obj1+ n_obj2 σ_c_obj2));h_shape=w_compact h_compact + (1 - w_compact) h_smooth; where w_color is the color weight, (1-w_color) is the shape weight; w_c is the weight of each band, with 1 for all three RGB bands; w_compact is the compactness weight, and (1-w_compact) is the smoothness weight; Step 3: Perform optimal merging, selecting the pair with the smallest f-value from all adjacent object pairs for merging; Step 4: Repeat steps 2 and 3. In each iteration, the f-values ​​of the newly generated object and its adjacent objects are recalculated. When the smallest f-value among all adjacent object pairs is greater than the square of the scale parameter, i.e., the threshold, the entire segmentation process terminates.

[0014] In one embodiment, the step of eliminating interference from non-asphalt pavement patches in the homogeneous patch to obtain asphalt pavement patches includes: marking a certain number of vegetation, shadows, vehicles, and road markings in the segmented asphalt pavement patches; generating training and testing sample sets; training the object convolutional neural network model OCNN using the training samples; verifying the model accuracy using the testing samples; obtaining the target model; inputting the asphalt pavement patches of each road segment into the target model to obtain predicted result patches; and excluding the four types of predicted result patches (vegetation, shadows, vehicles, and road markings) from the asphalt pavement patches.

[0015] This invention offers the following advantages: It utilizes publicly available RGB three-band true-color imagery, combining color and shape features to construct a road surface remote sensing fragmentation index. This quantitatively detects the degree of asphalt pavement distress at different levels—from pixel to patch to road segment—providing precise spatiotemporal distribution of distress and broad applicability, overcoming the limitations of existing technologies in terms of applicability and scope. By employing an object-based convolutional neural network model, interference from non-asphalt patches such as tree and building shadows, lampposts, and vehicles is eliminated, resulting in more accurate results. The physical meaning is clear, the operation process is simple, the operating efficiency is high, and the invention has strong universality. Attached Figure Description

[0016] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1 This is a flowchart illustrating the remote sensing dynamic assessment process for asphalt pavement defects; Figure 2 This is a schematic diagram showing the test results of suspected road defects; Figure 3 This shows the test results for suspected road defects; Figure 4 This shows the distribution of suspected road defects in suspected defect sections. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be understood that the terms "first," "second," "third," and "fourth," etc., in the claims, specification, and drawings of this invention are used to distinguish different objects, rather than to describe a specific order. The terms "comprising" and "including" used in the specification and claims of this invention indicate the presence of the described features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof. It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. As used in this specification and claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]." The first aspect of this invention provides a remote sensing dynamic assessment method for asphalt pavement distress. Specific embodiments of the invention are described in detail below with reference to the accompanying drawings. Figure 1 The diagram shown is a flowchart of the remote sensing dynamic assessment of asphalt pavement distress according to the present invention.

[0018] From the appendix Figure 1 As can be seen, the remote sensing dynamic assessment method for asphalt pavement distress of the present invention specifically includes the following steps.

[0019] S1. Data Acquisition and Preprocessing: Acquire high-resolution optical satellite imagery and road vector data of the work area, and perform necessary preprocessing on the satellite imagery and road vectors.

[0020] In one embodiment, if no road vector is available, the road centerline is digitized based on orthophoto satellite imagery.

[0021] In one embodiment, the above preprocessing includes: if the acquired optical satellite image is raw data, in order to eliminate radiation and geometric distortion caused by factors such as atmospheric conditions, solar elevation angle, satellite observation angle, and terrain, the remote sensing image needs to undergo preprocessing such as radiometric calibration, orthorectification, registration, fusion, color balancing, mosaicking, and projection conversion.

[0022] S2. Asphalt pavement generation and segmentation: The road centerline is divided into segments of approximately 200 meters in length. Using the road width as a parameter, a buffer zone operation is performed on the segmented road centerline to obtain the asphalt pavement.

[0023] In one embodiment, buffer analysis is a commonly used geographic proximity analysis method that creates a polygonal region at a specified distance around input features (points, lines, polygons). In this study area, which is a two-way four-lane road, buffer zones are created on both sides of the centerline of the two-way road. The buffer distance is set to 3.75m, the standard highway width, and the shape of the buffer zone ends is set to flat.

[0024] S3. Asphalt pavement segmentation: Based on color and shape features, cluster analysis is performed on each asphalt pavement segment to obtain homogeneous patches.

[0025] In one embodiment, the clustering analysis employs a bottom-up region merging algorithm. Its core logic is to minimize the increase in heterogeneity after each merge. The merging process stops when the minimum heterogeneity increase exceeds a preset scale parameter. Each time two adjacent objects are merged, the algorithm calculates the heterogeneity increase value *f*, i.e., the fusion value. This value is a weighted average of color heterogeneity and shape heterogeneity. f=w_color h_color+(1-w_color) h_shape In the formula, w_color is the color weight, and (1-w_color) is the shape weight.

[0026] Color heterogeneity (h_color) is the weighted standard deviation of the pixel spectral values ​​within the merged new object, minus the weighted standard deviation of each of the two objects before merging.

[0027] h_color=Σw_c (n_merge σ_c_merge-(n_obj1 σ_c_obj1+n_obj2 σ_c_obj2)) In the formula, w_c is the weight of each band (all three bands of RGB are 1).

[0028] Shape heterogeneity (h_shape) consists of two sub-criterions: compactness and smoothness. Smoothness heterogeneity (h_smooth) measures the smoothness of the merged object's boundary. It calculates the difference between the ratio of the object's actual perimeter to the perimeter of its smallest bounding rectangle. Compactness heterogeneity (h_compact) measures the compactness of the merged object. It calculates the difference between the ratio of the object's actual perimeter to the square root of its area.

[0029] h_shape=w_compact h_compact + (1 - w_compact) h_smooth In the formula, w_compact is the compactness weight, and (1-w_compact) is the smoothness weight.

[0030] In one embodiment, the parameters of the study area are set as follows: color 0.9, shape 0.1, compactness and smoothness 0.5 each, and scale 10. The specific steps are as follows: Initialization: The algorithm treats each pixel in the image as an independent initial object.

[0031] Calculate the cost of merging adjacent objects: The algorithm iterates through all pairs of adjacent objects in the image and, for each pair, calculates the total heterogeneity increase (f) that would result in the new object if they were merged.

[0032] Perform optimal merge: Select the pair with the smallest f value from all adjacent pairs and merge them.

[0033] Iteration and Termination: Repeat steps 2 and 3. In each iteration, the f-values ​​of the newly generated object and its neighbors are recalculated. The segmentation process terminates when the smallest f-value among all pairs of neighboring objects is greater than the square of the scale parameter (i.e., the threshold).

[0034] S4. To eliminate interference factors affecting the spectral signal, such as non-asphalt patches, vegetation, streetlights, building shadows, and moving vehicles, an object-based convolutional neural network model is used to detect and eliminate non-asphalt pavement patches such as shadows, road dividing lines, and moving vehicles.

[0035] In one embodiment, the above-mentioned exclusion of non-asphalt pavement patches includes: marking a certain number of vegetation, shadows, vehicles, and road markings in the segmented asphalt pavement patches, generating training samples and test sample sets, training the object convolutional neural network model OCNN with the training samples, verifying the model accuracy with the test samples, and obtaining the target model.

[0036] Input the asphalt pavement patch of each road segment into the target model to obtain the predicted patch.

[0037] The four types of predicted result patches—vegetation, shadow, vehicles, and road markings—are excluded from the asphalt pavement patches.

[0038] S5. Brightness Feature Statistics: Brightness values ​​of asphalt pavement patches are statistically analyzed, including the mean brightness, median brightness, stddev brightness variance, and 95th quantile brightness value. For three-band true-color imagery, the patch brightness value is the mean reflectance of the three RGB bands, calculated using the following formula: brightness=(ρBlue+ρGreen+ρRed) / 3 Among them, ρBlue, ρGreen, and ρRed are the average reflectance values ​​of the blue, green, and red bands of each pixel within the image patch, respectively.

[0039] The combination of mean or median and variance is calculated (e.g., mean + 1.5Stddev, median + 2Stddev). The 95th percentile is the threshold used for the following threshold classification, and any one of them can be chosen.

[0040] S6. Detection of suspected defects: Road patches whose brightness values ​​exceed the 95th percentile or are 1.5 standard deviations higher than the average brightness are identified using a threshold classification method and defined as suspected defects (SD).

[0041] In one embodiment, if brightness > 95th percentile or brightness > mean + 1.5Stddev, it is determined as a suspected disease patch (SD).

[0042] S7, Inspection of suspected road sections with defects S7.1 Characteristics of remote sensing fragmentation index of road surface: In this invention, the index feature is an indicator for measuring the degree of road segment damage. It is a measurement of the results obtained in step S6 on a segment-by-segment basis, while the threshold in S6 is the basis for classifying suspected damaged patches based on patch brightness. Specifically, it includes: Paved Damage Area (PDA) is the sum of the areas of damaged patches on a road section, and it directly reflects the severity of road damage.

[0043] Paved Damage Percentage (PDP) = Sum of damaged patch areas / Total road section area.

[0044] Paved Damage Density (PDD) = Number of damaged patches / Total area of ​​the road segment. It visually reflects whether pavement damage consists of scattered small cracks or concentrated large potholes.

[0045] S7.2 Calculate the suspected defect patch rate for each road segment separately, and find the 95% and 90% quantiles of the suspected defect patch rate.

[0046] The S7.3 threshold classification method defines road sections with a suspected damage rate exceeding the 95th percentile as poor road sections (suspected damaged 1); and road sections with a suspected damage rate exceeding the 90th percentile in other road sections are defined as medium road sections (suspected damaged 2).

[0047] S8. Results Export: Export the poor and medium grade road sections and their corresponding suspected disease patches along with their remote sensing fragmentation index characteristics to provide decision support for further detailed road maintenance investigation and verification.

[0048] In this invention, the implementation logic at different levels of "pixel-map-road segment" is as follows: Pixels are clustered into homogeneous patches through cluster analysis, providing raw materials for subsequent elimination of interfering patch elements and detection of suspected defect patches. The severity of defects is measured based on remote sensing fragmentation index characteristics, on a segment-by-segment basis, consistent with current road maintenance and management standards.

[0049] In one specific embodiment, a section of a highway was selected as the work area, using 0.5-meter resolution RGB three-band true-color imagery from April 2024. The detection results for suspected road defects are shown below. Figure 2 , Figure 2 Red indicates poor quality, green indicates medium quality; and the results of the detection of suspected road defects are shown in the 0.5-meter resolution RGB three-band true-color image from September 2024. Figure 3 , Figure 3 Medium red indicates poor condition, and green indicates medium condition. Using the road surface remote sensing fragmentation index characteristics, a threshold classification method is employed to dynamically assess the degree and spatiotemporal distribution of asphalt pavement damage at different levels from "pixel-pattern-road segment," as detailed below.

[0050] S1 Data Acquisition and Preprocessing: Acquire high-resolution optical satellite imagery and road vector data, and perform necessary preprocessing. If road vectors are unavailable, digitize road centerlines based on satellite imagery.

[0051] The S2 asphalt pavement generation and segmentation process involves dividing the 45-kilometer-long, two-lane Tuhe Expressway into 400 segments of 226 meters each, using the road centerline as a reference. A buffer zone operation is then performed on the centerline of each segment, using a road width of 3.75 meters as a parameter, to obtain the asphalt pavement.

[0052] S3 asphalt pavement patch segmentation: Based on color and shape features, cluster analysis is performed on each road segment to obtain homogeneous patches.

[0053] S4 excludes non-asphalt patches, vegetation, streetlights, building shadows, and moving vehicles that affect the spectral signal. It uses an object-based convolutional neural network model to detect and exclude non-asphalt pavement patches such as shadows, road dividing lines, and moving vehicles.

[0054] S5 Feature Statistics: Brightness statistics for asphalt pavement patches, including mean brightness, median brightness, stddev brightness variance, and 95th quantile brightness value. For three-band true-color imagery, the patch brightness value is the mean reflectance of the blue, green, and red bands, calculated using the following formula: in , and These represent the average reflectance values ​​of each pixel within the image patch for the blue, green, and red bands, respectively.

[0055] S6 Suspected Defect Patch Detection: Using a threshold classification method, road patches with asphalt pavement brightness values ​​exceeding the 95th quantile are defined as suspected defect patches (SD).

[0056] S7 suspected road defect detection S7.1 Characteristics of remote sensing fragmentation index of road surface: Paved Damage Area (PDA) is the sum of the areas of damaged patches on a road section, and it directly reflects the severity of road damage.

[0057] Paved Damage Percentage (PDP) = Sum of damaged patch areas / Total road section area.

[0058] Paved Damage Density (PDD) = Number of damaged patches / Total area of ​​the road segment. It visually reflects whether pavement damage consists of scattered small cracks or concentrated large potholes.

[0059] S7.2 Calculate the suspected defect rate for each road segment separately, and find the 95% and 90% quantiles of the suspected defect rate.

[0060] S7.3 Threshold classification method: road sections with a suspected damage patch rate exceeding the 95th percentile are defined as poor road sections (suspected damaged 1); road sections with a suspected damage patch rate exceeding the 90th percentile in other road sections are defined as medium road sections (suspected damaged 2).

[0061] S8 results export includes data on poor and medium grade road sections, along with corresponding pavement damage area, damage rate, and damage density, providing decision support for further detailed road maintenance investigations and design. The distribution of suspected road defects in suspected defect sections is shown in the image. Figure 4 .

[0062] A second aspect of the present invention provides a remote sensing dynamic assessment system for asphalt pavement defects, comprising: an image acquisition module for acquiring RGB optical images of asphalt pavement and dividing the RGB optical images into multiple asphalt road segments of equal length; a clustering analysis module for performing clustering analysis on each asphalt road segment based on color and shape features to obtain homogeneous patches; an interference removal module for removing interference from non-asphalt pavement patches in the homogeneous patches to obtain asphalt road segment patches, wherein the non-asphalt pavement patches include patches caused by vegetation, streetlights, building shadows, and interference from vehicles; a brightness calculation module for calculating the brightness values ​​of the asphalt road segment patches and performing brightness value statistics, including the mean brightness, median brightness, standard deviation brightness, and 95th percentile brightness value; a threshold classification module for detecting road patches whose asphalt pavement brightness values ​​exceed the 95th percentile or are 1.5 standard deviations higher than the average brightness using a threshold classification method, defining them as suspected defect patches; and an assessment module for assessing the road segments based on the suspected defect patches.

[0063] In one embodiment, the evaluation of road segments based on the suspected disease patches includes: calculating the suspected disease patch rate for each road segment, finding the 95th and 90th percentiles of the suspected disease patch rate; identifying road segments with a suspected disease patch rate exceeding the 95th percentile and defining them as poor-quality road segments; and monitoring road segments with a suspected disease patch rate exceeding the 90th percentile in the remaining road segments and defining them as medium-quality road segments.

[0064] In one embodiment, the assessment of road sections based on the suspected defect patches further includes: calculating the road surface remote sensing fragmentation index characteristics, and deriving them together with the remote sensing fragmentation index characteristics of the poor and medium-grade road sections and their corresponding suspected defect patches, providing decision support for further detailed road maintenance investigation and verification; the road surface remote sensing fragmentation index characteristics include: road surface damage area, which is the sum of the areas of defect patches in the road section, intuitively reflecting the severity of road surface defects; road surface damage rate = sum of defect patch areas / total road section area; road surface damage density = number of defect patches / total road section area, intuitively reflecting whether the road surface defects are scattered small cracks or concentrated large potholes.

[0065] In one embodiment, the clustering analysis of each asphalt pavement segment to obtain homogeneous patches includes: Step 1: Initialization, treating each pixel in the image as an independent initial object; Step 2: Calculating the merging cost of adjacent objects, traversing all adjacent object pairs in the image, and for each pair, calculating the total heterogeneity increase value f that would be generated if they were merged; the total heterogeneity increase value f is composed of a weighted average of color heterogeneity and shape heterogeneity: f = w_color h_color+(1-w_color) h_shape; h_color=Σw_c (n_merge σ_c_merge-(n_obj1 σ_c_obj1+ n_obj2 σ_c_obj2));h_shape=w_compact h_compact + (1 - w_compact) h_smooth; where w_color is the color weight, (1-w_color) is the shape weight; w_c is the weight of each band, with 1 for all three RGB bands; w_compact is the compactness weight, and (1-w_compact) is the smoothness weight; Step 3: Perform optimal merging, selecting the pair with the smallest f-value from all adjacent object pairs for merging; Step 4: Repeat steps 2 and 3. In each iteration, the f-values ​​of the newly generated object and its adjacent objects are recalculated. When the smallest f-value among all adjacent object pairs is greater than the square of the scale parameter, i.e., the threshold, the entire segmentation process terminates.

[0066] In one embodiment, the step of eliminating interference from non-asphalt pavement patches in the homogeneous patch to obtain asphalt pavement patches includes: marking a certain number of vegetation, shadows, vehicles, and road markings in the segmented asphalt pavement patches; generating training and testing sample sets; training the object convolutional neural network model OCNN using the training samples; verifying the model accuracy using the testing samples; obtaining the target model; inputting the asphalt pavement patches of each road segment into the target model to obtain predicted result patches; and excluding the four types of predicted result patches (vegetation, shadows, vehicles, and road markings) from the asphalt pavement patches.

[0067] While various embodiments of the invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and intent of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of the invention. The appended claims are intended to define the scope of protection of the invention and therefore cover modular compositions, equivalents, or alternatives within the scope of these claims.

Claims

1. A remote sensing dynamic assessment method for asphalt pavement distress, characterized in that, include: Acquire RGB optical images of asphalt pavement and divide the RGB optical images of asphalt pavement into multiple asphalt road segments of equal length; Based on color and shape features, cluster analysis was performed on each of the asphalt road segments to obtain homogeneous patches. The asphalt road section patch is obtained by excluding the interference of non-asphalt pavement patches in the homogeneous patch. The non-asphalt pavement patches include patches caused by vegetation, street lights, building shadows, and interference from driving vehicles. The brightness values ​​of the asphalt road sections are calculated, and the brightness values ​​are statistically analyzed, including the mean brightness, median brightness, standard deviation brightness, and 95th quantile brightness value. Using a threshold classification method, road patches whose asphalt pavement brightness values ​​exceed the 95th percentile or are 1.5 standard deviations higher than the average brightness are defined as suspected defects. The road section was assessed based on the suspected defect patches.

2. The remote sensing dynamic assessment method for asphalt pavement distress according to claim 1, characterized in that, The assessment of road sections based on the suspected defect patches includes: The rate of suspected defects was calculated for each road segment, and the 95th and 90th percentiles of the suspected defect rate were identified. Road sections with a suspected disease patch rate exceeding the 95th percentile are defined as poor road sections; road sections with a suspected disease patch rate exceeding the 90th percentile in other road sections are defined as medium road sections.

3. The remote sensing dynamic assessment method for asphalt pavement distress according to claim 2, characterized in that, The assessment of road sections based on the suspected road defect patches also includes: The remote sensing fragmentation index characteristics of the road surface are calculated, and the characteristics of the remote sensing fragmentation index are derived together with the poor and medium grade road sections and the corresponding suspected defects, so as to provide decision support for further detailed road maintenance investigation and verification. The road surface remote sensing fragmentation index features include: The damaged area of ​​the road surface is the sum of the areas of the damaged patches in the road section, which directly reflects the severity of the road surface damage. Road surface damage rate = Sum of damaged patch areas / Total road section area; Road surface damage density = number of damaged patches / total area of ​​road section, which directly reflects whether the road surface damage is scattered small cracks or concentrated large potholes.

4. The remote sensing dynamic assessment method for asphalt pavement distress according to any one of claims 1-3, characterized in that, The cluster analysis performed on each asphalt pavement segment yields homogeneous patches, including: Step 1: Initialization, treating each pixel in the image as an independent initial object; Step 2: Calculate the cost of merging adjacent objects. Traverse all pairs of adjacent objects in the image. For each pair, calculate the total heterogeneity increase f that would be generated if they were merged into a new object. The total heterogeneity growth value f is composed of a weighted average of color heterogeneity and shape heterogeneity. f=w_color h_color+(1-w_color) h_shape; h_color=Σw_c (n_merge σ_c_merge-(n_obj1 σ_c_obj1+ n_obj2 σ_c_obj2)); h_shape=w_compact h_compact+(1-w_compact) h_smooth; Where w_color is the color weight, (1-w_color) is the shape weight; w_c is the weight of each band, with 1 for each of the three RGB bands; w_compact is the compactness weight, and (1-w_compact) is the smoothness weight. Step 3: Perform optimal merge, selecting the pair with the smallest f value from all adjacent object pairs for merging; Step 4: Repeat steps 2 and 3. In each iteration, the f-values ​​of the newly generated object and its neighboring objects will be recalculated. The entire segmentation process terminates when the smallest f-value among all adjacent object pairs is greater than the square of the scale parameter, i.e., the threshold.

5. The remote sensing dynamic assessment method for asphalt pavement distress according to claim 4, characterized in that, The process of eliminating interference from non-asphalt pavement patches in the homogeneous patch to obtain asphalt pavement patches includes: In the segmented asphalt pavement patches, a certain number of vegetation, shadows, vehicles, and road markings are marked to generate training and test sample sets. The training samples are used to train the object convolutional neural network model OCNN, and the test samples are used to verify the model accuracy to obtain the target model. Input the asphalt pavement patch of each road segment into the target model to obtain the predicted patch; The four types of predicted result patches—vegetation, shadow, vehicles, and road markings—are excluded from the asphalt pavement patches.

6. A remote sensing dynamic assessment system for asphalt pavement defects, characterized in that, include: The image acquisition module is used to acquire RGB optical images of asphalt pavement and divide the RGB optical images of asphalt pavement into multiple asphalt road segments of equal length. The clustering analysis module is used to perform clustering analysis on each of the asphalt road segments based on color and shape features to obtain homogeneous patches; The interference elimination module is used to eliminate interference from non-asphalt pavement patches in the homogeneous patch to obtain asphalt pavement patches. The non-asphalt pavement patches include patches caused by vegetation, streetlights, building shadows, and interference from vehicles. The brightness calculation module calculates the brightness value of the asphalt road section patches and performs brightness value statistics, including the mean brightness, median brightness, standard deviation brightness, and 95th percentile brightness value. The threshold classification module is used to detect road patches whose brightness values ​​exceed the 95th percentile or are 1.5 standard deviations higher than the average brightness using a threshold classification method, and define them as suspected defects. The assessment module evaluates the road section based on the suspected defect patches.

7. The remote sensing dynamic assessment method for asphalt pavement distress according to claim 6, characterized in that, The assessment of road sections based on the suspected defect patches includes: The rate of suspected defects was calculated for each road segment, and the 95th and 90th percentiles of the suspected defect rate were identified. Road sections with a suspected disease patch rate exceeding the 95th percentile are defined as poor road sections; road sections with a suspected disease patch rate exceeding the 90th percentile in other road sections are defined as medium road sections.

8. The remote sensing dynamic assessment method for asphalt pavement distress according to claim 7, characterized in that, The assessment of road sections based on the suspected road defect patches also includes: The remote sensing fragmentation index characteristics of the road surface are calculated, and the characteristics of the remote sensing fragmentation index are derived together with the poor and medium grade road sections and the corresponding suspected defects, so as to provide decision support for further detailed road maintenance investigation and verification. The road surface remote sensing fragmentation index features include: The damaged area of ​​the road surface is the sum of the areas of the damaged patches in the road section, which directly reflects the severity of the road surface damage. Road surface damage rate = Sum of damaged patch areas / Total road section area; Road surface damage density = number of damaged patches / total area of ​​road section, which directly reflects whether the road surface damage is scattered small cracks or concentrated large potholes.

9. The remote sensing dynamic assessment method for asphalt pavement distress according to any one of claims 6-8, characterized in that, The cluster analysis performed on each asphalt pavement segment yields homogeneous patches, including: Step 1: Initialization, treating each pixel in the image as an independent initial object; Step 2: Calculate the cost of merging adjacent objects. Traverse all pairs of adjacent objects in the image. For each pair, calculate the total heterogeneity increase f that would be generated if they were merged into a new object. The total heterogeneity growth value f is composed of a weighted average of color heterogeneity and shape heterogeneity. f=w_color h_color+(1-w_color) h_shape; h_color=Σw_c (n_merge σ_c_merge-(n_obj1 σ_c_obj1+ n_obj2 σ_c_obj2)); h_shape=w_compact h_compact+(1-w_compact) h_smooth; Where w_color is the color weight, (1-w_color) is the shape weight; w_c is the weight of each band, with 1 for each of the three RGB bands; w_compact is the compactness weight, and (1-w_compact) is the smoothness weight. Step 3: Perform optimal merge, selecting the pair with the smallest f value from all adjacent object pairs for merging; Step 4: Repeat steps 2 and 3. In each iteration, the f-values ​​of the newly generated object and its neighboring objects will be recalculated. The entire segmentation process terminates when the smallest f-value among all adjacent object pairs is greater than the square of the scale parameter, i.e., the threshold.

10. The remote sensing dynamic assessment method for asphalt pavement distress according to claim 9, characterized in that, The process of eliminating interference from non-asphalt pavement patches in the homogeneous patch to obtain asphalt pavement patches includes: In the segmented asphalt pavement patches, a certain number of vegetation, shadows, vehicles, and road markings are marked to generate training and test sample sets. The training samples are used to train the object convolutional neural network model OCNN, and the test samples are used to verify the model accuracy to obtain the target model. Input the asphalt pavement patch of each road segment into the target model to obtain the predicted patch; The four types of predicted result patches—vegetation, shadow, vehicles, and road markings—are excluded from the asphalt pavement patches.

Images