An asphalt pavement disease identification method based on unmanned aerial vehicle scanning

Abstract

This invention discloses a method for identifying asphalt pavement defects based on UAV scanning, relating to the field of asphalt pavement defect identification. The method includes: using a UAV equipped with a multispectral imager to acquire orthophotos of the entire road section via a preset flight path for preliminary feature extraction and identification of suspected areas; then, for suspected areas, intelligently planning a re-examination flight path, optimizing the flight path according to the degree of suspicion and spatial distribution, and collecting specific band spectral reflectance data for secondary verification; explicitly extracting color features, gloss features, and texture features from the target asphalt pavement area; and quantifying the physical features related to the defects through methods such as brightness and saturation distribution, highlight area detection, and texture structure analysis, thereby improving the accuracy and reliability of bleeding defect identification and generating a defect severity level, providing accurate data support for asphalt pavement maintenance decisions.

Description

Technical Field

[0001] This invention belongs to the field of asphalt pavement distress identification technology, and relates to a method for asphalt pavement distress identification based on UAV scanning. Background Technology

[0002] Asphalt pavement, as a widely used pavement structure in current highway construction, is susceptible to various diseases during long-term service due to repeated traffic loads, changes in ambient temperature, and rainwater erosion. Among these, bleeding is a typical and serious type of disease.

[0003] Bleeding in asphalt pavements refers to the phenomenon where asphalt binder within the pavement gradually migrates upwards and accumulates on the surface under conditions of increased temperature and compaction, resulting in an asphalt content on the surface exceeding the normal design range. This condition causes the pavement surface to exhibit an abnormal gloss and significantly reduces its skid resistance, especially in rainy weather, making it prone to causing vehicle skidding and directly threatening driving safety. Therefore, identifying this condition is crucial.

[0004] Currently, existing technologies have proposed methods for identifying asphalt pavement defects. For example, the invention patent with publication number CN112489026A proposes an asphalt pavement defect detection method based on a multi-branch parallel convolutional neural network. This method collects a large number of pavement images and constructs a dataset. Each image is labeled with the defect bounding box location, defect type category, and pixel-level segmentation mask. A multi-branch parallel convolutional neural network structure is constructed, with three branches connected in parallel. During training, a gradient descent algorithm is used. By calculating the deviation between the output of each branch and the true label, the weights and bias parameters of each layer in the network are jointly optimized using the backpropagation mechanism until the model converges. This method can simultaneously complete defect localization, classification, and pixel-level segmentation tasks, improving detection accuracy and efficiency.

[0005] However, although the existing solutions have achieved certain results in the identification of asphalt pavement defects, they still have the following shortcomings: First, the initial images are easily affected by environmental factors such as lighting, rain, and pavement stains, resulting in unclear defect features. Existing technologies only explicitly use pavement images as the data source and rely solely on an initial single batch of images to construct the dataset. They perform a single collection process for areas with potentially blurred features, lacking targeted data supplementation for suspected areas. This makes it impossible for the dataset to fully cover the diverse manifestations of defects, affecting the sample quality of model training and ultimately leading to missed detections or misjudgments in the detection results.

[0006] Secondly, existing technologies rely solely on network structures such as convolutional layers and pooling layers to automatically learn image features through data-driven methods. They do not explicitly define, extract, and quantify the core features required for asphalt pavement defect identification. Instead, they depend entirely on the network to autonomously mine feature associations from massive images, resulting in the feature learning process being a black box. When faced with insufficient image data, uneven data distribution, or interference scenarios such as temporary stains or sudden changes in lighting on the road surface, the model struggles to accurately identify key defect features with physical significance, reducing the relevance and accuracy of feature learning. Summary of the Invention

[0007] In view of this, in order to solve the problems mentioned in the background art, the present invention provides a method for identifying asphalt pavement defects based on unmanned aerial vehicle (UAV) scanning.

[0008] The objective of this invention can be achieved through the following technical solution: a method for identifying asphalt pavement defects based on UAV scanning, comprising: S1, using a UAV equipped with a multispectral imager to fly along a preset route to collect continuous images of the target asphalt pavement and generate a continuous orthophoto map of the pavement covering the entire road section.

[0009] S2. Extract the color features, gloss features, and texture features of the target asphalt pavement area from the orthophoto image.

[0010] S3. Establish an oil bloom disease identification model based on deep learning. Based on color features, gloss features, and texture features, the oil bloom disease identification model outputs preliminary suspected oil bloom disease areas and their location information, forming a set of suspected areas.

[0011] S4. Based on the set of suspected areas, plan the drone re-inspection route and control the drone to fly over each suspected area to collect spectral reflectance data of specific bands in the corresponding area.

[0012] S5. Screen suspected areas based on spectral reflectance data of specific bands to confirm the oil spill area.

[0013] S6. Calculate the effective area of ​​the bleeding zone, and integrate the location information, severity level, and effective area statistics to generate the severity level of bleeding damage on asphalt pavement.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention uses a drone equipped with a multispectral imager to obtain orthophotos of the entire road section through a preset route for preliminary feature extraction and identification of suspected areas. Then, for suspected areas, intelligent planning of re-inspection routes is used to optimize the flight path according to the degree of suspicion and spatial distribution, and specific band spectral reflectance data is collected for secondary verification. This effectively solves the problem of feature blurring caused by environmental interference such as light, rain, and stains in the initial image. By specifically supplementing the data, the accuracy and robustness of disease identification are improved, and missed detection and misjudgment are avoided.

[0015] (2) This invention explicitly extracts color features, gloss features and texture features for the target asphalt pavement area. It quantitatively extracts the physical features related to the disease through means such as brightness and saturation distribution, highlight area detection and texture structure analysis. This avoids the limitations of relying entirely on black box neural networks to autonomously mine features and improves the ability to stably identify key disease features when the amount of data is insufficient or there is interference. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a diagram illustrating the implementation steps of the method of the present invention.

[0018] Figure 2 This is a flowchart of the UAV re-inspection route planning process of the present invention.

[0019] Figure 3 This is a flowchart of the spectral identification process for oil seepage areas on the road surface according to the present invention. Detailed Implementation

[0020] 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 some embodiments of the present invention, and not all embodiments. 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.

[0021] Please see Figure 1 As shown, the present invention provides a method for identifying asphalt pavement defects based on UAV scanning, including: S1, using a UAV equipped with a multispectral imager to fly along a preset route to collect continuous images of the target asphalt pavement and generate a continuous orthophoto map of the pavement covering the entire road section.

[0022] It should be noted that multispectral imagers can provide multidimensional spectral information that ordinary optical cameras cannot cover. Ordinary cameras can only capture color information in the visible light band, while multispectral imagers can cover multiple bands such as visible light and near infrared, and can better separate brightness saturation and capture pixels with strong reflection. On the other hand, the high-resolution spectral data of multispectral imagers can also reduce the interference of lighting changes on texture features, ensuring the accuracy of subsequent texture structure recognition.

[0023] A specific step in obtaining a preset flight path is as follows: First, it is necessary to collect the vector boundaries of the target asphalt pavement, such as the road centerline, pavement width, coordinate system, and on-site survey data, such as the location of obstacles around the road.

[0024] Based on the hardware parameters of the drone and the multispectral imager: the flight altitude is calculated from the imager's focal length and sensor size, and the flight speed is determined from the imager's frame rate.

[0025] The drone's flight path is parallel to the road centerline to ensure full coverage of the road surface without any edges being missed. Based on obstacle data from on-site surveys, the flight path's turning points are manually adjusted, such as avoiding utility poles, and local flight path altitudes are fine-tuned. Takeoff and landing points are also marked.

[0026] A specific implementation step for generating a continuous orthophoto map of the entire road surface is as follows: perform preprocessing on the continuous images collected by the UAV to remove lens distortion correction, radiometric correction and noise reduction.

[0027] Pre-set ground control points in the target road section, such as known coordinate marks on the road surface, obtain the exterior orientation elements of each image, and project the pixels of each image from the image coordinate system to the target road surface plane coordinate system through bundle adjustment.

[0028] Based on the feature points of the overlapping areas of adjacent images identified by the corrected discrete images, the pixel values ​​of the overlapping areas are smoothed, and the corrected discrete images are stitched together as a whole.

[0029] Based on the vector boundary of the target road segment, the overall image after cropping and stitching is removed, and non-target areas outside the road surface are removed, finally obtaining an orthophoto map that only covers the target asphalt road surface.

[0030] S2. Extract the color features, gloss features, and texture features of the target asphalt pavement area from the orthophoto image.

[0031] The color feature extraction steps are as follows: preprocess the orthophoto image to separate the target asphalt pavement area from the non-pavement area, and obtain the image area containing only the target asphalt pavement.

[0032] Specifically, the continuous orthophoto map of the road surface covering the entire road section is divided into regions. The image map is divided into several basic analysis units of equal size. The channel parameters of each basic analysis unit in the preset color space are extracted, including the lightness channel and two chroma channels. Basic analysis units whose lightness channel parameters are within the road surface area determination threshold range are selected and marked as candidate road surface units. Adjacent candidate road surface units are merged by connectivity analysis to form a continuous region. Continuous regions with an area smaller than the preset minimum area threshold are removed. The retained continuous region is the image region of the target asphalt road surface.

[0033] Convert the image of the target asphalt pavement area to a color space that includes brightness and saturation parameters.

[0034] The image pixels of the target asphalt pavement area are traversed, and the brightness and saturation values ​​of each pixel are extracted.

[0035] Specifically, each pixel in the target asphalt pavement area image is accessed sequentially in a preset order, such as from left to right or from top to bottom, using a line-by-line scanning method. For each pixel traversed, its corresponding brightness parameter channel value, such as the V value of the HSV space, and saturation parameter channel value, such as the S value of the HSV space, are directly read from the color space after conversion in the second step.

[0036] The extracted brightness and saturation values ​​are divided into regions, and pixels with similar brightness and saturation values ​​are grouped into the same feature region.

[0037] The intervals formed by the minimum and maximum values ​​of all pixel brightness and saturation values ​​within each feature region are respectively denoted as the brightness distribution range and the saturation distribution range.

[0038] The brightness and saturation distribution ranges of each feature region are statistically analyzed to form the color characteristics of that feature region.

[0039] Bleeding areas typically have a significantly higher brightness value than normal pavement due to the enrichment of asphalt components, and their saturation value also falls within a specific range. By extracting color features, the unique color patterns of bleeding can be transformed into quantitative data that the model can recognize. This allows the model to locate potential areas that match the color characteristics of bleeding, thus completing the initial screening of suspected areas.

[0040] The above gloss feature extraction steps are as follows: Select a local area of ​​a preset size from the separated target asphalt pavement area as the analysis unit.

[0041] It should be noted that the preset size should be based on the smallest contiguous area that can completely cover the typical strong reflective area of ​​asphalt pavement bleeding defects, so as to ensure that a single test can capture the complete shape and boundary transition characteristics of the effective gloss area.

[0042] Highlight region detection is performed on each analysis unit to identify sets of pixels with strong reflective properties within the unit, forming highlight candidate regions.

[0043] It should be noted that normal asphalt pavement, due to the exposed aggregate particles and rough surface, mainly reflects light diffusely, with weak reflection intensity and uniform distribution without obvious strong reflection areas. However, the surface of the asphalt-rich and relatively smooth asphalt pavement mainly reflects light specularly, which will form local strong reflection areas.

[0044] If the difference in reflectance characteristics between adjacent pixels within the highlight candidate area is within a preset range, and its boundary exhibits a transitional shape that gradually weakens from the center to the edge, then it is marked as an effective gloss area.

[0045] It should be noted that the specular reflection of the oil-bleeded area is affected by the angle of light incidence and the road surface slope. Its strong reflection intensity exhibits a natural transition pattern, being stronger at the center and weaker at the edges, eventually merging with the diffuse reflection characteristics of the normal road surface to form a gradual transition boundary. In contrast, the reflective boundary of interfering highlights is steep and abrupt. Therefore, the boundary exhibiting a gradual weakening from the center to the edge is a unique morphological characteristic of oil-bleeded highlight areas, which can further eliminate non-oil-bleed interference and ensure that the marked effective gloss area completely corresponds to the oil-bleeding defect.

[0046] The preset range of the difference in reflectance characteristics between adjacent pixels was obtained through statistical analysis and verification of oily samples.

[0047] The proportion of the effective gloss area in the corresponding analysis cell is used as the gloss coverage characteristic of that cell.

[0048] Analyze the brightness gradient changes of pixels within the effective gloss area, and determine the gloss diffusion range characteristics based on the smoothness of the gradient changes.

[0049] Specifically, the pixel coordinate range is cropped, retaining only pixels within the effective gloss area. For each pixel in the cropped area, an 8-neighbor traversal method is used, which involves traversing the eight neighboring pixels above, below, left, right, and diagonally. The brightness difference between the target pixel and each neighboring pixel is calculated. The ratio of the brightness difference to the ground sampling distance of the orthophoto is used as the brightness gradient value per unit physical distance. The arithmetic mean of the eight neighboring gradient values ​​of each pixel is taken as the final gradient value of that pixel, and a gradient heatmap of the effective gloss area is generated.

[0050] Based on the training data of oily samples, the gradient values ​​of several groups of real oily effective gloss areas were statistically analyzed to determine the range of gradient value sets for 95% of the samples. A smoothness judgment threshold was set, and all pixels within the effective gloss area were traversed. If the final gradient value of a pixel is less than or equal to the smoothness judgment threshold, it is marked as a smooth pixel; if the final gradient value is greater than the smoothness judgment threshold, it is marked as a steep pixel. A region growing method was used to perform connectivity analysis on all smooth pixels, retaining only continuous smooth pixel clusters and excluding isolated smooth pixels.

[0051] By integrating the gloss coverage characteristics and gloss diffusion range characteristics of each analysis unit, the gloss characteristics of the target asphalt pavement area are formed.

[0052] Normal asphalt pavement has exposed aggregates and a rough surface, so the reflection is mainly diffuse and lacks gloss. Adding gloss features ensures the accuracy of identifying suspicious areas.

[0053] The above texture feature extraction steps are as follows: Traverse the pixels in the target asphalt pavement area, and mark the pixels with gray values ​​of adjacent pixels as potential texture structures if the variation range of gray values ​​is within a preset stable range and shows a directional regularity.

[0054] It should be noted that the aggregate particle size and distribution density of normal road surfaces are relatively uniform, and the road surface material transitions smoothly between adjacent pixels. Therefore, the grayscale value changes of adjacent pixels will not be abrupt, and the change range will remain within a stable range. In contrast, the grayscale changes of pixels without texture interference are irregular, and the range often exceeds this stable range.

[0055] The preset stability range is determined by statistical data on the variation of gray values ​​of adjacent pixels in a large number of normal asphalt pavement samples without bleeding defects, combined with the results of the consistency test between the validation sample set and the independent sample set.

[0056] If the extension direction deviation between adjacent texture segments of a potential texture structure exceeds a preset angle range, it is determined that there is an interruption and its start and end coordinates and the straight-line distance between the two texture segments are recorded as the interval distance.

[0057] It should be noted that the texture of normal asphalt pavement is continuous due to the construction process. The extension direction of adjacent texture segments will be consistent or slightly deviated, and the spacing will remain within a stable range without obvious abrupt changes. However, bleeding disease will cause asphalt to cover aggregate particles, destroy the continuous structure of the original texture, and cause the adjacent texture segments to show a large deviation in extension direction or a significant increase in spacing.

[0058] The preset angle range was determined by statistical data on the extension direction deviation of normal continuous texture segments in a large number of normal asphalt pavement samples without bleeding defects, combined with the results of the consistency test between the verification sample set and the independent sample set.

[0059] The number of textures whose main extension direction falls into the corresponding interval is counted, and the proportion of textures in each interval to the total number of textures is calculated to form texture direction features.

[0060] Divide the target asphalt pavement area into several equal-area sub-regions, count the total number of texture structures in each sub-region, and form texture density characteristics by comparing the differences in the number of different sub-regions.

[0061] Calculate the variance between the texture orientation eigenvalues ​​of all texture structures, and use the reciprocal of this variance as the texture uniformity feature.

[0062] It should be noted that a small variance means that the orientation angles of all textures are closely clustered around the average value, indicating that the road surface texture orientation is highly consistent and uniform. A large variance means that the texture orientation angles are scattered in various corners and far from the average value, indicating that the road surface texture orientation is disordered and has a very low degree of uniformity.

[0063] Texture features are formed by integrating texture direction features, texture density features, and texture uniformity features.

[0064] By integrating three types of characteristics—direction, density, and uniformity—the textural differences between normal road surfaces and asphalt-covered road surfaces can be comprehensively quantified from three core dimensions: directional regularity, quantity distribution, and morphological consistency.

[0065] S3. Establish an oil bloom disease identification model based on deep learning. Based on color features, gloss features, and texture features, the oil bloom disease identification model outputs preliminary suspected oil bloom disease areas and their location information, forming a set of suspected areas.

[0066] The specific content of the above-mentioned asphalt pavement disease identification model is as follows: Collect samples under different asphalt pavement scenarios, extract the color features, gloss features, and texture features of each sample, and label the actual state of the samples to form a model training dataset.

[0067] The training dataset is divided into a training subset and a validation subset according to a preset ratio.

[0068] It should be noted that the preset ratio is determined based on the premise that the training subset needs to be sufficient to support the model in learning the features of different asphalt pavement scenarios, while the validation subset needs to have sufficient sample size and scenario coverage to accurately verify the model's fit.

[0069] The training subset features are input into the model for training, and the consistency between the calculated output of the subset and the actual annotation is verified.

[0070] It should be noted that the accuracy of the oil spill area location is calculated using the intersection-union ratio (IUGR), which is the ratio of the intersection area to the union area between the predicted oil spill area and the actual labeled oil spill area.

[0071] The oil penetration level consistency is calculated by measuring the proportion of the number of levels that match the predicted levels from the model to the actual labeled levels out of the total number of validation subsets.

[0072] The preliminary model was tested using an independent sample set that was not involved in the partitioning, and the output of the suspected area was compared with the actual oil spill situation.

[0073] If the deviation exceeds the preset range, the training subset features are returned to the model for retraining and readjustment; otherwise, the model construction is completed.

[0074] It should be noted that a deviation exceeding the preset range indicates that the model has not fully grasped the characteristics and patterns of oil bloom damage. This may be due to insufficient learning of certain scene samples during training, or improper model parameter settings, resulting in an inability to accurately determine the oil bloom state based on features. If the model is used directly in this case, misclassification or omission of oil bloom areas may occur.

[0075] The steps for forming the above suspected area set are as follows: input the color feature, gloss feature, and texture feature data into the oil bloom disease identification model to obtain the preliminary boundary range and location coordinates of the suspected oil bloom disease area.

[0076] If each initially suspected area has complete boundaries, meets the accuracy of its positioning coordinates, and its area meets the preset threshold, it is determined to be a valid area; otherwise, it is excluded.

[0077] It should be noted that incomplete boundaries are most likely false alarms caused by image sensor noise, road surface stains, or model segmentation errors, rather than actual bleeding areas. Bleeding defects, due to the continuous upward movement of lightweight asphalt components, typically exhibit complete boundaries. Excluding areas with incomplete boundaries can prevent subsequent re-inspection resources from being wasted on invalid areas.

[0078] If the positioning coordinate accuracy is not up to standard, such as if the coordinates deviate too much from the actual road surface position, the drone will not be able to fly accurately over the suspected area and collect effective spectral data, which will affect the confirmation of the oil spill area. In the implementation of this invention, the positioning coordinate accuracy is up to standard when the deviation of each positioning coordinate does not exceed 0.5 cm.

[0079] Bleeding is a continuous area formed by the floating of lightweight components in asphalt. True bleeding areas usually have a certain size, while initially suspected areas that are too small are often isolated noise spots, reflections from small stones, or localized stains, and are not true bleeding. Screening by area thresholds can further reduce invalid suspected areas.

[0080] The preset area threshold was determined based on sample statistics and validation optimization of real oil spill areas.

[0081] If the boundaries of valid regions overlap or the spacing is too close, they are merged into one region and the information is updated; otherwise, they are retained as independent regions.

[0082] All valid areas are aggregated, assigned a unique identifier, and associated with their boundaries and location information to form a set of suspected areas.

[0083] S4. Based on the set of suspected areas, plan the drone re-inspection route and control the drone to fly over each suspected area to collect spectral reflectance data of specific bands in the corresponding area.

[0084] See Figure 2 As shown, the specific steps for planning the UAV re-inspection route based on the suspected area set are as follows: obtain the positioning coordinates and boundary range of each effective suspected area from the suspected area set, and read the recognition confidence level of each area output by the oil spill disease identification model, and mark the confidence level as the degree of disease suspicion of each area.

[0085] Suspected areas are ranked from highest to lowest suspicion level, and areas with the same level of suspicion are categorized according to the concentration of their spatial distribution.

[0086] After sorting, suspected areas that are spatially adjacent are divided into the same re-examination unit, while spatially dispersed areas are treated as independent re-examination units.

[0087] Within each re-inspection unit, the order of accessing collection points is planned from near to far according to the regional location, forming a continuous flight path.

[0088] Connect the starting collection points of each re-inspection unit according to the re-inspection priority order and use the shortest path to form a complete flight path.

[0089] The second round of inspections improves the accuracy and robustness of disease identification, avoids missed detections and misjudgments, and sorts the suspected areas according to the degree of suspicion of the disease, divides the spatially adjacent inspection units and plans the shortest flight path. This reduces the invalid mileage and energy consumption of the drones, and prioritizes the inspection of highly suspected areas, ensuring that spectral data of key areas are obtained first, avoiding resource waste and improving the timeliness of the inspection process.

[0090] The specific content of the spectral reflectance data of the corresponding regions collected above is as follows: Based on the characteristics of each region in the suspected region set, a specific combination of spectral bands to be collected is assigned to it.

[0091] Spectral data of a specific band assigned to each suspected area were collected sequentially.

[0092] If the reflectance data acquired in the current band is not within the normal physical range for that band, the acquisition is deemed invalid, the sensor parameters are reconfigured, and the data for that band is acquired again.

[0093] It should be noted that the reflectivity of different materials at specific wavelengths will exhibit a fixed range due to differences in their composition and structure. This range is an objective physical law derived from extensive experimental statistics. If the collected data exceeds this range, it indicates that the data does not accurately reflect the optical properties of the road surface material. This is most likely due to incorrect sensor parameter configuration or abnormal data caused by external interference, and cannot be used for subsequent oil seepage assessment.

[0094] Conversely, if the data in that band is not valid, the data in that band will be marked as valid, and the acquisition of the next band will continue.

[0095] Once all specified band data for a suspected area have been collected and are valid, these data are integrated into a complete spectral reflectance dataset for that area.

[0096] Repeat the above collection and judgment process for all suspected areas until a complete spectral reflectance dataset of all suspected areas is obtained.

[0097] By extracting specific band reflectance data of the suspected area, it is determined whether it falls within the preset oil spill reflectance threshold range. Then, by combining the difference characteristics of the second specific band reflectance, the consistency between the area to be confirmed and the oil spill standard is verified. Finally, the real oil spill area and the non-oil spill interference area are accurately distinguished, avoiding misjudgment caused by relying solely on image features.

[0098] S5. Screen suspected areas based on spectral reflectance data of specific bands to confirm the oil spill area.

[0099] See Figure 3 As shown, the steps for determining the above-mentioned oil spill area are as follows: bind the unique identifier of each suspected area to its specific band spectral reflectance data.

[0100] For each suspected area, extract its specific band reflectance data. If the specific band reflectance data falls within the preset oil spill reflectance threshold range, it is marked as an oil spill area to be confirmed.

[0101] It should be noted that the preset oil spill reflectance threshold range is determined based on statistical data of specific band reflectance samples of real oil spills and normal road surfaces under different scenarios, combined with verification and testing optimization.

[0102] Objective physical laws verified through numerous asphalt pavement spectral experiments have shown that the reflectance of asphalt-rich bleeding pavement is significantly higher than that of normal pavement dominated by aggregates. Therefore, the above method can be used to make a judgment.

[0103] For areas to be confirmed as oil spills, retrieve the reflectivity data of the second specific band. If the reflectivity of the first specific band is always higher than that of the second specific band, and the difference between the two is within the stable range unique to oil spill pavements, then mark the area as pending confirmation.

[0104] It should be noted that the first specific band is sensitive to the asphalt composition, and the reflectivity of asphalt-rich bleeding pavement is significantly higher in this band. The second specific band is sensitive to the aggregate composition, and the reflectivity of normal pavement is relatively higher in this band due to the exposure of aggregate.

[0105] Asphalt is unaffected by minor road surface stains, changes in light and shadow, etc. However, on normal roads or in areas without oil spillage, the difference in reflectivity between the two bands either exceeds this range or has no stable pattern.

[0106] Verify the consistency between the spectral data of the suspected oil spill area and the oil spill standard. If there is no contradiction, it is confirmed as an oil spill area.

[0107] Interference areas are filtered out based on criteria such as whether the reflectance falls within the oil bloom threshold range, the relationship and difference between the two bands of reflectance, etc., and only the real oil bloom areas are retained. This avoids deviations in subsequent disease statistics and grade determination, and ensures the accuracy of oil bloom identification results.

[0108] S6. Calculate the effective area of ​​the bleeding zone, integrate the location information of the disease, the degree of deviation of the spectral reflectance data and the statistical results of the effective area to generate the degree level of bleeding disease of asphalt pavement.

[0109] Specifically, based on the ratio of the oil spill area to the overall road surface and the degree of deviation of the spectral reflectance from the normal road surface, three levels of judgment criteria are set: minor, moderate, and severe.

[0110] More specifically, the judgment criteria are set based on an area ratio coefficient and a spectral deviation coefficient. The area ratio coefficient is the ratio of the total area of ​​the bleeding zone to the total area of ​​the target asphalt pavement, and the spectral deviation coefficient is the ratio or difference between the average reflectance value of the bleeding zone in a specific band and the standard reflectance value of the normal pavement in that band.

[0111] After collecting historical data on road surfaces without defects and performing statistical analysis, a preset range of characteristic intervals for normal road surfaces is determined, and the following level determination rules are established: if both the area ratio coefficient and the spectral deviation coefficient are within the preset range, the area is determined to be of a slight level.

[0112] If only one of the area ratio coefficient and the spectral deviation coefficient is within the preset range, the region is determined to be of medium grade.

[0113] If both the area ratio coefficient and the spectral deviation coefficient deviate from the preset range, the region is determined to be of a severe level.

[0114] An anomaly in a single indicator may be a false alarm caused by other factors, such as shadows, water stains, or sensor noise. Using two indicators at a time improves the accuracy and reliability of identification and reduces false alarms.

[0115] The area proportion and spatial distribution of oil spills of different grades are statistically analyzed for the entire road section. If high-grade areas are concentrated, the overall road section evaluation grade is improved.

[0116] A specific implementation step for the above-mentioned high-level regional distribution concentration is as follows: calculate the centroid coordinates of all severe oil spill areas, and calculate the average nearest neighbor distance between these centroids to calculate their spatial clustering index. If the average nearest neighbor distance is lower than the preset threshold distance, it is determined to be a high-level regional distribution concentration.

[0117] The specific details of the above-mentioned improvement of the overall road section evaluation level are as follows: The overall road section evaluation level is initially determined to be Level L based on the proportion of the total oil spill area of ​​the entire road section, where L ∈ {minor, moderate, severe}.

[0118] If the spatial clustering index indicates a concentrated distribution, the overall road segment evaluation level will be finalized as (L+1). For example, if the initial rating is moderate, it will be upgraded to severe; if the initial rating is already severe, it will not be upgraded further.

[0119] The final result includes both individual region ratings and overall evaluation ratings.

[0120] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0121] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0122] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0123] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0124] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying asphalt pavement defects based on unmanned aerial vehicle (UAV) scanning, characterized in that: include: S1. By using a drone equipped with a multispectral imager to fly along a preset route, continuous images of the target asphalt pavement are collected, and a continuous orthophoto map of the pavement covering the entire road section is generated. S2. Extract the color features, gloss features, and texture features of the target asphalt pavement area from the orthophoto image; S3. Based on deep learning, establish an oil bloom disease identification model. According to color features, gloss features and texture features, the oil bloom disease identification model outputs the preliminary suspected oil bloom disease areas and their location information, forming a set of suspected areas. S4. Based on the set of suspected areas, plan the drone re-inspection route and control the drone to fly over each suspected area to collect spectral reflectance data of specific bands in the corresponding area. S5. Screen suspected areas based on spectral reflectance data of specific bands to confirm the oil spill area; S6. Calculate the effective area of ​​the bleeding zone, and integrate the location information of the disease, the degree of deviation of the spectral reflectance data and the statistical results of the effective area to generate the degree level of bleeding disease in asphalt pavement. The steps for determining the oil spill area are as follows: Each suspected area is uniquely identified and its specific band spectral reflectance data is bound together; for each suspected area, its specific band reflectance data is extracted; if the specific band reflectance data falls within a preset oil spill reflectance threshold range, it is marked as an oil spill area to be confirmed; for the oil spill area to be confirmed, its second specific band reflectance data is retrieved; if the reflectance of the first specific band is consistently higher than that of the second specific band, and the difference between the two is within the stable range characteristic of the oil spill road surface, it is marked as to be confirmed; the consistency between the spectral data of the oil spill area to be confirmed and the oil spill standard is verified; if there is no contradiction, it is confirmed as an oil spill area.

2. The method for identifying asphalt pavement defects based on UAV scanning according to claim 1, characterized in that: The color feature extraction steps are as follows: Preprocess the orthophoto to separate the target asphalt pavement area from the non-pavement area, and obtain an image area containing only the target asphalt pavement. Convert the image of the target asphalt pavement area to a color space that includes brightness and saturation parameters; The image pixels of the target asphalt pavement area are traversed, and the brightness and saturation values ​​of each pixel are extracted; The extracted brightness and saturation values ​​are divided into regions, and pixels with similar brightness and saturation values ​​are grouped into the same feature region; The intervals formed by the minimum and maximum values ​​of all pixel brightness and saturation values ​​within each feature region are respectively denoted as the brightness distribution range and the saturation distribution range; The brightness and saturation distribution ranges of each feature region are statistically analyzed to form the color characteristics of that feature region.

3. The method for identifying asphalt pavement defects based on UAV scanning according to claim 1, characterized in that: The glossiness feature extraction steps are as follows: Select a local area of ​​a preset size as the analysis unit from the separated target asphalt pavement area; Detect highlight regions in each analysis unit, identify sets of pixels with strong reflective properties within the unit, and form highlight candidate regions; If the difference in reflectance characteristics between adjacent pixels in the highlight candidate area is within a preset range, and its boundary shows a transitional shape that gradually weakens from the center to the edge, it is marked as an effective gloss area. The proportion of the effective gloss area in the corresponding analysis unit is used as the gloss coverage characteristic of that unit. Analyze the brightness gradient changes of pixels within the effective gloss area, and determine the gloss diffusion range characteristics based on the smoothness of the gradient changes. By integrating the gloss coverage characteristics and gloss diffusion range characteristics of each analysis unit, the gloss characteristics of the target asphalt pavement area are formed.

4. The method for identifying asphalt pavement defects based on UAV scanning according to claim 1, characterized in that: The texture feature extraction steps are as follows: Traverse the pixels within the target asphalt pavement area. If the variation in grayscale values ​​of adjacent pixels is within a preset stable range and exhibits a directional regularity, mark them as potential texture structures. If the extension direction deviation of adjacent texture segments in a potential texture structure exceeds the preset angle range, it is determined that there is an interruption and its start and end coordinates and the straight-line distance between the two texture segments are recorded as the interval distance. Count the number of main extension directions of each texture structure that fall into the corresponding interval, calculate the proportion of the number of textures in each interval to the total number of textures, and form texture direction features; Divide the target asphalt pavement area into several equal-area sub-regions, count the total number of texture structures in each sub-region, and form texture density characteristics by comparing the differences in the number of different sub-regions; Calculate the variance between the texture orientation feature values ​​of all texture structures, and use the reciprocal of this variance as the texture uniformity feature; Texture features are formed by integrating texture direction features, texture density features, and texture uniformity features.

5. The method for identifying asphalt pavement defects based on UAV scanning according to claim 1, characterized in that: The specific details of the oilseed bleeding disease identification model are as follows: Collect samples from different asphalt pavement scenarios, extract color features, gloss features, and texture features of each sample, and label the actual state of the samples to form a model training dataset; The training dataset is divided into a training subset and a validation subset according to a preset ratio; The training subset features are input into the model for training, and the consistency between the calculated output of the subset and the actual annotation is verified. The preliminary model was tested using an independent sample set that was not involved in the partitioning, and the output of the suspected area was compared with the actual oil spill situation. If the deviation exceeds the preset range, the training subset features are returned to the model for retraining and readjustment; otherwise, the model construction is completed.

6. The method for identifying asphalt pavement defects based on UAV scanning according to claim 1, characterized in that: The steps for forming the suspected area set are as follows: Input the color features, gloss features, and texture features into the oil bleeding disease identification model to obtain the preliminary boundary range and location coordinates of the suspected oil bleeding disease area; If each initially suspected area has complete boundaries, meets the accuracy of its positioning coordinates, and its area meets the preset threshold, it is determined to be a valid area; otherwise, it is excluded. If the boundaries of valid regions overlap or the spacing is too close, they are merged into one region and the information is updated; otherwise, they are retained as independent regions. All valid areas are aggregated, assigned a unique identifier, and associated with their boundaries and location information to form a set of suspected areas.

7. The method for identifying asphalt pavement defects based on UAV scanning according to claim 1, characterized in that: The specific steps for planning the drone re-inspection route based on the suspected area set are as follows: The location coordinates and boundary range of each valid suspected area are obtained from the suspected area set, and the identification confidence of each area is read from the oil seepage disease identification model. The confidence is marked as the degree of disease suspicion of each area. Suspected areas are ranked from highest to lowest level of suspicion; areas with the same level of suspicion are categorized according to the concentration of their spatial distribution. After sorting, suspected areas that are spatially adjacent are divided into the same re-examination unit, while spatially dispersed areas are treated as independent re-examination units. Within each re-inspection unit, the order of accessing collection points is planned from near to far according to the regional location, forming a continuous flight path; Connect the starting collection points of each re-inspection unit according to the re-inspection priority order and use the shortest path to form a complete flight path.

8. The method for identifying asphalt pavement defects based on UAV scanning according to claim 1, characterized in that: The specific details of the spectral reflectance data for the corresponding region in the specified band are as follows: Based on the characteristics of each region in the suspected region set, assign it a specific combination of bands to be collected. For each suspected area, spectral data of its assigned specific band is collected sequentially; If the reflectivity data acquired in the current band is not within the normal physical range of that band, the acquisition is deemed invalid, the sensor parameters are reconfigured, and the data in that band is acquired again. Conversely, if the data in that band is not valid, the data in that band will be marked as valid, and the acquisition of the next band will continue. Once all specified band data for a suspected area have been collected and are valid, these data will be integrated into a complete spectral reflectance dataset for that area. Repeat the above collection and judgment process for all suspected areas until a complete spectral reflectance dataset of all suspected areas is obtained.

9. The method for identifying asphalt pavement defects based on UAV scanning according to claim 1, characterized in that: The specific details of the asphalt pavement bleeding damage severity levels are as follows: Based on the ratio of the oil spill area to the overall road surface and the degree of deviation of the spectral reflectance from the normal road surface, three levels of judgment criteria are set: minor, moderate, and severe. The area proportion and spatial distribution of oil spill areas of each grade are statistically analyzed for the entire road section. If high-grade areas are concentrated, the overall road section evaluation grade is improved. The final result includes both individual region ratings and overall evaluation ratings.

Images