Rapid large-range asphalt-pavement quality assessment method and system based on high-resolution remote sensing

Abstract

Disclosed in the present invention are a rapid large-range asphalt-pavement quality assessment method and system based on high-resolution remote sensing. The method comprises the following steps: acquiring a high-resolution multispectral image and panchromatic image of a pavement area of a road network area to be subjected to assessment; generating superpixels of the pavement area on the basis of the high-resolution multispectral image and panchromatic image of the pavement area; removing superpixels of a non-asphalt pavement from the superpixels of the pavement area, so as to obtain superpixels of an asphalt pavement in the pavement area; on the basis of the superpixels of the asphalt pavement, performing calculation to obtain a brightness indicator, texture indicator, structure indicator and spectrum indicator of each superpixel of the asphalt pavement; and according to the calculated brightness indicator, texture indicator, structure indicator and spectrum indicator, performing calculation to obtain a remote-sensing quality assessment index corresponding to each superpixel of the asphalt pavement, and then obtaining a quality situation assessment diagram for the asphalt pavement. By means of the present invention, various assessment indexes are comprehensively used, thereby improving the precision and visibility of an assessment result, and reducing the manual intervention.

Description

A method and system for rapid assessment of large-scale asphalt pavement quality based on high-resolution remote sensing Technical Field

[0001] The present invention belongs to the technical field of road surface quality, and in particular relates to a method and system for rapid assessment of large-scale asphalt road surface quality based on high-resolution remote sensing. Background Art

[0002] Highways, as transportation corridors, play a vital role in socioeconomic development. Asphalt pavement is the primary pavement structure on high-grade highways. However, factors such as temperature, moisture, weathering, and vehicle load can adversely affect the structural strength and surface finish of asphalt pavements, significantly reducing their performance and leading to significant economic losses and accident risks. Therefore, timely and accurate asphalt pavement quality assessment is crucial for ensuring the quality of highway transportation.

[0003] Currently, asphalt pavement quality monitoring relies primarily on manual inspections and multi-functional road inspection vehicles. While these methods are capable of accurately measuring the quality of individual roads, large-scale asphalt pavement quality monitoring faces challenges such as high costs and the difficulty of sampling surveys to fully reflect the overall pavement quality. Remote sensing offers advantages such as wide data acquisition range and high timeliness. With the continuous improvement in the resolution of satellite remote sensing imagery, the acquisition of detailed road surface information is becoming possible, providing a reliable data source for the development of large-scale asphalt pavement quality assessment models. Therefore, high-resolution remote sensing technology provides a rapid and powerful technical approach for assessing asphalt pavement quality across large highway networks.

[0004] At present, high-resolution remote sensing images have been preliminarily applied to the rapid assessment of asphalt pavement quality. For example, Pan et al. successfully mapped the aging conditions of asphalt pavements in Fangshan District, Beijing, using Wordview-2 multispectral road imagery. Emery et al. used the histogram threshold segmentation method to classify Worldview-2 panchromatic road imagery to obtain asphalt pavements with different aging conditions. Mei et al. used IKONOS and QuickBird multispectral satellite remote sensing images to monitor the health of asphalt roads in airports and urban areas. Chen et al. used an improved recurrent neural network to automatically identify aging areas of asphalt pavement from high-resolution satellite imagery. The results showed that high-resolution satellite imagery can be used to efficiently obtain the health conditions of asphalt pavements over a large area.

[0005] However, most existing remote sensing assessment methods for asphalt pavement quality only utilize pavement spectral information. However, the changes in asphalt pavement spectral information are complex and limited. It is difficult to fully and accurately assess the pavement quality condition based solely on spectral characteristics, and cannot meet the actual application needs of pavement quality remote sensing monitoring.

[0006] Summary of the Invention

[0007] The purpose of the present invention is to address the shortcomings of the above-mentioned background technology and provide a method and system for rapid assessment of asphalt pavement quality over a large area based on high-resolution remote sensing. Superpixels are used as asphalt pavement quality assessment units, and four assessment indicators, namely brightness, texture, structure and spectrum, are comprehensively utilized to construct a superpixel-based asphalt pavement quality remote sensing assessment index, thereby realizing rapid assessment of asphalt pavements over a large area.

[0008] The technical solution adopted by the present invention is: a method for rapid assessment of asphalt pavement quality over a large area based on high-resolution remote sensing, comprising the following steps:

[0009] Obtain high-resolution multispectral and panchromatic images of the road surface area of ​​the road network area to be evaluated;

[0010] Generate superpixels of the road surface area based on the high-resolution multispectral image and panchromatic image of the road surface area;

[0011] Eliminate superpixels of non-asphalt pavement from superpixels in the pavement area to obtain superpixels of asphalt pavement in the pavement area;

[0012] Based on the asphalt pavement superpixels, the brightness index, texture index, structure index and spectral index of each superpixel of the asphalt pavement are calculated;

[0013] Based on the calculated brightness index, texture index, structure index and spectral index, a remote sensing quality evaluation index corresponding to each superpixel of the asphalt pavement is calculated; the remote sensing quality evaluation index is used to characterize the condition of the asphalt pavement;

[0014] Based on the asphalt pavement quality remote sensing evaluation index of each superpixel, the superpixels are divided into different levels, and an asphalt pavement quality status evaluation map of the road network area is generated; the asphalt pavement quality status evaluation map is used to display the quality rating of each superpixel at the location of the asphalt pavement.

[0015] In the above technical solution, the process of obtaining high-resolution multispectral images and panchromatic images of the road area to be evaluated includes:

[0016] Obtain high-resolution multispectral images and panchromatic images of the road network area where the road surface area to be evaluated is located;

[0017] Radiometric correction is performed on the high-resolution multispectral and panchromatic images of the road network area: the DN values ​​of the high-resolution multispectral images of the road network area are radiometrically calibrated and converted into radiometric brightness data, and atmospheric correction is performed on the calibrated images; radiometric calibration is performed on the panchromatic images of the road network area to obtain the corresponding apparent reflectance;

[0018] Orthorectify the radiometrically corrected multispectral and panchromatic images;

[0019] Fuse the ortho-processed panchromatic image and multispectral image;

[0020] The fused multispectral image and panchromatic image are cropped using the road network vector file to obtain high-resolution multispectral remote sensing images and panchromatic images of the road area.

[0021] In the above technical solution, the process of generating superpixels of the road surface area includes:

[0022] Select the initial center of the superpixel based on the panchromatic image of the road area;

[0023] A simple linear iterative clustering algorithm is executed in a loop until convergence or the maximum number of iterations is reached: each pixel is divided into the superpixel with the closest similarity distance; the similarity distance is measured using spectral distance, GLCM texture distance and spatial distance; after all pixels are divided, the center of the superpixel is updated based on all pixels within each superpixel;

[0024] A superpixel with a number of pixels less than a certain number is regarded as a tiny superpixel, and the number of neighboring superpixels M of the tiny superpixel is counted; when M = 1, the tiny superpixel is directly merged with the adjacent superpixel; when M > 1, the similarity distance between the tiny superpixel center and the neighboring superpixel center is calculated respectively, and the neighboring superpixel with the smallest gap is selected for merging.

[0025] In the above technical solution, the expression of the similarity distance D is as follows:

[0026] In the formula, max(d s ) and max(d t ) is the maximum value of the spectral distance and texture distance in the last iteration, and S is the sampling step size;

[0027] Spectral distance d s , texture distance d t , spatial distance d o The calculation formulas are:

[0028] Where: n is the number of bands of the multispectral remote sensing image of the road surface, k is the number of GLCM texture features selected, is the spectral feature of the i-th superpixel center, is the texture feature of the center of the i-th superpixel, is the spectral feature of the j-th pixel, is the texture feature of the jth pixel; (x i ,y i ) is the spatial position of the center of the i-th superpixel; (xj ,y j ) is the spatial position of the j-th pixel.

[0029] In the above technical solution, the process of cyclically executing the simple linear iterative clustering algorithm until convergence or reaching the maximum number of iterations also includes:

[0030] After dividing each pixel into its corresponding superpixel, the spectral and texture distance D between each pixel and its corresponding superpixel is calculated. pt :

[0031] When D pt When it is greater than the set threshold, it is considered that the difference between the pixel and its corresponding superpixel is too large and does not participate in the update of the superpixel center.

[0032] In the above technical solution, the process of selecting the initial center of the superpixel includes: using the Sobel operator to calculate the gradient image of the full-color image of the road area; based on the gradient image, sampling according to the set step size to select the seed point as the initial center of the superpixel, and then adjusting the seed point to the lowest point of the gradient image within the local range.

[0033] In the above technical solution, the process of removing non-asphalt road superpixels from the superpixels in the road area includes:

[0034] Multiple pure samples of non-asphalt pavement type and asphalt pavement type are selected from the high-resolution multispectral remote sensing image of the pavement area, and the spectral information corresponding to each sample is extracted as training samples;

[0035] Use training samples to train the random forest model;

[0036] Obtain the spectral feature vector of each pixel and input it into the trained random forest model to obtain the number of votes for each category for each pixel;

[0037] The number of votes for each category of all pixels in the superpixel is counted in units of superpixels, and the superpixel category is divided into the category with the most votes. The superpixels belonging to non-asphalt road types are marked and eliminated.

[0038] In the above technical solution, the average value of the full-color image grayscale corresponding to all pixels in the superpixel is used as the brightness index of the superpixel;

[0039] The information entropy of the full-color image grayscale corresponding to all pixels in the superpixel is used as the texture index of the superpixel;

[0040] The variance of the reflectivity ratio of the blue band to the near-infrared band of each pixel in the superpixel is used to represent the structural index of the pavement crack structure change;

[0041] The spectral data in the optimal characteristic spectrum or spectral combination that characterizes the health status of the road surface are subjected to different mathematical processing to establish the spectral evaluation index of the asphalt pavement; the average value of the spectral evaluation index of each pixel in the superpixel is used as the spectral index.

[0042] In the above technical solution, the process of calculating the remote sensing quality assessment index corresponding to each superpixel of the asphalt pavement in the pavement area includes:

[0043] Calculate the brightness index, texture index, structure index and spectral index of each superpixel for normalization.

[0044] Perform PCA transformation on the normalized evaluation index and obtain the first principal component;

[0045] The first principal component is normalized to obtain the superpixel-based asphalt pavement quality remote sensing evaluation index.

[0046] The present invention also provides a large-scale asphalt pavement quality rapid assessment system based on high-resolution remote sensing, which is used in the method described in the above technical solution.

[0047] The beneficial effects of the present invention are as follows: compared with traditional methods based on manual labor or inspection vehicles, the use of high-resolution remote sensing technology to evaluate a large area of ​​asphalt pavement greatly improves the efficiency of pavement quality assessment; using superpixels as the pavement quality assessment unit eliminates the salt and pepper phenomenon in traditional pixel-based assessment results, and improves the assessment accuracy and visibility of the results; the comprehensive use of four evaluation indicators of brightness, texture, structure and spectrum, compared with the traditional remote sensing pavement quality assessment that only uses spectral characteristics, the obtained assessment results are more accurate.

[0048] Furthermore, the present invention generates high-resolution multispectral remote sensing images and panchromatic images of the pavement area by fusing the corrected high-resolution multispectral images with the panchromatic images, which can fully reflect the pavement conditions of the pavement area and fully characterize the changes in spectral information of asphalt pavements of different qualities, thereby ensuring the accuracy of subsequent superpixel extraction and asphalt pavement quality evaluation.

[0049] Furthermore, the present invention obtains superpixels of the road surface area through a simple linear iterative clustering algorithm, which is simple in calculation and highly efficient while taking into account the accuracy of superpixel generation in road surface areas of different qualities.

[0050] Furthermore, the present invention proposes a method for calculating the similarity distance between pixels and superpixel centers. This method uses spectral distance, GLCM texture distance, and spatial distance to measure similarity distance, ensuring that the road quality represented by each pixel within a superpixel is consistent, further improving the accuracy of superpixel clustering.

[0051] Furthermore, the present invention sets thresholds for the spectral and texture distances between pixels and superpixel centers to eliminate pixels that differ too much from superpixels, further avoiding interference from abnormal pixels.

[0052] Furthermore, the present invention selects seed points based on the gradient image according to a set step size, and adjusts the seed points to the lowest point of the gradient image in a local range to avoid using noise points or edge points as seed points, thereby further improving the accuracy of the evaluation results.

[0053] Furthermore, the present invention trains a random forest model by selecting samples from high-resolution multispectral remote sensing images of the road surface area, ensuring that the generated random forest model fully reflects the characteristics of the road surface area being evaluated. Furthermore, the random forest is used to classify road surface superpixels and eliminate non-asphalt superpixels, effectively improving the efficiency of the evaluation process.

[0054] Furthermore, the present invention proposes a characterization method for brightness index, texture index, structure index and spectral index, which can fully reflect the quality status of asphalt pavement.

[0055] Furthermore, the asphalt pavement quality remote sensing evaluation index constructed by the present invention using PCA technology comprehensively utilizes four evaluation indicators: brightness, texture, structure and spectrum, avoids the result deviation caused by traditional linear weighting, reduces human intervention, improves the accuracy and work efficiency of pavement quality evaluation results, and is more practical. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 is a flowchart of rapid assessment of large-scale asphalt pavement quality based on high-resolution remote sensing.

[0057] Figure 2 is a flowchart of high-resolution remote sensing image preprocessing.

[0058] Figure 3 is a flowchart of superpixel segmentation of road surface images.

[0059] FIG4 is a diagram of the asphalt pavement quality assessment of the regional road network in a specific embodiment.

[0060] FIG5 is a diagram showing the quality assessment of asphalt pavement in a local road network in a specific embodiment.

[0061] Figure 6 is a diagram of the asphalt pavement quality assessment of a local road network using only spectral features. DETAILED DESCRIPTION

[0062] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments to facilitate a clear understanding of the present invention, but they do not constitute a limitation to the present invention.

[0063] As shown in FIG1 , the present invention provides a method for rapid assessment of asphalt pavement quality over a large area based on high-resolution remote sensing, comprising the following steps:

[0064] Obtain high-resolution multispectral and panchromatic images of the road surface area of ​​the road network area to be evaluated;

[0065] Generate superpixels of the road surface area based on the high-resolution multispectral image and panchromatic image of the road surface area;

[0066] Eliminate superpixels of non-asphalt pavement from superpixels in the pavement area to obtain superpixels of asphalt pavement in the pavement area;

[0067] Based on the asphalt pavement superpixels, the brightness index, texture index, structure index and spectral index of each superpixel of the asphalt pavement are calculated;

[0068] Based on the calculated brightness index, texture index, structure index and spectral index, a remote sensing quality evaluation index corresponding to each superpixel of the asphalt pavement is calculated; the remote sensing quality evaluation index is used to characterize the condition of the asphalt pavement;

[0069] Based on the asphalt pavement quality remote sensing evaluation index of each superpixel, the superpixels are divided into different levels, and an asphalt pavement quality status evaluation map of the road network area is generated; the asphalt pavement quality status evaluation map is used to display the quality rating of each superpixel at the location of the asphalt pavement.

[0070] The present invention also provides a large-scale asphalt pavement quality rapid assessment system based on high-resolution remote sensing, which is used in the method described in the above technical solution.

[0071] This specific embodiment provides a large-scale asphalt pavement quality assessment method based on high-resolution remote sensing, comprising the following steps:

[0072] The first step is preprocessing of high-resolution remote sensing images: radiometric correction, orthorectification, and image fusion are performed on the high-resolution multispectral and panchromatic images of the road network area. The processed images are then cropped using the road network vector file to obtain high-resolution multispectral and panchromatic images of the road surface area.

[0073] The spectral characteristics of remote sensing images are crucial for evaluating road quality. High-resolution remote sensing images require some preprocessing to obtain the true reflectance of the road surface spectrum. As shown in Figure 2, the specific steps are as follows:

[0074] Radiometric correction: Professional remote sensing software is used to perform radiometric calibration on the DN values ​​of the acquired multispectral images and convert them into radiometric brightness data. The calibrated images are then atmospherically corrected using the FLAASH atmospheric correction model. The panchromatic images are radiometrically calibrated to obtain apparent reflectance.

[0075] Orthorectification: Professional remote sensing software is used to perform orthorectification on the multispectral and panchromatic images after radiation correction using the image's rational polynomial file and DEM data.

[0076] Image fusion: Professional remote sensing software is used to fuse the orthophoto processed panchromatic image and multispectral image using the Gram-Schmidt method.

[0077] Road surface remote sensing image extraction: If multiple remote sensing images are required to cover the road network, the remote sensing satellite images should be mosaicked after multi-view processing; using the road network vector file, the fused multispectral image and panchromatic image should be cropped to obtain high-resolution multispectral remote sensing images and panchromatic images of the road network pavement area.

[0078] The second step is superpixel segmentation of road surface images: using high-resolution multispectral images and panchromatic images of the road surface area, a simple linear iterative clustering algorithm suitable for road surface quality assessment is used to generate superpixels as road surface quality assessment units.

[0079] As shown in Figure 3, the specific steps are as follows:

[0080] (1) Select the initial seed point: Use the full-color image of the road surface and the Sobel operator to calculate the corresponding gradient image; select the seed point as the initial center of the superpixel according to the sampling step size S, and then adjust the seed point to the lowest point of the gradient image within a local 5×5 range to avoid using noise points or edge points as seed points.

[0081] (2) Local K-means iterative clustering:

[0082] ① Calculate the distance D from each pixel to the superpixel center in the range of 2S×2S, and divide the pixel into the superpixel closest to it. To ensure that the road quality represented by each pixel in the superpixel is consistent, the spectral distance d is used. s , GLCM texture distance d t and spatial distance d o To measure the similarity distance D, the specific expression is as follows:

[0083] In the formula, max(d s ) and max(d t ) is the maximum value of the spectral distance and texture distance in the last iteration process, and the spectral distance d s , texture distance d t , spatial distance d o The calculation formulas are:

[0084] Where: n is the number of bands of the multispectral remote sensing image of the road surface, k is the number of GLCM texture features selected, is the spectral feature of the i-th superpixel center, is the texture feature of the center of the i-th superpixel, is the spectral feature of the j-th pixel spatial position, is the texture feature of the j-th pixel spatial position; (x i ,y i ) is the spatial position of the center of the i-th superpixel; (x j ,y j ) is the spatial position of the jth pixel. The values ​​of the spectral features, texture features and spatial position can be directly obtained from the high-resolution multispectral image and the panchromatic image.

[0085] ② After all pixels are divided, the average value of the spectral characteristics, GLCM texture characteristics and spatial position of all pixels in each superpixel is calculated, and the average value is used as the new center of the superpixel.

[0086] ③ Repeat steps ①-② until convergence or the maximum number of iterations is reached. To further avoid the interference of abnormal pixels, use the d obtained in the previous step s with d t , calculate the spectral and texture distances between each pixel in the superpixel and the superpixel center:

[0087] When D pt When it is greater than the set threshold, it is considered that the difference between the pixel and the superpixel is too large and does not participate in the update of the superpixel center.

[0088] (3) Post-clustering processing: To avoid isolated pixels, superpixels with a number of pixels less than a certain number are considered small superpixels, and the number of neighboring superpixels M of the small superpixel is counted. When M = 1, the small superpixel is directly merged with the neighboring superpixel. When M > 1, the similarity distance D between the center of the small superpixel and the center of the neighboring superpixel is calculated, and the neighboring superpixel with the smallest difference is selected for merging. The new center of the superpixel is the average of the spectral characteristics, GLCM texture characteristics, and spatial position of all pixels within the superpixel.

[0089] The third step is to eliminate non-asphalt road superpixels based on random forest: using high-resolution multispectral images of the road area, non-asphalt road samples such as vehicles, vegetation, shadows, bare soil, lane markings, and central dividers, as well as asphalt road samples, are selected to train a random forest model, and on this basis, non-asphalt road superpixels are marked.

[0090] High-resolution remote sensing images often contain information such as vehicles, vegetation, shadows, bare soil, lane markings, and medians on the road surface. This can cause some superpixels generated in the second step to not represent asphalt pavement, thus interfering with subsequent asphalt pavement quality assessments. Therefore, it is necessary to remove non-asphalt pavement superpixels to obtain accurate road surface superpixels. This method uses a random forest algorithm to remove these non-asphalt pavement superpixels, which includes the following steps:

[0091] (1) Training sample selection: Multiple pure samples of non-asphalt pavement types and asphalt pavement types, such as vehicles, vegetation, shadows, bare soil, lane markings, and central dividers, are selected from the multispectral remote sensing images of the road area to be evaluated, and the spectral information corresponding to each sample is extracted as the training samples of the random forest model.

[0092] (2) Random forest model training: Multiple random and repeatable samplings are performed on the input training samples to obtain multiple bootstrap sample sets. Then, using each bootstrap sample set, the samples are iteratively split into two left and right subsets according to the feature rules, and the corresponding decision trees are constructed. The above iterative training process is repeated until the set maximum tree depth is reached or until no greater information gain can be obtained by further splitting, and a trained random forest model is obtained.

[0093] (3) Obtain the spectral feature vector of each pixel as the input of the random forest model trained in the second step, so as to obtain the number of votes for each pixel belonging to each category.

[0094] Taking superpixels as units, the number of pixels in each superpixel is obtained, and the number of votes in each category of all pixels in the superpixel is counted. The superpixel category is divided into the category with the most votes. On this basis, the superpixels belonging to non-asphalt pavement are marked and do not participate in the subsequent pavement quality evaluation.

[0095] The fourth step is to calculate the asphalt pavement quality evaluation indicators based on superpixels: using superpixels as units, calculate the brightness, texture, structure and spectrum of each asphalt pavement superpixel for asphalt pavement quality evaluation.

[0096] Asphalt pavements in different health conditions exhibit distinct characteristics in remote sensing images. For example, as the asphalt age increases, the asphalt content decreases, manifesting as increasing brightness in the image. The surface texture of an aged asphalt pavement becomes rougher, and the spectral reflectance of asphalt pavements at different stages of aging also changes. To address this issue, the present invention uses superpixel-based calculations to evaluate pavement quality, using four evaluation indicators: brightness, texture, structure, and spectrum. The specific steps are as follows:

[0097] Brightness evaluation index: The average value of the full-color image grayscale corresponding to all pixels in the superpixel is used as the brightness index of the superpixel. sp The calculation formula is as follows:

[0098] Where K is the number of pixels in a superpixel; b i is the grayscale value of the panchromatic image corresponding to the i-th pixel in the superpixel, which can be directly obtained from the influence. The greater the superpixel brightness, the less asphalt content in the road surface, and the worse the corresponding road surface quality.

[0099] Texture evaluation index: The information entropy of the full-color image grayscale corresponding to all pixels in the superpixel is used as the texture index of the superpixel. sp The calculation formula is as follows:

[0100] Where M is the number of gray levels, p m is the proportion of pixels with grayscale value m in the superpixel. The larger the information entropy of a superpixel, the more heterogeneous the superpixel is, and the worse the corresponding road quality.

[0101] Structural index: The variance of the reflectivity ratio of each pixel in the superpixel between the blue band and the near-infrared band is used to characterize structural changes such as pavement cracks. Pavement structural index C based on superpixels sp The calculation formula is as follows:

[0102] Where K is the number of pixels in the superpixel, R i is the blue band reflectance of the i-th pixel in the superpixel Reflectivity in the near-infrared band The ratio of R ave It is the average value of the reflectance ratio of the blue band to the near-infrared band of all pixels in the superpixel.

[0103] The pixel values ​​of the high-resolution multispectral remote sensing image and the panchromatic image of the road area are the reflectance values. The reflectance can be directly obtained from the pre-processed image.

[0104] The larger the variance of superpixels, the greater the change in road structure and the worse the corresponding road quality.

[0105] Spectral evaluation indicators: By combining the spectral range of high-resolution remote sensing images, the optimal characteristic spectrum or spectral combination for road health is determined. Then, various mathematical methods such as ratio, logarithm, difference, normalization, and multi-band combination are used to establish spectral evaluation indicators for asphalt pavement. Taking the Worldview-2 satellite as an example, the spectral indicators that can be constructed using the reflectance of bands 1, 2, 5, 7, and 8 of the satellite are:

[0106] wait.

[0107] According to the actual scene requirements, an indicator that describes the corresponding optimal characteristic spectrum or spectral combination mathematical characteristics can be selected as the spectral index of a single pixel i, S i , then the spectral evaluation index S based on superpixel sp The calculation formula is as follows:

[0108] Where K is the number of pixels in the superpixel, S i is the spectral evaluation index value corresponding to the i-th pixel.

[0109] The fifth step is to conduct a rapid assessment of large-scale asphalt pavement using comprehensive multiple features: PCA technology is used to comprehensively analyze the four extracted evaluation indicators, and the first principal component is used to construct a superpixel-based asphalt pavement quality remote sensing assessment index, thereby obtaining a large-scale asphalt pavement quality status assessment map.

[0110] Pavement brightness, texture, structure, and spectral indicators are used to evaluate asphalt pavement quality from different perspectives. To comprehensively utilize multiple evaluation indicators, this paper uses PCA technology to construct an asphalt pavement quality remote sensing evaluation index. This avoids the result deviation caused by traditional linear weighting due to individual differences and methodological differences, thereby achieving rapid evaluation of asphalt pavements over a large area. The specific steps are as follows:

[0111] (1) Normalization of evaluation indicators: Four evaluation indicators B based on superpixel extraction sp 、E sp 、C sp and S sp Normalization is performed to eliminate the impact caused by different data ranges and dimensions of different indicators. The following method is adopted:

[0112] In the above formula, and They are evaluation indicators f sp The top 1% and bottom 1% values ​​of are normalized to 1 if the value is greater than and to 0 if the value is less than .

[0113] (2) Construction of asphalt pavement quality evaluation index: Taking superpixels as units, the normalized evaluation index is transformed using PCA and the first principal component is obtained. And Normalized to obtain the superpixel-based asphalt pavement quality remote sensing assessment index ASQRSI sp , the expression is as follows:

[0114] In the above formula, and They are The top 1% and bottom 1% values ​​of are normalized to 1 if the value is greater than and to 0 if the value is less than .

[0115] Pavement quality grading: Use the above formula to calculate the ASQRSI of each asphalt superpixel sp The superpixels are then classified into five levels using the natural breaks method: excellent, good, fair, poor, and poor. On the map of the road area to be evaluated, each asphalt superpixel's spatial location is marked with a different color to represent the corresponding level, thereby generating an asphalt pavement quality assessment map for the road network area to be evaluated.

[0116] Subsequently, highway maintenance departments can conduct targeted inspections on sections of road with poor quality (i.e., assessment grades below medium) based on the asphalt pavement quality assessment map of a large road network area, thereby quickly identifying road surfaces that need maintenance, effectively saving manpower and material costs, and providing practical guidance and effective support for road maintenance and repair decisions in the highway network.

[0117] The effects of the present invention will be further described below with reference to specific embodiments:

[0118] Taking XX District of XX City as the study area, based on WorldView-3 satellite images (imaging time is XX / XX / 20XX, the panchromatic band resolution is 0.5 meters, and the 8 multispectral bands resolution is 2 meters), the asphalt pavement quality rapid assessment method based on high-resolution remote sensing of the present invention is adopted (the spectral index adopted is NDMAI) to evaluate the quality of the asphalt pavement in the study area. The overall assessment results are shown in Figure 4, and the asphalt pavement quality assessment results of the local road network in the study area are shown in Figure 5. Figure 6 is a road network quality assessment diagram of the corresponding area using only spectral features.

[0119] According to statistics, in the asphalt pavement quality assessment diagram shown in Figure 4, the asphalt pavement areas with asphalt pavement quality assessment of excellent, good, medium, poor and bad accounted for 46.04%, 23.82%, 15.70%, 8.92% and 5.52% respectively, among which the pavement with medium and above quality accounted for more than 85%, indicating that most of the asphalt pavements in the study area have timely maintenance and are in good condition.

[0120] In addition, according to the results shown in Figure 4, the sections with better road conditions are mainly the main roads in the study area and the roads newly built in recent years, and the sections with poorer quality are mainly located in the factory areas with more truck traffic, which is consistent with the actual distribution of road network quality in the study area; it shows that the asphalt pavement quality assessment results obtained based on high-resolution remote sensing technology are consistent with the actual situation of regional asphalt pavement quality, further verifying the effectiveness of the evaluation method of the present invention.

[0121] Comparing Figures 5 and 6, it can be seen that the results obtained by the asphalt pavement quality assessment method using only spectral characteristics are limited in accuracy. In particular, the pavement quality of some main roads and secondary roads in the region with better pavement quality is assessed as poor or bad, which is inconsistent with the actual situation in the region. This shows that the asphalt pavement evaluation index constructed by the present invention using a comprehensive approach of pavement brightness, texture, structure and spectral indicators is more accurate and reliable.

[0122] More accurate and reliable road network asphalt pavement assessment maps can better guide subsequent pavement maintenance work, avoiding situations where road surfaces requiring maintenance are not represented on the assessment maps, which can lead to continued pavement deterioration and increased subsequent maintenance costs. By improving the accuracy of road network asphalt pavement assessment maps, this invention can effectively reduce maintenance costs and extend the service life of roads, resulting in significant economic benefits.

[0123] The above examples show that the pavement quality assessment method proposed in the present invention can quickly and accurately obtain a large-scale asphalt pavement quality assessment map, providing accurate decision-making information for subsequent highway network maintenance.

Claims

1. A method for rapid assessment of large-scale asphalt pavement quality based on high-resolution remote sensing, characterized by: The following steps are involved: Obtain high-resolution multispectral images and panchromatic images of the road surface area of ​​the road network area to be evaluated; Generate superpixels of the road surface area based on the high-resolution multispectral image and panchromatic image of the road surface area; Remove superpixels of non-asphalt pavement from superpixels in the pavement area to obtain superpixels of asphalt pavement in the pavement area; Based on the asphalt pavement superpixels, the brightness index, texture index, structure index and spectral index of each superpixel of the asphalt pavement are calculated; According to the calculated brightness index, texture index, structure index and spectral index, a quality remote sensing evaluation index corresponding to each superpixel of the asphalt pavement is calculated; the quality remote sensing evaluation index is used to characterize the state of the asphalt pavement; According to the asphalt pavement quality remote sensing evaluation index of each superpixel, the superpixels are divided into different levels, and an asphalt pavement quality status evaluation map of the road network area is generated; the asphalt pavement quality status evaluation map is used to display the quality rating of each superpixel location of the asphalt pavement.

2. A method according to claim 1, characterized in that: The process of obtaining high-resolution multispectral and panchromatic images of the road area to be evaluated includes: Obtain high-resolution multispectral images and panchromatic images of the road network area where the road surface area to be evaluated is located; Radiometric correction is performed on the high-resolution multispectral image and panchromatic image of the road network area: the DN value of the high-resolution multispectral image of the road network area is radiometrically calibrated, converted into radiometric brightness data, and the calibrated image is atmospherically corrected; the panchromatic image of the road network area is radiometrically calibrated to obtain the corresponding apparent reflectance; Orthorectify the multispectral and panchromatic images after radiation correction; Fusion of ortho-processed panchromatic and multispectral images; The fused multispectral image and panchromatic image are cropped using the road network vector file to obtain high-resolution multispectral remote sensing images and panchromatic images of the road area.

3. A method according to claim 1, characterized in that: The process of generating superpixels for road surface areas includes: Select the initial center of the superpixel based on the panchromatic image of the road area; A simple linear iterative clustering algorithm is executed repeatedly until convergence or the maximum number of iterations is reached: each pixel is divided into the superpixel closest to it in similarity distance; the similarity distance is measured by spectral distance, GLCM texture distance and spatial distance; after all pixels are divided, the center of the superpixel is updated based on all pixels in each superpixel; Superpixels with a number of pixels less than a certain number are regarded as tiny superpixels, and the number of neighborhood superpixels M of the tiny superpixel is counted; when M=1, the tiny superpixel is directly merged with the adjacent superpixel; when M>1, the similarity distance between the tiny superpixel center and the neighborhood superpixel center is calculated respectively, and the neighborhood superpixel with the smallest gap is selected for merging.

4. A method according to claim 3, characterized in that: The expression of the similarity distance D is as follows: In the formula, max(d s ) and max(d t ) is the maximum value of the spectral distance and the texture distance in the previous iteration, and S is the sampling step size; Spectral distance d s , texture distance d t , spatial distance d o The calculation formulas are: Where: n is the number of bands of multispectral remote sensing images of the road surface, k is the number of GLCM texture features selected, is the spectral feature of the i-th superpixel center, is the texture feature of the i-th superpixel center, is the spectral feature of the jth pixel, is the texture feature of the jth pixel; (x i ,y i ) is the spatial position of the center of the i-th superpixel; (x j ,y j ) is the spatial position of the j-th pixel.

5. A method according to claim 4, characterized in that: The process of cyclically executing the simple linear iterative clustering algorithm until convergence or reaching the maximum number of iterations also includes: After dividing each pixel into its corresponding superpixel, the spectral and texture distances D between each pixel and its corresponding superpixel are calculated. pt : When D pt When it is greater than the set threshold, it is considered that the difference between the pixel and its corresponding superpixel is too large and does not participate in the update of the superpixel center.

6. A method according to claim 3, characterized in that: The process of selecting the initial center of the superpixel includes: using the Sobel operator to calculate the gradient image of the full-color image of the road area; based on the gradient image, sampling and selecting the seed point as the initial center of the superpixel according to the set step size, and then adjusting the seed point to the lowest point of the gradient image in the local range.

7. A method according to claim 1, characterized in that: The process of removing non-asphalt road superpixels from the superpixels in the road area includes: Select multiple pure samples of non-asphalt pavement type and asphalt pavement type from the high-resolution multispectral remote sensing images of the pavement area, and extract the spectral information corresponding to each sample as training samples; Use training samples to train the random forest model; Get the spectral feature vector of each pixel and input it into the input of the trained random forest model to get the number of votes for each pixel belonging to each category; The number of votes for each category of all pixels in the superpixel is counted in units of superpixels, and the superpixel category is divided into the category with the largest number of votes, and the superpixels belonging to non-asphalt road types are marked and eliminated.

8. A method according to claim 1, characterized in that: The average value of the panchromatic image grayscale corresponding to all pixels in the superpixel is used as the brightness index of the superpixel; The information entropy of the grayscale of the full-color image corresponding to all pixels in the super-pixel is used as the texture index of the super-pixel; The variance of the ratio of the reflectivity of the blue band to the near-infrared band of each pixel in the superpixel is used to characterize the structural index of the pavement crack structure change; The spectral data in the optimal characteristic spectrum or spectral combination that characterizes the health status of the road surface are subjected to different mathematical processing to establish the spectral evaluation index of the asphalt pavement; the average value of the spectral evaluation index of each pixel in the superpixel is used as the spectral index.

9. A method according to claim 1, characterized in that: The process of calculating the quality remote sensing assessment index corresponding to each superpixel of the asphalt pavement in the pavement area includes: Calculate the brightness index, texture index, structure index and spectral index of each superpixel for normalization. The normalized evaluation index is transformed using PCA, and the first principal component is obtained; The first principal component is normalized to obtain the superpixel-based asphalt pavement quality remote sensing evaluation index.

10. A large-scale asphalt pavement quality rapid assessment system based on high-resolution remote sensing, characterized by: The system is used to implement the method described in any one of claims 1 to 9.

Images