Nondestructive testing method for highway pavement defects based on unmanned aerial vehicle inspection
By using a dynamic energy function model and morphological operations, the accuracy problem of traditional graph cut algorithms in detecting road surface defects in UAV inspections has been solved, achieving efficient and accurate pothole defect identification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHAANXI TONGCHEN HEGUANG LOW TEMPERATURE TECH CO LTD
- Filing Date
- 2026-03-19
- Publication Date
- 2026-06-23
AI Technical Summary
Traditional graph cut algorithms are less accurate when used in UAV inspections to detect defects in highway pavement. They are affected by grayscale reference offset and lighting interference, resulting in blurred boundaries, missed detections, or incorrect fusion of segmentation results.
Threshold segmentation is performed by acquiring the grayscale distribution features of aerial images to obtain suspected defect areas. By combining the gradient features of pixels and the grayscale difference features of neighboring areas, a dynamic energy function model is constructed to perform energy minimization segmentation. Finally, morphological operations are used to screen the final defect areas.
It improves the accuracy of highway pavement defect detection, reduces the missed detection rate, enhances the continuity and accuracy of the segmentation boundary, and conforms to the physical morphology of potholes.
Smart Images
Figure CN121883491B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image recognition technology, and more specifically to a non-destructive testing method for highway pavement defects based on unmanned aerial vehicle (UAV) inspection. Background Technology
[0002] Due to long-term environmental erosion, material aging, and traffic loads, highway pavements are prone to typical local defects, such as potholes. These defects are mainly caused by asphalt mixture bonding failure, water damage, or structural fatigue, manifesting as surface material loss and the formation of irregular depressions. Potholes not only reduce pavement smoothness and driving comfort but can also affect vehicle handling and cause accidents. To safely and efficiently detect pavement defects, drones can be used to inspect and photograph roads. Existing graph cut algorithms based on energy minimization can initially separate candidate defect regions from road images for subsequent quantification of defect states.
[0003] However, when traditional graph cut algorithms detect potholes and defects in road images, their effectiveness relies heavily on a static energy function model built on a fixed grayscale reference of normal road surface samples. In actual drone inspections, aerial road images may be affected by material aging or lighting interference, resulting in overall grayscale reference shifts or local contrast distortions. This causes the static model to fail to accurately define the feature boundaries between potholes and road surfaces, leading to blurred boundaries, missed detections, or incorrect fusion with defect areas. Furthermore, segmentation results based on minimizing the energy function of local pixel features lack guidance on segmentation rationality, potentially producing fragmented or inconsistent segmentation results. Therefore, traditional graph cut algorithms are prone to low accuracy in road defect detection. Summary of the Invention
[0004] To address the aforementioned technical problems, the present invention aims to provide a non-destructive testing method for highway pavement defects based on unmanned aerial vehicle (UAV) inspection. The specific technical solution adopted is as follows:
[0005] Acquire aerial images of the road;
[0006] Threshold segmentation is performed on the grayscale distribution characteristics of the road image to obtain suspected defect areas; the pothole feature values of the pixels are obtained based on the gradient features and neighborhood grayscale difference features of the pixels in the suspected defect areas; a first confidence level and a second confidence level are obtained based on the grayscale distribution characteristics of the suspected defect areas and the grayscale features of the pixels.
[0007] The road surface value of the pixel is obtained based on the pothole feature value and the first confidence level; the pothole value of the pixel is obtained based on the pothole feature value and the second confidence level; the outlier value is obtained based on the grayscale difference feature between adjacent pixels; an undirected graph of the suspected defect region is constructed based on the road surface value, the pothole value, and the outlier value, and energy minimization segmentation is performed to obtain the pothole pixels;
[0008] The pit region is obtained based on the positional distribution characteristics of the pit pixels; the final defect region is obtained based on the shape characteristics of the pit region.
[0009] Further, the step of thresholding based on the grayscale distribution features of the road image to obtain suspected defect areas includes:
[0010] The grayscale threshold of the road image is obtained by the Otsu's method, and the product of the preset adjustment factor and the grayscale threshold is calculated to obtain the final coarse segmentation threshold. The contours of different regions in the road image whose grayscale does not exceed the final coarse segmentation threshold are extracted, and the region range of the smallest bounding rectangle of each contour is taken as the suspected defect region.
[0011] Further, the step of obtaining the pit feature value of the pixel based on the gradient features and neighborhood gray-level difference features of the pixel in the suspected defect region includes:
[0012] In the formula, H represents the pit feature value of a pixel, and a represents the preset weight coefficient. Let F represent the hyperbolic tangent function, and let F represent the gradient value of the pixel. The maximum gradient value in the suspected defect region is represented by N, the number of pixels in the preset neighborhood of the pixel is represented by r, and the position of the pixel is represented by the gray level in the preset neighborhood sorted from smallest to largest.
[0013] Further, the step of obtaining the first confidence level and the second confidence level based on the grayscale distribution characteristics of the suspected defective region and the grayscale characteristics of the pixel includes:
[0014] The grayscale histogram of the suspected defective region is smoothed. In the smoothed grayscale histogram, the minimum grayscale value among all local maxima is taken as the low grayscale value, and the maximum grayscale value among all local maxima is taken as the high grayscale value. The absolute value of the difference between the grayscale value of the pixel and the low grayscale value is calculated and negatively correlated to obtain a first confidence level. The absolute value of the difference between the grayscale value of the pixel and the high grayscale value is calculated and negatively correlated to obtain a second confidence level.
[0015] Further, the step of obtaining the road surface cost of the pixel based on the pothole feature value and the first confidence level includes:
[0016] In the formula, D represents the road surface cost of a pixel. Let G represent the natural logarithm, H represent the first confidence level of the pixel, H represent the pit feature value of the pixel, and x represent the preset minimum positive number.
[0017] Further, the step of obtaining the pit cost of the pixel based on the pit feature value and the second confidence level includes:
[0018] In the formula, E represents the pit cost of a pixel. R represents the natural logarithm, H represents the second confidence level of the pixel, H represents the pit feature value of the pixel, and x represents the preset minimum positive number.
[0019] Furthermore, the step of obtaining the heterogeneous value based on the grayscale difference features between adjacent pixels includes:
[0020] Calculate the absolute value of the grayscale difference between adjacent pixels and perform negative correlation mapping to obtain the heterogeneous value between adjacent pixels.
[0021] Further, the step of constructing an undirected graph of the suspected defect region based on the road surface value, the pothole value, and the outlier value, and performing energy-minimum segmentation to obtain pothole pixels includes:
[0022] The two terminal nodes of the undirected graph are the source node representing the pothole feature and the sink node representing the road surface image, respectively. The weight of the connection line between each pixel and the source node is the pothole value, the weight of the connection line between each pixel and the sink node is the road surface value, and the weight of the connection line between adjacent pixels is the heterogeneous value. The undirected graph is segmented by minimizing energy according to the graph cut algorithm, and the pixels connected to the source node are taken as the pothole pixels.
[0023] Furthermore, the step of obtaining the pit region based on the positional distribution features of the pit pixels includes:
[0024] Morphological closing operations are performed on the pitted pixels to obtain different pitted regions.
[0025] Furthermore, the step of obtaining the final defect region based on the shape characteristics of the pit region includes:
[0026] Calculate the area ratio of the pit region to the area of the smallest circumcircle of the pit region to obtain the shape feature value; the pit region whose shape feature value exceeds the preset shape threshold is taken as the final defect region.
[0027] The present invention has the following beneficial effects:
[0028] In this invention, acquiring suspected defect areas can initially and quickly determine areas where potholes may exist based on the grayscale difference characteristics between potholes and normal road surfaces, reducing subsequent computational load. Acquiring pothole feature values allows analysis of the probability that a pixel is located within a pothole region based on its gradient and grayscale characteristics. Acquiring first and second confidence scores further analyzes the region type of a pixel based on the grayscale difference between the pothole region and the road surface region. Acquiring road surface cost values characterizes the cost of classifying a pixel as a road surface region; acquiring pothole cost values characterizes the cost of classifying a pixel as a pothole region; combining road surface cost values and pothole cost values allows for accurate classification of pixels in different regions. Acquiring out-of-type cost values characterizes the probability that adjacent pixels belong to the same type of region, further improving segmentation accuracy. Acquiring pothole pixels allows for accurate identification of pothole defect regions based on energy minimization segmentation. Finally, obtaining the final defect region based on the shape characteristics of the pothole region allows for the selection of more realistic pothole defects based on these shape characteristics, improving the accuracy of defect detection. Attached Figure Description
[0029] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.
[0030] Figure 1 The flowchart illustrates a method for non-destructive testing of road surface defects based on unmanned aerial vehicle (UAV) inspection, as provided in one embodiment of the present invention. Detailed Implementation
[0031] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a non-destructive testing method for highway pavement defects based on unmanned aerial vehicle (UAV) inspection proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0032] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0033] The following description, in conjunction with the accompanying drawings, details a specific scheme for a non-destructive testing method for highway pavement defects based on unmanned aerial vehicle (UAV) inspection, provided by this invention.
[0034] Please see Figure 1 The diagram illustrates a flowchart of a non-destructive testing method for highway pavement defects based on unmanned aerial vehicle (UAV) inspection, according to an embodiment of the present invention. The method includes the following steps:
[0035] Step S1: Obtain aerial images of the road.
[0036] In this embodiment of the invention, the scenario involves using a drone to detect road surface defects, thereby improving the accuracy of road defect detection. First, a high-resolution industrial camera is mounted on the drone to capture images of the road surface under normal lighting conditions. After capturing the images, geometric correction is performed on the road surface images to reduce the impact of distortion and geometric discrepancies. The corrected road surface images are then filtered for noise reduction. In this embodiment, bilateral filtering is used for smoothing. This algorithm considers both pixel grayscale similarity and spatial distance weights, which can suppress Gaussian noise and salt-and-pepper noise caused by drone flight vibrations while fully preserving the grayscale gradient information of pothole edges, avoiding the edge blurring problem caused by traditional Gaussian filtering. After noise reduction, road surface areas are extracted. A simple color-based segmentation method automatically extracts the road area, effectively eliminating interference from vegetation, guardrails, vehicles, and road markings, generating an image containing only the road area. After road surface extraction, a contrast-limited adaptive histogram equalization algorithm is used to enhance the local contrast between the road surface and potential defect areas, improving the visibility of pothole edges, ultimately obtaining an aerial image of the road.
[0037] Step S2: Threshold segmentation is performed based on the grayscale distribution characteristics of the road image to obtain suspected defect areas; the pothole feature values of the pixels are obtained based on the gradient features and neighborhood grayscale difference features of the pixels in the suspected defect areas; and the first confidence level and the second confidence level are obtained based on the grayscale distribution characteristics of the suspected defect areas and the grayscale features of the pixels.
[0038] Traditional graph cut algorithms typically construct their energy functions based on statistical models (e.g., histograms) trained with fixed samples, relying on the histogram differences between preset samples and the region to be detected to construct data terms. However, energy functions containing only data and smoothing terms lack guidance on the rationality of the segmentation results, failing to adapt to complex situations such as road surface grayscale reference shifts and illumination fluctuations in UAV inspection scenarios. This leads to incomplete or fragmented segmentation boundaries, affecting the accuracy of defect detection. Therefore, the purpose of this invention is to overcome the limitations of traditional static energy function models in pothole region segmentation. By analyzing the statistical characteristics of the current road surface image and designing a cost correction method focusing on edge pixel classification, an energy model for data and smoothing terms is dynamically constructed. Because pothole regions have lower grayscale values than surrounding normal road surfaces due to depressions, shadows, or material differences, and the texture uniformity of pothole boundaries differs significantly from that of the road surface, with grayscale fluctuations in the neighborhood being much greater than in the uniform areas of normal road surfaces, the accuracy of edge pixel attribution directly determines the continuity and accuracy of the segmentation boundaries, which is crucial for improving the overall accuracy of the segmentation results. The embodiments of the present invention perform independent analysis on each suspected defect region, so that the energy function can not only accurately respond to the real feature differences in the current defect detection scenario, but also solve the problems of misclassification and omission of edge pixels, thereby improving the accuracy of segmentation and defect detection.
[0039] First, threshold segmentation is performed based on the grayscale distribution characteristics of the road image to obtain suspected defect areas. Preferably, in this embodiment of the invention, the step of obtaining suspected defect areas includes: obtaining the grayscale threshold of the road image using the maximum inter-class variance method. It should be noted that the maximum inter-class variance method is existing technology, and the specific steps will not be elaborated here. The product of a preset adjustment factor and the grayscale threshold is calculated to obtain the final coarse segmentation threshold. In this embodiment of the invention, the preset adjustment factor is 1.2, which can be determined by the implementer according to the implementation scenario. The preset adjustment factor must be greater than 1. The final coarse segmentation threshold is larger than the grayscale threshold. Since the grayscale value of road potholes is low, the final coarse segmentation threshold can cover more areas with low grayscale values, reducing the false negative rate. The contours of different areas in the road image whose grayscale values do not exceed the final coarse segmentation threshold are extracted, and the area of the smallest bounding rectangle of each contour is taken as the suspected defect area. Obtaining suspected defect areas can initially and quickly locate areas where potholes may exist, effectively eliminating a large amount of non-defective road surface interference and reducing the subsequent computational load.
[0040] Furthermore, the data term in the energy function of the graph cut algorithm is used to quantify the probability that a pixel belongs to a pothole feature or a normal road surface feature, and can be constructed based on grayscale features and gradient features; therefore, the pothole feature value of the pixel is obtained according to the gradient features and neighborhood grayscale difference features of the pixel in the suspected defect area; preferably, in this embodiment of the invention, the step of obtaining the pothole feature value includes:
[0041] ;
[0042] In the formula, H represents the pit feature value of a pixel, and a represents the preset weight coefficient. Let F represent the hyperbolic tangent function, and let F represent the gradient value at a pixel. This represents the maximum gradient value in the suspected defect area. The gradient value reflects the degree of drastic change in the grayscale of the pixel itself. The larger the gradient, the higher the probability that the pixel is located at an edge or in a region of abrupt texture change, and the more likely it is to be located at the edge of a pothole. N represents the number of pixels in the preset neighborhood of the pixel, and r represents the position of the pixel when sorted by grayscale from smallest to largest within the preset neighborhood. In this embodiment of the invention, the preset neighborhood is the range formed by the pixel and its eight neighbors, which can be determined by the implementer according to the implementation scenario. The grayscale value of pixels on the edge of a pothole is usually lower than that of the surrounding road surface, and their position is relatively earlier. The closer the value is to 1, the more likely the pixel is to be located at the edge of a pothole. In this embodiment of the invention, the preset weight coefficient is 0.6. This value can adapt to detection scenarios with complex textures and significant gradient changes at pothole edges, improving the sensitivity of the graph cut algorithm to pothole edges and texture abrupt changes. Implementers can determine the value according to their specific implementation scenarios. The pothole feature value measures the probability of a pixel being located in a pothole. For pixels located at the edge of a pothole, the gradient is larger and the brightness is darker. For pixels inside a pothole, even if the brightness order inside the pothole is affected, the gradient feature contributes to the pothole feature value, resulting in a relatively high pothole feature value. For normal road surfaces, the gradient is smaller and the order is not prominent, resulting in a relatively low pothole feature value. Therefore, the larger the pothole feature value, the less likely the pixel is to be located in a normal road surface area.
[0043] Furthermore, the grayscale difference between the pothole area and the normal road surface area in the suspected defect area is significant. Therefore, the grayscale histogram of the local area containing potholes should exhibit a bimodal shape, with the peak of larger grayscale corresponding to the normal road surface and the peak of smaller grayscale corresponding to potholes and shadowed pixel areas. The type of a pixel can then be determined based on the difference in grayscale between the pixel and the corresponding peak. Therefore, a first confidence level and a second confidence level are obtained based on the grayscale distribution characteristics of the suspected defect area and the grayscale characteristics of the pixel. Preferably, in this embodiment of the invention, the steps of obtaining the first confidence level and the second confidence level include: smoothing the grayscale histogram of the suspected defect area, which can be done using Gaussian filtering to suppress minor fluctuations caused by noise and improve the accuracy of selecting local maxima. In the smoothed grayscale histogram, the minimum grayscale value among all local maxima is taken as the low grayscale value, which can serve as the grayscale benchmark for determining whether a pixel belongs to a pothole area; the maximum grayscale value among all local maxima is taken as the high grayscale value, which can serve as the grayscale benchmark for determining whether a pixel belongs to a normal road surface area. The absolute value of the difference between the grayscale value of a pixel and its lower grayscale value is calculated and negatively correlated to obtain a first confidence level. The closer the grayscale value of a pixel is to its lower grayscale value, the higher the first confidence level, and the more likely the pixel belongs to a pothole region; conversely, the lower the first confidence level, the more likely the pixel belongs to a road surface region. The absolute value of the difference between the grayscale value of a pixel and its higher grayscale value is calculated and negatively correlated to obtain a second confidence level. The closer the grayscale value of a pixel is to its higher grayscale value, the higher the second confidence level, and the more likely the pixel belongs to a road surface region; conversely, the lower the second confidence level, the more likely the pixel belongs to a pothole region. In this embodiment of the invention, using... Perform negative correlation mapping. Let i represent an exponential function with the natural constant as the base, and let i represent the independent variable.
[0044] Step S3: Obtain the road surface value of the pixel based on the pothole feature value and the first confidence level; obtain the pothole value of the pixel based on the pothole feature value and the second confidence level; obtain the outlier value based on the grayscale difference feature between adjacent pixels; construct an undirected graph of the suspected defect area based on the road surface value, pothole value, and outlier value, and perform energy minimization segmentation to obtain the pothole pixels.
[0045] After obtaining the pothole feature values, first confidence level, and second confidence level of the pixels in the suspected defect area, the road surface cost of the pixels can be obtained based on the pothole feature values and the first confidence level; preferably, in this embodiment of the invention, the step of obtaining the road surface cost includes:
[0046] ;
[0047] In the formula, D represents the road surface cost of a pixel. Let G represent the natural logarithm. This function is a standard practice in the energy minimization framework, converting probability confidence into cost. G represents the first confidence of a pixel, and H represents the pothole feature value of a pixel. The independent variable ranges from 0 to 1, and x represents a preset minimum positive number to avoid the independent variable being 0. The road surface cost quantifies the cost of a pixel being classified as part of the road surface. The higher the pothole feature value and the first confidence of a pixel, the more likely the pixel is to belong to a pothole region, resulting in a lower road surface cost. In an undirected graph, this means it is more likely to be assigned to a pothole endpoint. Pixels located at the edge of a pothole also have a lower road surface cost due to the addition of the pothole feature value. A lower road surface cost means a higher probability of marking the pixel as a pothole pixel in the graph cut algorithm; conversely, a higher road surface cost means a higher cost for a pixel on a normal road surface to be marked as a pothole pixel.
[0048] Furthermore, a complete binary segmentation model typically requires two different types of data items simultaneously. Using road surface cost as the road surface data item, the pothole cost of a pixel can be obtained based on pothole feature values and a second confidence level. Preferably, in this embodiment of the invention, the step of obtaining the pothole cost includes:
[0049] ;
[0050] In the formula, E represents the pit cost of a pixel. Let R represent the natural logarithm, H represent the second confidence level of a pixel, H represent the pothole feature value of a pixel, and x represent a preset minimum positive number. The higher the second confidence level and the lower the pothole feature value of a pixel, the more likely the pixel is located in a normal road surface area, resulting in a lower pothole value and a higher probability of the pixel being assigned to a road surface endpoint in the undirected graph. Conversely, the lower the second confidence level and the lower the pothole feature value, the less likely a pixel in a pothole region is to be labeled as a road surface region. The pothole value can be used as a pothole data item, and pixel assignment can be determined using both the pothole value and the road surface value.
[0051] Furthermore, the smoothing term in the energy function model can be obtained. The smoothing term can be defined as the cost of quantifying adjacent pixels belonging to different categories. Therefore, the out-of-category cost is obtained based on the grayscale difference characteristics between adjacent pixels. Preferably, in this embodiment, the step of obtaining the out-of-category cost includes: calculating the absolute value of the grayscale difference between adjacent pixels and performing a negative correlation mapping to obtain the out-of-category cost between adjacent pixels. This formula quantifies the grayscale difference between adjacent pixels. The smaller the grayscale difference, the more likely the two pixels are to represent the same type of region, thus the larger the out-of-category cost, and the greater the cost of classifying them into different categories. Conversely, the smaller the out-of-category cost, the larger the grayscale difference between adjacent pixels, and the smaller the cost of classifying them into different categories. Adjacent pixels refer to pixels that are directly connected in position.
[0052] Obtaining the road surface value, pothole value, and outlier value of pixels enables scene adaptation of the energy function, making it independent of absolute sample models. Instead, it utilizes the statistical characteristics and local context information of the current image to adaptively adjust the energy function of different pixels, effectively overcoming the overall grayscale shift caused by illumination changes or other factors, and improving segmentation accuracy. Furthermore, an undirected graph of suspected defect regions can be constructed based on the road surface value, pothole value, and outlier value, and energy-minimum segmentation can be performed to obtain pothole pixels. Preferably, in this embodiment, the step of obtaining pothole pixels includes: the two terminal nodes of the undirected graph are the source node representing pothole features and the sink node representing the road surface image, respectively; the weight of the connection line between each pixel and the source node is the pothole value, the weight of the connection line with the sink node is the road surface value, and the weight of the connection line between adjacent pixels is the outlier value; energy-minimum segmentation is performed on the undirected graph using a graph cut algorithm, and pixels connected to the source node are taken as pothole pixels. The sum of the weights of all edges cut by a cut is the total energy corresponding to that cut. According to graph cut theory, finding the minimum cut (the cut with the minimum total weight) is equivalent to minimizing the energy function. For example, if the pothole cost of a pixel to the source node is greater than the road surface cost to the sink node, then based on the minimum cut method, the connection between that pixel and the sink node will be cut, and that pixel will be assigned to the pothole source node. The minimum cut will divide all nodes into two disjoint sets. It should be noted that the graph cut algorithm is an existing technology, and the specific steps will not be elaborated here. Graph cut can accurately distinguish between pothole pixels and road surface pixels.
[0053] Step S4: Obtain the pit region based on the positional distribution characteristics of the pit pixels; obtain the final defect region based on the shape characteristics of the pit region.
[0054] While adaptive data terms and smoothing terms can effectively distinguish between potholes and road surfaces by utilizing local pixel features and neighborhood relationships, they are limited to local information modeling and fail to introduce macroscopic constraints on the overall shape of the region. Therefore, even if the graph cut algorithm based on this energy function achieves global energy minimization, it can only guarantee local optimality of pixel-level classification and cannot ensure that the segmentation result conforms to the physical characteristics of potholes in terms of macroscopic shape. For example, potholes should be circular or elliptical, rather than elongated shapes. Therefore, it is necessary to further introduce compactness constraints to guide the segmentation result towards a reasonable physical shape by penalizing non-compact shapes, thereby effectively filtering out mis-segmented regions with abnormal shapes and significantly improving the physical credibility and structural integrity of the final defect region. Therefore, the pothole region is obtained based on the positional distribution characteristics of pothole pixels. Preferably, in this embodiment of the invention, the step of obtaining the pothole region includes: performing morphological closing operations on pothole pixels to obtain different pothole regions; morphological operations can connect neighboring pothole pixels, fill small internal holes and smooth boundaries, thereby obtaining a more complete and connected segmentation result. Then, the final defect region can be obtained based on the shape characteristics of the pothole region.
[0055] Preferably, in this embodiment of the invention, the step of obtaining the final defect region includes: calculating the area ratio of the pit region to the area of the smallest circumcircle of the pit region to obtain a shape feature value; when the shape feature value is closer to 1, the region shape is closer to a circle or ellipse, which better matches the actual pit defect; by calculating the shape feature value, some fragmented and non-physical shape segmentation results can be filtered out, making the pit defect more consistent with the actual physical characteristics. Furthermore, pit regions with shape feature values exceeding a preset shape threshold can be taken as the final defect region. Obtaining the final defect region further improves the detection accuracy. In this embodiment of the invention, the preset shape threshold is 0.3, which can be determined by the implementer according to the implementation scenario. After obtaining a final defect region with clear boundaries and a reasonable shape, such regions can be analyzed at a higher level to determine countermeasures, which are not limited here.
[0056] In summary, this invention provides a non-destructive testing method for highway pavement defects based on UAV inspection. The method involves threshold segmentation of road images to obtain suspected defect regions; obtaining pothole feature values based on the gradient and neighborhood grayscale of pixels within the suspected defect regions; obtaining a first confidence level and a second confidence level based on the grayscale distribution of the suspected defect regions and the grayscale of the pixels; obtaining the pavement cost value of the pixels based on the pothole feature values and the first confidence level; obtaining the pothole cost value of the pixels based on the pothole feature values and the second confidence level; and obtaining the out-of-class cost value based on the grayscale difference features between adjacent pixels. This invention constructs an undirected graph of suspected defect regions based on the pavement cost value, pothole cost value, and out-of-class cost value, and segments it to obtain pothole pixels; obtains pothole regions based on the pothole pixels; and obtains the final defect region based on the shape features of the pothole regions, thus improving the accuracy of pavement defect detection.
[0057] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0058] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
Claims
1. A non-destructive testing method for highway pavement defects based on unmanned aerial vehicle (UAV) inspection, characterized in that, The method includes the following steps: Acquire aerial images of the road; Threshold segmentation is performed on the grayscale distribution characteristics of the road image to obtain suspected defect areas; the pothole feature values of the pixels are obtained based on the gradient features and neighborhood grayscale difference features of the pixels in the suspected defect areas; a first confidence level and a second confidence level are obtained based on the grayscale distribution characteristics of the suspected defect areas and the grayscale features of the pixels. The road surface value of the pixel is obtained based on the pothole feature value and the first confidence level; the pothole value of the pixel is obtained based on the pothole feature value and the second confidence level; the outlier value is obtained based on the grayscale difference feature between adjacent pixels; an undirected graph of the suspected defect region is constructed based on the road surface value, the pothole value, and the outlier value, and energy minimization segmentation is performed to obtain the pothole pixels; The pit region is obtained based on the positional distribution characteristics of the pit pixels; the final defect region is obtained based on the shape characteristics of the pit region. The step of obtaining the road surface cost of the pixel based on the pothole feature value and the first confidence level includes: In the formula, D represents the road surface cost of a pixel. Let G represent the natural logarithm, H represent the first confidence level of the pixel, and x represent the pit feature value of the pixel. The step of obtaining the pit cost of the pixel based on the pit feature value and the second confidence level includes: In the formula, E represents the pit cost of a pixel. R represents the natural logarithm, H represents the second confidence level of the pixel, and x represents the pit feature value of the pixel. The step of obtaining the heterogeneous value based on the grayscale difference feature between adjacent pixels includes: Calculate the absolute value of the grayscale difference between adjacent pixels and perform negative correlation mapping to obtain the heterogeneous value between adjacent pixels.
2. The method for non-destructive testing of highway pavement defects based on unmanned aerial vehicle (UAV) inspection according to claim 1, characterized in that, The step of thresholding based on the grayscale distribution features of the road image to obtain suspected defect areas includes: The grayscale threshold of the road image is obtained by the Otsu's method, and the product of the preset adjustment factor and the grayscale threshold is calculated to obtain the final coarse segmentation threshold. The contours of different regions in the road image whose grayscale does not exceed the final coarse segmentation threshold are extracted, and the region range of the smallest bounding rectangle of each contour is taken as the suspected defect region.
3. The method for non-destructive testing of highway pavement defects based on unmanned aerial vehicle (UAV) inspection according to claim 1, characterized in that, The step of obtaining the pit feature value of the pixel based on the gradient features and neighborhood gray-level difference features of the pixel in the suspected defect region includes: In the formula, H represents the pit feature value of a pixel, and a represents the preset weight coefficient. Let F represent the hyperbolic tangent function, and let F represent the gradient value of the pixel. The maximum gradient value in the suspected defect region is represented by N, the number of pixels in the preset neighborhood of the pixel is represented by r, and the position of the pixel is represented by the gray level in the preset neighborhood sorted from smallest to largest.
4. The method for non-destructive testing of highway pavement defects based on unmanned aerial vehicle (UAV) inspection according to claim 1, characterized in that, The steps of obtaining the first confidence level and the second confidence level based on the grayscale distribution characteristics of the suspected defect area and the grayscale characteristics of the pixel include: The grayscale histogram of the suspected defective region is smoothed. In the smoothed grayscale histogram, the minimum grayscale value among all local maxima is taken as the low grayscale value, and the maximum grayscale value among all local maxima is taken as the high grayscale value. The absolute value of the difference between the grayscale value of the pixel and the low grayscale value is calculated and negatively correlated to obtain a first confidence level. The absolute value of the difference between the grayscale value of the pixel and the high grayscale value is calculated and negatively correlated to obtain a second confidence level.
5. The method for non-destructive testing of highway pavement defects based on unmanned aerial vehicle (UAV) inspection according to claim 1, characterized in that, The step of constructing an undirected graph of suspected defect regions based on the road surface value, the pothole value, and the outlier value, and performing energy-minimum segmentation to obtain pothole pixels includes: The two terminal nodes of the undirected graph are the source node representing the pothole feature and the sink node representing the road surface image, respectively. The weight of the connection line between each pixel and the source node is the pothole value, the weight of the connection line between each pixel and the sink node is the road surface value, and the weight of the connection line between adjacent pixels is the heterogeneous value. The undirected graph is segmented by minimizing energy according to the graph cut algorithm, and the pixels connected to the source node are taken as the pothole pixels.
6. The method for non-destructive testing of highway pavement defects based on unmanned aerial vehicle (UAV) inspection according to claim 1, characterized in that, The step of obtaining the pit region based on the positional distribution features of pit pixels includes: Morphological closing operations are performed on the pitted pixels to obtain different pitted regions.
7. The method for non-destructive testing of highway pavement defects based on unmanned aerial vehicle (UAV) inspection according to claim 1, characterized in that, The step of obtaining the final defect region based on the shape characteristics of the pit region includes: Calculate the area ratio of the pit region to the area of the smallest circumcircle of the pit region to obtain the shape feature value; the pit region whose shape feature value exceeds the preset shape threshold is taken as the final defect region.