Crack detection method and system for expressway construction

Abstract

The invention relates to a crack detection method and system for expressway construction. The method comprises the following steps: acquiring a multi-modal sensing data set containing multispectral, three-dimensional point cloud and polarized light data on the surface of a highway to be detected, generating a multi-dimensional fusion image, and classifying to obtain a detection scene type; obtaining interference judgment parameters in combination with a preset scene-threshold mapping model; then layering and filtering interference to obtain an effective fusion image; geometric and thermodynamic parameters of the crack are extracted, working condition background data such as historical detection, environment and traffic are combined, a complete-cycle comprehensive risk value and roadbed residual bearing capacity are obtained, and finally a grading detection report is generated. By adopting the method, the crack risk and the roadbed bearing capacity can be quantified, and the reliability of complex scene detection and the scientificity of maintenance decision are improved.

Description

Technical Field

[0001] This invention belongs to the field of construction crack detection technology, and in particular relates to a method and system for detecting cracks in highway construction. Background Technology

[0002] With the development of highway construction and maintenance technologies, crack detection, as a crucial link in ensuring highway structural safety and extending service life, has received widespread attention in the industry. Highway cracks are mostly caused by construction defects, material aging, excessive loads, or environmental factors. If not detected and addressed promptly and accurately, they can rapidly expand, leading to roadbed damage and pavement collapse, directly threatening traffic safety. Against this backdrop, imaging-based crack detection methods are gradually replacing manual inspections and becoming the mainstream detection method. These technologies acquire pavement images and combine them with data processing and analysis to achieve visual identification and preliminary assessment of cracks, offering advantages such as higher efficiency and wider coverage compared to manual inspections.

[0003] Existing crack detection solutions mainly revolve around the fusion of infrared thermal imaging and visible light images. For example, some technologies use guide rails to fix infrared thermal imaging and visible light imaging equipment, simultaneously acquiring infrared thermal images and visible light images of the road surface. The infrared data is then manually processed to generate temperature contour maps, which are then compared with the visible light images to identify hidden cracks. Other technologies establish crack development models, using infrared thermal images to obtain measured temperature difference data between the cracked area and the road surface. These are combined with atmospheric temperature correction models to calculate the crack development level, while simultaneously fusing grayscale and temperature information for detection. These methods have achieved preliminary crack identification in specific scenarios, providing some data support for highway maintenance.

[0004] However, current detection methods still have significant limitations: First, the data fusion dimensions are insufficient. Existing solutions only integrate infrared thermal imaging and visible light images, failing to fully utilize multispectral data and thermodynamic dynamic parameters, resulting in limited ability to identify cracks in complex environments. Second, the anti-interference capability is weak, relying on static temperature thresholds or manual corrections, and cannot respond in real time to dynamic interferences such as steam, dust, optical shadows, and surface deposits. In complex scenarios such as steam and oil stains mixing after heavy rain, the false alarm rate is high. Third, there is a lack of a quantitative assessment system. Cracks are qualitatively judged only through temperature difference or grayscale information, making it difficult to accurately quantify the degree of damage and expansion trend of cracks, causing inconvenience to subsequent maintenance decisions and easily leading to engineering safety hazards due to assessment deviations. Summary of the Invention

[0005] Therefore, it is necessary to provide a method and system for crack detection in highway construction to address the aforementioned technical problems.

[0006] Firstly, this application provides a method for detecting cracks in highway construction, including:

[0007] Acquire a multimodal sensing dataset of the surface of the highway to be inspected; the multimodal sensing dataset includes multispectral data, 3D point cloud data, and polarized light data;

[0008] Generate multi-dimensional fused images based on multimodal sensing datasets;

[0009] The multi-dimensional fused image is classified to obtain the current detected scene type;

[0010] Based on the detection scene type, and combined with the preset scene-threshold mapping model, interference judgment parameters are obtained;

[0011] Based on the interference determination parameters, the multi-dimensional fused image is subjected to hierarchical interference filtering according to priority logic to obtain an effective fused image;

[0012] Geometric and thermodynamic parameters of cracks are extracted from the effectively fused images and combined with background data to obtain the comprehensive risk value and remaining bearing capacity of the subgrade throughout the entire cycle; the background data includes historical monitoring data, environmental data and traffic data.

[0013] Based on the comprehensive risk value throughout the entire life cycle and the remaining bearing capacity of the roadbed, a graded inspection report is generated for the expressway to be inspected.

[0014] In one embodiment, a multi-dimensional fused image is generated based on a multimodal sensing dataset, including:

[0015] The crack region point cloud is segmented from the 3D point cloud data to obtain independent point cloud clusters, and the independent point cloud clusters are reconstructed in 3D to generate a 3D surface model of the crack.

[0016] Extract the three-dimensional geometric parameters of the three-dimensional surface model;

[0017] Based on polarized light data, a polarization degree distribution map of the crack region is obtained; the polarization degree distribution map is used to characterize the difference in polarization reflection between the crack and the road matrix.

[0018] Multispectral data, three-dimensional geometric parameters, and polarization degree distribution maps are fused to generate a multi-dimensional fused image.

[0019] In one embodiment, multispectral data, three-dimensional geometric parameters, and polarization degree distribution maps are fused to generate a multidimensional fused image, including:

[0020] Multispectral data is reconstructed to generate multispectral images;

[0021] The RANSAC registration algorithm is used to map and match the pixel coordinates of the crack's three-dimensional spatial coordinates and polarization degree distribution map corresponding to the three-dimensional geometric parameters with the pixels of the multispectral image, thereby obtaining a multimodal spatial alignment feature set;

[0022] A multi-modal spatially aligned feature set is subjected to pixel-level feature weighting and fusion to generate a multi-dimensional fused image.

[0023] In one embodiment, based on the detection scene type and combined with a preset scene-threshold mapping model, interference determination parameters are obtained, including:

[0024] Calculate the interference determination parameters using the following formula:

[0025]

[0026] in, These are interference detection parameters. It is the initial value of the correlation coefficient threshold obtained by calibration using historical data under a standard dry, sunny day scenario. It is the near-infrared dust concentration factor. It is the mean of the near-infrared dust concentration factor in the historical dataset. It is the standard deviation of the near-infrared dust concentration factor in the historical dataset. It is the long-wave infrared temperature gradient. This is a typical temperature gradient value under standard dry, sunny conditions. It is the influence factor of polarized light droplets. It is the scene-adaptive weighting coefficient.

[0027] In one embodiment, scene classification is performed on the multi-dimensional fused image to obtain the current detected scene type, including:

[0028] Near-infrared band reflection intensity, long-wave infrared temperature distribution, and polarized light scattering intensity of the multi-dimensional fused image are extracted as environmental feature vectors.

[0029] The environmental feature vector is input into a pre-trained convolutional neural network classification model to obtain the current detection scene type; the current detection scene type includes at least one of the following: dry sunny day, after rainstorm, dusty day, foggy day, and icy day.

[0030] In one embodiment, the geometric and thermodynamic parameters of the cracks are extracted from the effectively fused image, and combined with the background data of the working conditions, the comprehensive risk value and the remaining bearing capacity of the subgrade throughout the entire cycle are obtained, including:

[0031] Based on the effectively fused images, the initial geometric morphology parameters and initial thermodynamic parameters of the crack are extracted; the initial thermodynamic parameters include the heat diffusion area calculated based on the standard heat conduction model.

[0032] Obtain the pavement material type of the highway to be tested, and determine the corresponding material thermal conductivity calibration factor according to the preset material type-thermal conductivity mapping relationship;

[0033] The thermal diffusion area is calibrated based on the material thermal conductivity calibration factor to obtain the calibrated thermal diffusion area.

[0034] From the long-wave infrared band data of the effectively fused image, the temperature gradient variance of the crack region is calculated, and the temperature gradient variance is determined as a parameter of heat conduction uniformity.

[0035] Based on near-infrared band data, measurement data from areas obscured by foreign objects in the layered measurement of crack depth are removed to obtain effective crack width-depth sequence data. The crack width-depth sequence data is then subjected to piecewise smooth fitting to obtain the variation gradient. The variation gradient is used to characterize the crack morphology's expansion characteristics in the vertical direction.

[0036] Based on the calibrated thermal diffusion area, thermal conduction uniformity parameters and variation gradient, combined with background data of the working conditions, the comprehensive risk value and the remaining bearing capacity of the subgrade throughout the entire cycle are obtained.

[0037] In one embodiment, the method further includes:

[0038] Based on the comprehensive risk value throughout the entire cycle and three-dimensional geometric parameters, combined with the preset subgrade type, recommended repair schemes are obtained by matching from the preset crack feature-repair scheme mapping database; the recommended repair schemes include material selection, construction technology and estimated construction period parameters;

[0039] After the repair work is completed, a second inspection is conducted on the repaired area at a predetermined inspection period to obtain the comprehensive risk factor after repair.

[0040] Based on the comprehensive risk coefficient after repair and the comprehensive risk value throughout the entire cycle, the risk coefficient reduction rate is obtained;

[0041] If the risk reduction rate is lower than the preset standard threshold, a secondary repair plan adjustment instruction is generated and sent to the maintenance decision terminal; the secondary repair plan adjustment instruction is used to instruct the initiation of the repair plan re-matching process based on the post-repair detection data.

[0042] Secondly, this application also provides a crack detection system for highway construction, comprising:

[0043] The data acquisition module is used to acquire multimodal sensing datasets of the surface of the highway to be inspected; the multimodal sensing datasets include multispectral data, three-dimensional point cloud data, and polarized light data;

[0044] The fusion processing module is used to generate multi-dimensional fused images based on multimodal sensor datasets;

[0045] The scene intelligent recognition module is used to classify scenes from multi-dimensional fused images to obtain the current detected scene type;

[0046] The threshold decision module is used to obtain interference judgment parameters based on the detection scene type and a preset scene-threshold mapping model.

[0047] The interference filtering module is used to perform hierarchical interference filtering on the multi-dimensional fused image according to the interference determination parameters, so as to obtain an effective fused image.

[0048] The analysis module is used to extract the geometric and thermodynamic parameters of cracks from the effectively fused images, and combine them with the background data of the working conditions to obtain the comprehensive risk value and the remaining bearing capacity of the subgrade throughout the entire cycle; the background data of the working conditions includes historical detection data, environmental data and traffic data;

[0049] The report generation module is used to generate a graded inspection report for the highway to be inspected based on the comprehensive risk value throughout the entire cycle and the remaining bearing capacity of the roadbed.

[0050] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0051] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0052] The aforementioned method, system, equipment, and medium for crack detection in highway construction involves acquiring a multimodal sensing dataset of the highway surface to be inspected. This dataset includes multispectral data, 3D point cloud data, and polarized light data. Based on the multimodal sensing dataset, a multidimensional fused image is generated. The multidimensional fused image is then classified to determine the current detection scene type. Based on the detection scene type and a pre-defined scene-threshold mapping model, interference judgment parameters are obtained. According to the interference judgment parameters, the multidimensional fused image undergoes hierarchical interference filtering based on priority logic to obtain an effective fused image. The geometric and thermodynamic parameters of the cracks are extracted from the effective fused image, and combined with background data, the overall risk value and remaining bearing capacity of the subgrade are obtained throughout the entire lifecycle. The background data includes historical inspection data, environmental data, and traffic data.

[0053] Based on the comprehensive risk value throughout the entire life cycle and the remaining bearing capacity of the subgrade, a graded inspection report is generated for the highway to be inspected. This method can quantify crack risk and subgrade bearing capacity, improving the reliability of inspection in complex scenarios and the scientific nature of maintenance decisions. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 A schematic diagram illustrating the implementation environment of a crack detection method for highway construction, provided as an exemplary embodiment of this application;

[0056] Figure 2 A schematic flowchart illustrating a crack detection method for highway construction, provided as an exemplary embodiment of this application;

[0057] Figure 3 for Figure 2 This is a schematic diagram of the structure of a crack detection system for highway construction, provided as an exemplary embodiment of this application. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0059] The implementation environment of the embodiments of this application will be described. For illustrative purposes, please refer to... Figure 1 The implementation environment includes a detection terminal 101, a data acquisition terminal 102, and a maintenance decision terminal 103. The detection terminal 101 is connected to the data acquisition terminal 102 and the maintenance decision terminal 103 via a network to receive multimodal sensor data sets of the highway surface to be inspected from the data acquisition terminal 102, and promptly sends graded inspection reports, recommended repair plans, and secondary repair plan adjustment instructions to the maintenance decision terminal 103. The network connection can be a wired network or a wireless network.

[0060] The detection terminal 101 can be a computer device, such as a desktop computer or laptop, used to receive, process, and analyze the multimodal sensor dataset of the highway surface to be inspected from the acquisition terminal 102. The detection terminal 101 can have strong computing and data processing capabilities to meet the data processing needs during the inspection process.

[0061] The acquisition terminal 102 can be a multimodal sensing integrated detection device, used to acquire multimodal sensing datasets of the surface of the highway to be inspected at close range, and transmit the acquired multimodal sensing datasets to the detection terminal 101 in real time via a network.

[0062] The maintenance decision terminal 103 can be a computer device, such as a desktop computer or a laptop, used to receive graded inspection reports, recommended repair plans, and secondary repair plan adjustment instructions sent by the inspection terminal 101 in real time via the network, providing maintenance personnel with decision-making basis and realizing closed-loop management of inspection and maintenance.

[0063] The crack detection method for highway construction provided in this application embodiment can be used for various stages such as pavement paving / pouring quality inspection, subgrade compaction quality verification, bridge / tunnel connection section construction, key project special acceptance, daily inspection, emergency detection after special weather, special monitoring of high-load road sections, repair plan formulation and crack expansion early warning.

[0064] This embodiment of the invention provides a method for detecting cracks in highway construction, which can also be applied to other application scenarios. It is only used as an example and is not intended to limit the specific application scenarios.

[0065] In one embodiment, such as Figure 2 As shown, a method for crack detection in highway construction is provided. This embodiment applies this method to... Figure 1 Taking the detection terminal 101 as an example, in this embodiment, the method includes the following steps:

[0066] Step S201: Obtain the multimodal sensing dataset of the highway surface to be detected; the multimodal sensing dataset includes multispectral data, three-dimensional point cloud data and polarized light data.

[0067] The multimodal sensing dataset can be a comprehensive dataset formed by simultaneously collecting multispectral data, 3D point cloud data, and polarized light data from the surface of the highway to be inspected using various types of sensing devices. During the acquisition process, the acquisition terminal can use mobile inspection vehicles, drones, or roadside fixed supports to capture multidimensional information of the highway surface at close range. The multispectral data reflects the spectral response of the road surface in different bands, the 3D point cloud data records the 3D spatial morphology of the road surface, and the polarized light data presents the polarization reflection characteristics of the road surface.

[0068] Specifically, the detection terminal can receive multimodal sensing datasets transmitted in real time by the acquisition terminal via the network. The multimodal sensing datasets are generated synchronously by the multispectral sensor, three-dimensional lidar, and polarization imaging equipment mounted on the acquisition terminal.

[0069] Step S202: Generate a multi-dimensional fused image based on the multimodal sensing dataset.

[0070] Among them, the multi-dimensional fused image can be a comprehensive image generated by the detection terminal after processing the collected multispectral data, three-dimensional point cloud data and polarized light data.

[0071] Specifically, the detection terminal can extract the core features of various types of data from the multimodal sensing dataset, then use spatial alignment processing to accurately match the feature positions of data from different dimensions, and then use feature weighted fusion to organically integrate these multi-dimensional features into the same image frame, ultimately generating a multi-dimensional fused image.

[0072] Step S203: Perform scene classification on the multi-dimensional fused image to obtain the current detected scene type.

[0073] Among them, the current detection scenario type can be the actual environmental category in which the highway crack detection is located, determined by the detection terminal after performing environmental feature analysis on the multi-dimensional fused image.

[0074] Specifically, the detection terminal can extract key features that reflect the environmental conditions from multi-dimensional fused images, generate a set of feature information that characterizes the current detection environment, and then the detection terminal completes the analysis and matching of environmental features, automatically determines the scene type in which the current detection is taking place, and obtains the current detection scene type.

[0075] Step S204: Based on the detection scene type and combined with the preset scene-threshold mapping model, obtain the interference judgment parameters.

[0076] The preset scene-threshold mapping model can adopt a layered architecture of baseline threshold, scene adaptation, and dynamic calibration. The bottom layer is the baseline threshold layer, which stores the initial threshold calibrated with historical data under standard environmental scenarios as the basic baseline for threshold adjustment. The middle layer is the scene mapping rule layer, which has built-in logic for the association between different detection scene types and environmental interference factors, as well as a scene adaptive weight configuration mechanism, to establish the correspondence between scenes and threshold adjustment rules. The top layer is the dynamic calculation layer, which integrates a quantization algorithm module to perform calculations on the input environmental feature data to achieve dynamic calibration and optimization of the baseline threshold and generate interference judgment parameters adapted to the current scene. Basic configuration input: baseline threshold calibrated with historical data under standard environmental scenarios; Scene type input: current detection scene type identified by the detection terminal; Environmental feature input: key environmental data reflecting the current scene interference status; Rule parameter input: preset scene adaptive weight coefficients and calculation rules for various environmental interference factors, outputting interference judgment parameters adapted to the current detection scene.

[0077] Interference determination parameters are core quantitative indicators used to quantify the degree of environmental interference in multi-dimensional fused images. These parameters are calculated by the detection terminal based on the current detection scene type using a preset scene-threshold mapping model.

[0078] Specifically, the detection terminal can retrieve a preset scene-threshold mapping model, input the baseline threshold under standard conditions, the identified current detection scene type, key environmental data reflecting scene interference, and preset rule parameters. Through the model's hierarchical architecture and quantization calculations, the baseline threshold is adapted to the scene and dynamically calibrated, ultimately generating interference judgment parameters that can accurately match the current level of environmental interference.

[0079] Step S205: Based on the interference determination parameters, perform hierarchical interference filtering on the multi-dimensional fused image according to priority logic to obtain an effective fused image.

[0080] Among them, layered interference filtering can be a processing method in which the detection terminal, based on interference judgment parameters and according to preset priority logic, performs layered and targeted removal of environmental interference information of different types and intensities in the multi-dimensional fused image. The preset priority logic can prioritize the interference factors in the current scene (such as dust, steam, fog droplets, surface attachments, etc.) according to their degree of influence, and then adapt the corresponding filtering strategy according to the characteristics of different levels of interference—prioritizing the deep filtering of major interferences that affect the detection accuracy (such as high concentration of dust), moderately correcting minor interferences (such as slight shadows), and finally stripping away all interference information that is irrelevant to the crack features, outputting an effective fused image that retains only the true features of the crack.

[0081] An effective fused image can be a high-purity comprehensive image that retains only the true crack features on the highway surface after the detection terminal performs hierarchical interference filtering on the multi-dimensional fused image according to the interference judgment parameters and priority logic.

[0082] Specifically, the detection terminal can identify the degree of influence of various environmental interferences in the multi-dimensional fused image and classify them according to the interference judgment parameters. It can prioritize the deep filtering of major interferences with high influence and make appropriate corrections to minor interferences with low influence. Through targeted and adapted filtering strategies, it can process layer by layer, remove all interference information that is not related to crack features, and finally obtain a high-purity effective fused image that clearly retains the true crack features of the highway surface.

[0083] Step S206: Extract the geometric and thermodynamic parameters of the cracks from the effectively fused images, and combine them with the background data of the working conditions to obtain the comprehensive risk value of the whole cycle and the remaining bearing capacity of the subgrade; the background data of the working conditions includes historical detection data, environmental data and traffic data.

[0084] Among them, the geometric and thermodynamic parameters of the crack can be a set of core parameters extracted from the effectively fused images, which can quantitatively characterize the physical morphology and thermal conductivity of the crack.

[0085] Background data can be a set of full-cycle auxiliary information related to the generation, development and bearing capacity of cracks in the highway under test.

[0086] The full-cycle comprehensive risk value can be a comprehensive quantitative indicator used to characterize the degree of damage and the risk of expansion of cracks in highways throughout their entire life cycle.

[0087] The remaining bearing capacity of a roadbed can be a quantitative indicator of the remaining capacity of a highway roadbed to safely withstand the design load under the current state of crack damage and subsequent service conditions.

[0088] Specifically, the detection terminal can extract core parameters of the physical morphology and thermal conductivity of cracks from effectively fused images, and then retrieve full-cycle background information including historical detection, environmental and traffic data, perform correlation analysis and quantitative calculation, and finally generate a full-cycle comprehensive risk value that characterizes the damage and expansion risk of cracks throughout their entire life cycle, as well as the remaining bearing capacity of the roadbed that reflects the current remaining safe bearing capacity of the roadbed.

[0089] Step S207: Based on the comprehensive risk value throughout the entire cycle and the remaining bearing capacity of the roadbed, generate a graded inspection report for the expressway to be inspected.

[0090] Among them, the graded inspection report can be a comprehensive report that characterizes the crack condition and subgrade safety level of the highway to be inspected.

[0091] Specifically, the testing terminal can classify the degree of crack damage, expansion risk, and roadbed safety level of different road sections based on the comprehensive risk value throughout the entire life cycle and the remaining bearing capacity of the roadbed, referring to preset grading standards. It integrates key information such as the testing scope, scenario type, and core parameters to generate a clearly structured and detailed graded testing report. The preset grading standards can be a pre-defined and stored quantitative evaluation rule system for classifying the degree of crack damage, expansion risk, and roadbed safety level of highways. The preset grading standards establish the correlation between the numerical range, grade labels, and corresponding assessment conclusions of the comprehensive risk value throughout the entire life cycle and the remaining bearing capacity of the roadbed. By setting multiple sets of critical numerical thresholds, the combined results of the two core indicators are divided into different levels (such as high, medium, and low risk / safety levels), clearly defining the degree of crack hazard, roadbed bearing redundancy status, and maintenance priority requirements corresponding to different levels.

[0092] The aforementioned method for crack detection during highway construction involves a detection terminal acquiring a multimodal sensor dataset of the highway surface to be inspected. This dataset includes multispectral data, 3D point cloud data, and polarized light data. Based on the multimodal sensor dataset, a multidimensional fused image is generated. The multidimensional fused image is then classified to determine the current detection scene type. Based on the detection scene type and a preset scene-threshold mapping model, interference judgment parameters are obtained. According to the interference judgment parameters, the multidimensional fused image undergoes hierarchical interference filtering based on priority logic to obtain an effective fused image. Geometric and thermodynamic parameters of the cracks are extracted from the effective fused image, and combined with background data, a comprehensive risk value and the remaining bearing capacity of the subgrade are obtained over the entire lifecycle. The background data includes historical inspection data, environmental data, and traffic data. Based on the comprehensive risk value and the remaining bearing capacity of the subgrade, a graded inspection report for the highway to be inspected is generated. In one embodiment...

[0093] In one embodiment, generating a multi-dimensional fused image based on a multimodal sensing dataset may include the following steps:

[0094] Step S301: The crack region point cloud is segmented from the three-dimensional point cloud data to obtain independent point cloud clusters, and the independent point cloud clusters are reconstructed in three dimensions to generate a three-dimensional surface model of the crack.

[0095] Among them, three-dimensional point cloud data can be a dataset composed of massive discrete three-dimensional coordinate points, used to accurately characterize the three-dimensional spatial morphological features of the surface of the highway to be detected.

[0096] An independent point cloud cluster can be a discrete point cloud set that contains only a single crack region after the detection terminal performs crack region point cloud segmentation on the original 3D point cloud data.

[0097] A three-dimensional surface model can be a digital model that is based on independent point cloud clusters and constructed through a three-dimensional reconstruction algorithm, and can intuitively reflect the three-dimensional morphology of cracks.

[0098] Specifically, the detection terminal first processes the massive amount of discrete three-dimensional coordinate points to form a three-dimensional point cloud. It then uses a point cloud segmentation algorithm to identify and separate the point cloud belonging to the crack area, removes the road background and irrelevant point clouds, and obtains an independent point cloud cluster containing only a single crack area. Subsequently, based on the independent point cloud cluster, it performs spatial fitting and contour restoration on the discrete points to construct a digital three-dimensional surface model that can intuitively present the three-dimensional morphology, surface contour and depth distribution of the crack.

[0099] Step S302: Extract the three-dimensional geometric parameters of the three-dimensional surface model.

[0100] Among them, the three-dimensional geometric parameters can be a set of core parameters extracted from the three-dimensional surface model of the crack, which can accurately quantify the spatial structure and morphological characteristics of the crack.

[0101] Specifically, the detection terminal can analyze the three-dimensional surface model of the crack and extract core parameters that can accurately quantify the spatial structure and morphological characteristics of the crack, including basic dimensional parameters such as the actual length of the crack extending along the road surface, the maximum and average width of the opening, and the maximum and average depth perpendicular to the road surface, as well as spatial morphological parameters such as the direction of the crack's extension, surface undulation, and cross-sectional shape characteristics, to generate a complete set of three-dimensional geometric parameters.

[0102] Step S303: Based on polarized light data, obtain the polarization degree distribution map of the crack area; the polarization degree distribution map is used to characterize the difference in polarization reflection between the crack and the road substrate.

[0103] Among them, polarized light data can be optical data that characterizes the polarization reflection characteristics of the surface of the highway to be tested.

[0104] A polarization degree distribution map can be a visual result generated by the detection terminal based on polarized light data processing, which intuitively presents the difference in polarization reflection between the crack area and the road matrix in the form of an image.

[0105] Specifically, the detection terminal can extract key information reflecting the polarization reflection state of different areas of the road surface, and then analyze and calculate the difference in polarization reflection between the crack area and the road matrix. The calculation results are converted into images that represent the degree of polarization with different pixel gray levels or colors, generating a polarization distribution map that can intuitively present the crack boundary, outline and extension range.

[0106] Step S304: Feature fusion of multispectral data, three-dimensional geometric parameters and polarization degree distribution map to generate a multi-dimensional fused image.

[0107] Specifically, the detection terminal can map and match the three-dimensional spatial coordinates of the crack corresponding to the three-dimensional geometric parameters, the pixel coordinates of the polarization degree distribution map, and multispectral data to obtain a spatially aligned multimodal feature set. Finally, the multimodal feature set is subjected to pixel-level feature weighted fusion, integrating the three core features of spectroscopy, geometry, and polarization to generate a multidimensional fused image that can comprehensively characterize the multidimensional information of the crack.

[0108] In this embodiment, the detection terminal performs layered processing and multi-dimensional feature fusion on three-dimensional point cloud data, polarized light data, and multispectral data. It integrates the core information of crack spatial morphology, polarization reflection, and spectral response to generate a fused image that can comprehensively and accurately characterize the multi-dimensional features of cracks, laying a reliable data foundation for crack identification and quantitative analysis in subsequent complex scenarios.

[0109] In one embodiment, multispectral data, three-dimensional geometric parameters, and polarization degree distribution maps are fused to generate a multidimensional fused image, including:

[0110] Step S401: The multispectral data is reconstructed to generate a multispectral image.

[0111] Among them, the multispectral image can be a visualized image generated by the detection terminal after performing image reconstruction processing on the multispectral data, which can simultaneously present the spectral response characteristics of multiple bands on the surface of the highway to be detected.

[0112] Specifically, the detection terminal can receive multispectral data of the highway surface to be inspected. The multispectral data contains spectral response information of multiple different wavelengths (such as visible light, near infrared, long-wave infrared, etc.). Then, the discrete spectral data is preprocessed and pixel mapped by the image reconstruction algorithm, and the spectral signals of different bands are converted into corresponding image pixel information. The multi-band spectral response values ​​of each pixel are integrated to generate a visualized multispectral image that can simultaneously present the multi-band spectral characteristics of the road surface and clearly distinguish the differences between cracks and background matrix.

[0113] Step S402: The RANSAC registration algorithm is used to map and match the pixel coordinates of the crack three-dimensional spatial coordinates and polarization degree distribution map corresponding to the three-dimensional geometric parameters with the pixels of the multispectral image to obtain a multimodal spatial alignment feature set.

[0114] Among them, the RANSAC registration algorithm is a robust algorithm for estimating mathematical model parameters and achieving spatial alignment of multi-source data in datasets with noise and outliers.

[0115] A multimodal spatial alignment feature set can be a unified feature set generated by matching features from multiple sources in spatial location.

[0116] Specifically, the detection terminal can use the RANSAC registration algorithm to obtain the optimal matching model through random sampling fitting and interior point screening iteration. It can accurately spatially map and match the three-dimensional spatial coordinates of the crack corresponding to the three-dimensional geometric parameters and the pixel coordinates of the polarization degree distribution map with the pixels of the multispectral image, eliminate matching outliers caused by interference, and make the feature points of the same crack position in different modal data correspond one by one, generating a multimodal spatial alignment feature set that integrates spectral, geometric and polarization features.

[0117] Step S403: Perform pixel-level feature weighting and fusion on the multimodal spatial alignment feature set to generate a multidimensional fused image.

[0118] Specifically, the detection terminal can assign weights to the spectral features, three-dimensional geometric features, and polarization reflection features in the multimodal spatial alignment feature set (based on the preset weight coefficients of each feature's contribution to crack identification). Then, for each pixel of the multispectral image, it fuses its corresponding three-dimensional spatial coordinate information and polarization degree data, integrates the effective information of the three types of features through pixel-level weighted operations, weakens redundant interference, and generates a multi-dimensional fusion image that can comprehensively and accurately characterize the multi-dimensional attributes of cracks.

[0119] In this embodiment, the detection terminal generates a multispectral image by reconstructing multispectral data, uses the RANSAC registration algorithm to achieve spatial alignment of multi-source data, and then integrates the three core features of spectrum, geometry and polarization through pixel-level feature weighting and fusion to generate a multi-dimensional fusion image that can comprehensively and accurately characterize the multi-dimensional attributes of cracks, providing high-quality data support for subsequent scene classification and accurate crack identification.

[0120] In one embodiment, based on the detection scene type and combined with a preset scene-threshold mapping model, interference determination parameters are obtained, including:

[0121] Calculate the interference determination parameters using the following formula:

[0122]

[0123] in, These are interference detection parameters. It is the initial value of the correlation coefficient threshold obtained by calibration using historical data under a standard dry, sunny day scenario. It is the near-infrared dust concentration factor. It is the mean of the near-infrared dust concentration factor in the historical dataset. It is the standard deviation of the near-infrared dust concentration factor in the historical dataset. It is the long-wave infrared temperature gradient. This is a typical temperature gradient value under standard dry, sunny conditions. It is the influence factor of polarized light droplets. It is the scene-adaptive weighting coefficient.

[0124] It can be generated based on historical detection datasets. By collecting a large amount of crack detection data under standard dry sunny day scenarios, analyzing the feature distinction threshold between cracks and background, and determining the benchmark threshold after statistical verification, it serves as the basic reference value for threshold adjustment under different scenarios and is stored in the preset parameter library of the detection terminal.

[0125] It can be extracted and calculated from near-infrared band data of multi-dimensional fused images. The detection terminal performs spectral analysis on the near-infrared band of the fused image, and calculates the intensity of dust interference in the current scene based on the scattering and absorption characteristics of near-infrared light by dust using a spectral intensity attenuation model.

[0126] and It can be obtained based on historical detection datasets. By extracting all near-infrared dust concentration factor samples from detection data of different past scenarios (including dry sunny days, dusty days, etc.), the mean value can be obtained through statistical calculation. and standard deviation As the current The normalized reference standard is pre-stored in the testing terminal.

[0127] It can be calculated from the long-wave infrared band data of the multi-dimensional fused image. The detection terminal extracts the long-wave infrared temperature data of the crack area and the surrounding road matrix in the fused image, and calculates the temperature gradient by the temperature difference between the two, which is used to reflect the impact of temperature interference on crack recognition in the current scene.

[0128] Calibration can be based on historical standard scenario data. This involves collecting a large number of long-wave infrared temperature gradient samples of the cracked area and the road surface substrate under standard dry, sunny weather conditions, and then statistically averaging these samples to determine typical values, which can then be used as the current benchmark. The reference standard is stored in the testing terminal.

[0129] It can be extracted and calculated from the polarization degree distribution map. The detection terminal analyzes the influence of fog droplets on the polarization light reflection characteristics in the polarization degree distribution map, and obtains it through the polarization degree attenuation coefficient. It is used to quantify the degree of interference of fog droplets on the identification of crack polarization features in the current scene.

[0130] In this embodiment, the detection terminal calls a preset formula, integrates historical calibration benchmark values, real-time extracted scene interference factors and scene adaptive weight coefficients, dynamically calculates interference judgment parameters adapted to the current scene, achieves accurate adaptation of interference threshold, and improves anti-interference capability in complex scenes.

[0131] In one embodiment, classifying the multi-dimensional fused image to obtain the current detected scene type may include the following steps:

[0132] Step S501: Extract the near-infrared band reflection intensity, long-wave infrared temperature distribution, and polarized light scattering intensity from the multi-dimensional fused image as environmental feature vectors.

[0133] Among them, the environmental feature vector can be a high-dimensional data vector extracted by the detection terminal from the multi-dimensional fused image to characterize the current environmental state of the detection scene. Its core consists of near-infrared band reflection intensity, long-wave infrared temperature distribution, and polarized light scattering intensity.

[0134] Specifically, the detection terminal can perform band-wise analysis on multi-dimensional fused images, extract reflection intensity data reflecting the degree of optical interference such as dust and adhering objects from the near-infrared band, obtain temperature distribution information reflecting the temperature field characteristics of the scene from the long-wave infrared band, and extract scattering intensity parameters characterizing the polarization interference state such as fog droplets and humidity from the polarization-related band. These data are then integrated into a structured high-dimensional data vector to generate an environmental feature vector.

[0135] Step S502: Input the environmental feature vector into the pre-trained convolutional neural network classification model to obtain the current detection scene type; the current detection scene type includes at least one of the following: dry sunny day, after rainstorm, dusty day, foggy day, and icy day.

[0136] The pre-trained convolutional neural network classification model can be a deep learning model trained and optimized from a large number of environmental feature vector samples of different scenarios (including dry sunny days, after rainstorms, dusty days, foggy days, and icy days). Its structure includes an input layer, multiple convolutional and pooling layers, fully connected layers, and an output layer. The convolutional layers are responsible for extracting deep semantic features from the environmental feature vectors, the pooling layers achieve feature dimensionality reduction and key information preservation, and the fully connected layers complete feature mapping and classification decisions. The function of the convolutional neural network classification model is to identify the environmental category of the current detection scene based on the input environmental feature vector. The input is an environmental feature vector representing the environmental state of the detection scene (including near-infrared band reflection intensity, long-wave infrared temperature distribution, and polarized light scattering intensity data), and the output is any or a combination of scene types such as dry sunny days, after rainstorms, dusty days, foggy days, and icy days.

[0137] Specifically, the detection terminal can call a pre-trained convolutional neural network classification model that has been trained and optimized by a large number of feature vector samples of different scene environments to obtain any or a combination of scene types such as dry sunny day, after rainstorm, dusty day, foggy day and freezing day, and determine the current detection scene type.

[0138] In this embodiment, the detection terminal extracts multi-dimensional environmental features from multi-dimensional fused images to construct an environmental feature vector, and then inputs it into a pre-trained convolutional neural network classification model for scene recognition. This enables the determination of various detection scenarios, such as dry sunny days and after heavy rain, providing a reliable basis for interference suppression in subsequent scene adaptation.

[0139] In one embodiment, extracting the geometric and thermodynamic parameters of the cracks from the effectively fused image and combining them with background data of the working conditions to obtain the comprehensive risk value and the remaining bearing capacity of the subgrade throughout the entire cycle may include the following steps:

[0140] Step S601: Based on the effectively fused image, extract the initial geometric morphology parameters and initial thermodynamic parameters of the crack; the initial thermodynamic parameters include the heat diffusion area calculated based on the standard heat conduction model.

[0141] Among them, the initial geometric morphology parameters can be a set of parameters that can be directly extracted by the detection terminal from the effectively fused image and can preliminarily quantify the surface and spatial basic morphological features of the crack.

[0142] The initial thermodynamic parameters can be a set of core parameters that characterize the thermal conductivity of the crack region, calculated by the detection terminal based on the effectively fused image and a preset standard thermal conduction model. These parameters reflect the difference in heat exchange between the crack and the surrounding pavement matrix. Illustratively, the preset standard thermal conduction model can be a physical-mathematical model based on Fourier's law of heat conduction, used to quantify the heat diffusion and transfer characteristics of the crack region. The standard thermal conduction model structure includes a thermal conduction control equation, a boundary condition setting module, and a parameter solving module. The control equation describes the heat transfer law between the crack and the pavement matrix, the boundary condition module defines the heat exchange boundary between the crack and the surrounding environment (such as thermal radiation and thermal convection coefficients), and the solving module performs numerical calculations using the finite element method or finite difference method. The function of the standard thermal conduction model is to calculate the heat diffusion-related parameters of the crack region based on the temperature information extracted from the effectively fused image. The inputs are the temperature distribution data of the crack region and the surrounding pavement matrix in the effectively fused image, and the basic thermal conductivity of the pavement matrix (pre-stored in the detection terminal). The outputs are the heat diffusion area (core output) characterizing the thermal conduction characteristics of the crack, as well as auxiliary parameters such as heat flux density and temperature gradient.

[0143] Specifically, the detection terminal can perform feature analysis on the effectively fused images and directly extract initial geometric parameters that can preliminarily quantify the surface and spatial foundation morphology of the cracks; at the same time, it can extract the temperature distribution data of the crack area and the surrounding pavement matrix in the image, call the pre-stored pavement matrix foundation thermal conductivity, input the standard thermal conduction model based on Fourier's law of thermal conduction, and calculate the initial thermodynamic parameters such as the thermal diffusion area characterizing the thermal conduction properties of the cracks.

[0144] Step S602: Obtain the pavement material type of the highway to be tested, and determine the corresponding material thermal conductivity calibration factor according to the preset material type-thermal conductivity coefficient mapping relationship.

[0145] Among them, the road surface material type can be the type of material used in the road surface structure of the highway to be tested.

[0146] The preset material type-thermal conductivity coefficient mapping relationship can be a structured association data (such as a mapping table or database) pre-stored in the testing terminal that establishes calibration rules for road material types and corresponding thermal conductivity coefficients.

[0147] The material thermal conductivity calibration factor can be a proportional coefficient or correction parameter that is matched from a preset mapping relationship based on the road material type and used to correct the initial thermal diffusion area.

[0148] Specifically, the testing terminal can retrieve the construction archives of the highway to be tested, the preset road section material information database, or receive material data collected on-site to obtain the material type (i.e., road material type) used in the road surface; then, it calls the preset material type-thermal conductivity coefficient mapping relationship (structured association data) stored in the terminal, and uses the obtained material type as an index to quickly match the corresponding proportional coefficient or correction parameter to determine the material thermal conductivity coefficient calibration factor used to correct the initial heat diffusion area.

[0149] Step S603: The thermal diffusion area is calibrated based on the material thermal conductivity calibration factor to obtain the calibrated thermal diffusion area.

[0150] Specifically, the detection terminal can retrieve the obtained material thermal conductivity calibration factor (proportional coefficient or correction parameter) and the initial thermal diffusion area calculated in step S601. Through a preset calibration algorithm (such as coefficient multiplication scaling, difference correction, etc.), the calibration factor is applied to the initial thermal diffusion area to eliminate the influence of the difference in thermal conductivity of different road materials on the calculation of thermal diffusion area, correct the material-specific deviations not considered in the initial parameters, and obtain a calibrated thermal diffusion area that conforms to the actual thermal conductivity characteristics of the current road material.

[0151] Step S604: Calculate the temperature gradient variance of the crack region from the long-wave infrared band data of the effectively fused image, and determine the temperature gradient variance as a parameter for heat conduction uniformity.

[0152] Specifically, the detection terminal can extract complete data of the long-wave infrared band from the effectively fused image, focus on the crack area and divide it into multiple sampling sub-regions, calculate the temperature gradient (temperature difference between the crack and the surrounding matrix) in each sub-region, and then calculate the dispersion of the temperature gradient in all sampling sub-regions through a statistical variance algorithm to obtain the temperature gradient variance; finally, the temperature gradient variance is determined as the heat conduction uniformity parameter.

[0153] Step S605: Based on near-infrared band data, the measurement data of the area obstructed by foreign objects in the layered measurement of crack depth is removed to obtain effective crack width-depth sequence data. The crack width-depth sequence data is then subjected to piecewise smooth fitting to obtain the variation gradient. The variation gradient is used to characterize the crack morphology's expansion characteristics in the vertical direction.

[0154] Among them, the crack width-depth sequence data can be the width-depth correspondence data sequence obtained by the detection terminal based on the near-infrared band data of the effectively fused image, after performing depth layer measurement of the crack and removing invalid data such as foreign objects blocking the crack, and arranging the width-depth correspondence data sequence along the crack extension direction according to the depth level.

[0155] The gradient of change can be a quantitative parameter that characterizes the rate and trend of change of crack width with depth after piecewise smooth fitting of effective crack width-depth sequence data.

[0156] Specifically, the detection terminal can extract near-infrared band data from the effectively fused image, use the near-infrared penetration characteristics to perform depth layer measurement of the crack, identify and remove invalid data in areas obscured by foreign objects, and obtain a width-depth correspondence data sequence arranged in depth hierarchy along the crack extension direction (i.e., effective crack width-depth sequence data). Subsequently, the sequence data is processed by a preset piecewise smoothing fitting algorithm to calculate the quantitative parameter characterizing the rate and trend of crack width change with depth, i.e., the change gradient.

[0157] Step S606: Based on the calibrated heat diffusion area, heat conduction uniformity parameters and variation gradient, combined with the background data of the working conditions, the comprehensive risk value of the whole cycle and the remaining bearing capacity of the subgrade are obtained.

[0158] Specifically, the detection terminal can retrieve the calibrated thermal diffusion area, thermal conduction uniformity parameters and variation gradient, and load background data including historical detection data, environmental data and traffic data. Then, by quantitatively analyzing the correlation between crack thermal conduction characteristics, vertical propagation trend and full-cycle operating conditions, it obtains the full-cycle comprehensive risk value characterizing the crack's full-life-cycle damage and propagation risk, as well as the roadbed's remaining bearing capacity reflecting the current safe bearing capacity of the roadbed.

[0159] In this embodiment, the detection terminal extracts and calibrates the core geometric and thermodynamic parameters of the crack in stages, and combines them with the background data of the whole cycle to obtain the comprehensive risk value and the remaining bearing capacity of the subgrade, providing a scientific basis for crack risk assessment and maintenance decision-making.

[0160] In one embodiment, the method may further include the following steps:

[0161] Step S701: Based on the full-cycle comprehensive risk value and three-dimensional geometric parameters, and combined with the preset subgrade type, a recommended repair scheme is obtained by matching from the preset crack feature-repair scheme mapping database; the recommended repair scheme includes material selection, construction technology and estimated construction period parameters.

[0162] The preset roadbed type can be a roadbed category pre-stored in the testing terminal, based on highway roadbed structure design standards and engineering practices.

[0163] The preset crack feature-repair scheme mapping database can be a structured data set (such as a database, association table, etc.) that is pre-stored in the detection terminal and establishes the relationship between crack features, subgrade type and repair scheme.

[0164] The recommended repair solution can be a comprehensive repair execution plan that is matched by the detection terminal from the preset mapping database and customized for the current crack characteristics and roadbed type.

[0165] Specifically, the detection terminal can use the full-cycle comprehensive risk value, extracted three-dimensional geometric parameters, and the preset subgrade type of the road section to be detected as a joint index, call the pre-stored crack feature-subgrade type-repair scheme structured mapping database, and through retrieval and matching, select a comprehensive repair execution scheme customized for the current crack risk, morphology and subgrade type, that is, a recommended repair scheme.

[0166] Step S702: After the repair work is completed, a second inspection is carried out on the repaired area at the preset inspection cycle to obtain the comprehensive risk factor after repair.

[0167] Specifically, the detection terminal can perform a second detection after the repair work is completed, according to the pre-stored preset detection cycle. The terminal re-acquires the multimodal sensor dataset of the repair area, and through the same process as the initial detection, such as multi-dimensional fusion image generation, interference filtering and parameter extraction, it quantitatively analyzes the residual state of cracks, structural integrity and performance recovery of the repair area, and finally calculates the comprehensive risk coefficient after repair, which represents the safety risk level of the repaired area.

[0168] Step S703: Based on the comprehensive risk coefficient after repair and the comprehensive risk value throughout the entire cycle, the risk coefficient reduction rate is obtained.

[0169] The risk reduction rate can be an evaluation index used to quantify the effect of repair measures on crack risk suppression, obtained by calculating the ratio of the difference between the comprehensive risk coefficient after repair and the comprehensive risk value of the whole cycle before repair.

[0170] Specifically, the detection terminal can retrieve the comprehensive risk coefficient obtained from the secondary detection after the repair construction is completed and the comprehensive risk value of the whole cycle calculated before the repair for quantitative calculation, so as to obtain the risk coefficient reduction rate that characterizes the effect of the repair measures on crack risk suppression.

[0171] Step S704: If the risk reduction rate is lower than the preset standard threshold, a secondary repair plan adjustment instruction is generated and sent to the maintenance decision terminal; the secondary repair plan adjustment instruction is used to instruct the initiation of the repair plan re-matching process based on the post-repair detection data.

[0172] Specifically, the detection terminal can compare the calculated risk reduction rate with the pre-stored preset standard threshold. If it is determined that the reduction rate has not reached the preset standard threshold, it indicates that the original repair plan has not achieved the expected risk suppression effect. Then, it automatically generates a secondary repair plan adjustment instruction containing the repair effect evaluation result, specific data on the threshold failure, and the plan rematch requirements. This instruction is sent to the maintenance decision terminal in real time via the network, instructing it to start the repair plan rematch process based on the new data such as the residual state of the crack and structural performance obtained from the secondary detection after repair.

[0173] In this embodiment, the detection terminal matches a customized recommended repair plan, performs a second inspection to evaluate the repair effect, and determines whether to trigger a rematch of the plan based on the risk reduction rate, forming a closed-loop management system of detection-repair-evaluation-optimization to ensure the effectiveness of crack repair.

[0174] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0175] Based on the same inventive concept, this application also provides a crack detection system for highway construction as described above. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of a crack detection system for highway construction provided below can be found in the limitations of the crack detection method for highway construction described above, and will not be repeated here.

[0176] In one exemplary embodiment, such as Figure 3 As shown, a crack detection system 800 for highway construction is provided, comprising:

[0177] The data acquisition module 801 is used to acquire a multimodal sensing dataset of the surface of the highway to be inspected; the multimodal sensing dataset includes multispectral data, three-dimensional point cloud data and polarized light data;

[0178] The fusion processing module 802 is used to generate multi-dimensional fused images based on a multimodal sensing dataset;

[0179] The scene intelligent recognition module 803 is used to classify multi-dimensional fused images to obtain the current detected scene type;

[0180] The threshold decision module 804 is used to obtain interference judgment parameters based on the detection scene type and in combination with the preset scene-threshold mapping model.

[0181] The interference filtering module 805 is used to perform hierarchical interference filtering on the multi-dimensional fused image according to the interference judgment parameters, so as to obtain an effective fused image.

[0182] Analysis module 806 is used to extract the geometric and thermodynamic parameters of cracks from the effectively fused images, and combine them with the background data of the working conditions to obtain the comprehensive risk value of the whole cycle and the remaining bearing capacity of the subgrade; the background data of the working conditions includes historical detection data, environmental data and traffic data;

[0183] The report generation module 807 is used to generate a graded inspection report for the highway to be inspected based on the comprehensive risk value throughout the entire cycle and the remaining bearing capacity of the roadbed.

[0184] In one embodiment, the fusion processing module generates a multi-dimensional fused image based on a multimodal sensing dataset, including:

[0185] The modeling unit is used to segment the crack region point cloud data into independent point cloud clusters, and to reconstruct the independent point cloud clusters in three dimensions to generate a three-dimensional surface model of the crack.

[0186] The parameter extraction unit is used to extract the three-dimensional geometric parameters of the three-dimensional surface model.

[0187] The distribution map generation unit is used to obtain the polarization degree distribution map of the crack region based on polarized light data; the polarization degree distribution map is used to characterize the difference in polarization reflection between the crack and the road matrix;

[0188] The feature fusion unit is used to fuse multispectral data, three-dimensional geometric parameters, and polarization degree distribution maps to generate a multi-dimensional fused image.

[0189] In one embodiment, the feature fusion unit fuses multispectral data, three-dimensional geometric parameters, and polarization degree distribution maps to generate a multi-dimensional fused image, including:

[0190] The multispectral image reconstruction subunit is used to reconstruct multispectral data from images to generate multispectral images.

[0191] The spatial registration subunit is used to map and match the pixel coordinates of the crack's three-dimensional spatial coordinates and polarization degree distribution map corresponding to the three-dimensional geometric parameters with the pixels of the multispectral image using the RANSAC registration algorithm, thereby obtaining a multimodal spatial alignment feature set;

[0192] The weighted fusion subunit is used to perform pixel-level feature weighting and fusion on the multimodal spatially aligned feature set to generate a multidimensional fused image.

[0193] In one embodiment, the threshold decision module, based on the detection scene type and combined with a preset scene-threshold mapping model, obtains interference determination parameters, including:

[0194] Calculate the interference determination parameters using the following formula:

[0195]

[0196] in, These are interference detection parameters. It is the initial value of the correlation coefficient threshold obtained by calibration using historical data under a standard dry, sunny day scenario. It is the near-infrared dust concentration factor. It is the mean of the near-infrared dust concentration factor in the historical dataset. It is the standard deviation of the near-infrared dust concentration factor in the historical dataset. It is the long-wave infrared temperature gradient. This is a typical temperature gradient value under standard dry, sunny conditions. It is the influence factor of polarized light droplets. It is the scene-adaptive weighting coefficient.

[0197] In one embodiment, the scene intelligent recognition module performs scene classification on the multi-dimensional fused image to obtain the current detected scene type, including:

[0198] The environmental feature vector extraction unit is used to extract the near-infrared band reflection intensity, long-wave infrared temperature distribution, and polarized light scattering intensity of the multi-dimensional fused image as environmental feature vectors.

[0199] The scene type classification unit is used to input environmental feature vectors into a pre-trained convolutional neural network classification model to obtain the current detected scene type; the current detected scene type includes at least one of the following: dry sunny day, after rainstorm, dusty day, foggy day, and icy day.

[0200] In one embodiment, the analysis module extracts the geometric and thermodynamic parameters of the cracks from the effectively fused images and combines them with background data to obtain the comprehensive risk value and remaining bearing capacity of the subgrade over the entire life cycle, including:

[0201] The parameter extraction unit is used to extract the initial geometric morphology parameters and initial thermodynamic parameters of the crack based on the effectively fused image; the initial thermodynamic parameters include the heat diffusion area calculated based on the standard heat conduction model.

[0202] The factor determination unit is used to obtain the pavement material type of the highway to be tested and determine the corresponding material thermal conductivity calibration factor according to the preset material type-thermal conductivity coefficient mapping relationship.

[0203] The area calibration unit is used to calibrate the thermal diffusion area based on the material's thermal conductivity calibration factor to obtain the calibrated thermal diffusion area.

[0204] The parameter calculation unit is used to calculate the temperature gradient variance of the crack region from the long-wave infrared band data of the effectively fused image, and to determine the temperature gradient variance as the heat conduction uniformity parameter.

[0205] The gradient acquisition unit is used to remove measurement data from areas obscured by foreign objects in the layered measurement of crack depth based on near-infrared band data, thereby obtaining effective crack width-depth sequence data. The crack width-depth sequence data is then segmented and smoothly fitted to obtain the changing gradient. The changing gradient is used to characterize the crack morphology's expansion characteristics in the vertical direction.

[0206] The integrated risk and bearing capacity calculation unit is used to obtain the full-cycle integrated risk value and the remaining bearing capacity of the subgrade based on the calibrated heat diffusion area, heat conduction uniformity parameters and variation gradient, combined with the background data of the working conditions.

[0207] In one embodiment, the system further includes:

[0208] The scheme matching module is used to match recommended repair schemes from a preset crack feature-repair scheme mapping database based on the full-cycle comprehensive risk value and three-dimensional geometric parameters, combined with the preset subgrade type. The recommended repair schemes include material selection, construction technology and estimated construction period parameters.

[0209] The secondary inspection module is used to conduct secondary inspections on the repaired area at a preset inspection cycle after the repair work is completed, so as to obtain the comprehensive risk factor after the repair.

[0210] The effect evaluation module is used to obtain the risk reduction rate based on the comprehensive risk coefficient after repair and the comprehensive risk value throughout the entire cycle.

[0211] The instruction sending module is used to generate a secondary repair plan adjustment instruction and send it to the maintenance decision terminal if the risk reduction rate is lower than the preset standard threshold. The secondary repair plan adjustment instruction is used to instruct the initiation of the repair plan re-matching process based on the post-repair detection data.

[0212] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a crack detection method for highway construction as described above.

[0213] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0214] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0215] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for detecting cracks during highway construction, characterized in that, The method includes: A multimodal sensing dataset is acquired from the surface of the highway to be inspected; the multimodal sensing dataset includes multispectral data, three-dimensional point cloud data, and polarized light data. Based on the multimodal sensing dataset, a multidimensional fused image is generated; The multi-dimensional fused image is classified to obtain the current detected scene type; Based on the detection scenario type, and combined with the preset scenario-threshold mapping model, interference determination parameters are obtained; Based on the interference determination parameters, the multi-dimensional fused image is subjected to hierarchical interference filtering according to priority logic to obtain an effective fused image; The geometric and thermodynamic parameters of the cracks are extracted from the effective fused images, and combined with the background data of the working conditions, the comprehensive risk value and the remaining bearing capacity of the subgrade are obtained throughout the entire cycle; the background data of the working conditions includes historical detection data, environmental data and traffic data; Based on the comprehensive risk value throughout the entire life cycle and the remaining bearing capacity of the roadbed, a graded inspection report for the highway to be inspected is generated.

2. The method according to claim 1, characterized in that, The generation of a multi-dimensional fused image based on the multimodal sensing dataset includes: The three-dimensional point cloud data is segmented into crack region point cloud to obtain independent point cloud clusters, and the independent point cloud clusters are reconstructed in three dimensions to generate a three-dimensional surface model of the crack. Extract the three-dimensional geometric parameters of the three-dimensional surface model; Based on the polarized light data, a polarization degree distribution map of the crack region is obtained; the polarization degree distribution map is used to characterize the difference in polarization reflection between the crack and the road substrate; The multispectral data, the three-dimensional geometric parameters, and the polarization degree distribution map are fused to generate a multi-dimensional fused image.

3. The method according to claim 2, characterized in that, The step of fusing the multispectral data, the three-dimensional geometric parameters, and the polarization degree distribution map to generate a multidimensional fused image includes: The multispectral data is reconstructed to generate a multispectral image; The RANSAC registration algorithm is used to map and match the three-dimensional spatial coordinates of the cracks corresponding to the three-dimensional geometric parameters and the pixel coordinates of the polarization degree distribution map with the pixels of the multispectral image to obtain a multimodal spatial alignment feature set; The multimodal spatial alignment feature set is subjected to pixel-level feature weighted fusion to generate a multidimensional fused image.

4. The method according to claim 1, characterized in that, The interference determination parameters are obtained based on the detection scene type and a preset scene-threshold mapping model, including: The interference determination parameter is calculated using the following formula: in, These are interference detection parameters. It is the initial value of the correlation coefficient threshold obtained by calibration using historical data under a standard dry, sunny day scenario. It is the near-infrared dust concentration factor. It is the mean of the near-infrared dust concentration factor in the historical dataset. It is the standard deviation of the near-infrared dust concentration factor in the historical dataset. It is the long-wave infrared temperature gradient. This is a typical temperature gradient value under standard dry, sunny conditions. It is the influence factor of polarized light droplets. It is the scene-adaptive weighting coefficient.

5. The method according to any one of claims 1 to 4, characterized in that, The step of classifying the multi-dimensional fused image to obtain the current detected scene type includes: The near-infrared band reflection intensity, long-wave infrared temperature distribution, and polarized light scattering intensity of the multi-dimensional fused image are extracted as environmental feature vectors. The environmental feature vector is input into a pre-trained convolutional neural network classification model to obtain the current detection scene type; the current detection scene type includes at least one of dry sunny day, after rainstorm, dusty day, foggy day and icy day.

6. The method according to claim 1, characterized in that, The process of extracting the geometric and thermodynamic parameters of the cracks from the effectively fused image and combining them with background data to obtain the comprehensive risk value and remaining bearing capacity of the subgrade throughout the entire cycle includes: Based on the effectively fused image, the initial geometric morphology parameters and initial thermodynamic parameters of the crack are extracted; the initial thermodynamic parameters include the heat diffusion area calculated based on the standard heat conduction model. The road surface material type of the highway to be tested is obtained, and the corresponding material thermal conductivity calibration factor is determined according to the preset material type-thermal conductivity mapping relationship. The thermal diffusion area is calibrated based on the material's thermal conductivity calibration factor to obtain the calibrated thermal diffusion area. From the long-wave infrared band data of the effectively fused image, the temperature gradient variance of the crack region is calculated, and the temperature gradient variance is determined as a parameter for heat conduction uniformity. Based on the near-infrared band data, the measurement data of the area blocked by foreign objects in the crack depth layer measurement are removed to obtain effective crack width-depth sequence data. The crack width-depth sequence data is then subjected to piecewise smooth fitting to obtain the variation gradient. The variation gradient is used to characterize the crack morphology's expansion characteristics in the vertical direction. Based on the calibrated heat diffusion area, the heat conduction uniformity parameter, and the change gradient, combined with the working condition background data, the comprehensive risk value for the entire cycle and the remaining bearing capacity of the roadbed are obtained.

7. The method according to claim 1, characterized in that, The method further includes: Based on the comprehensive risk value throughout the entire life cycle and the three-dimensional geometric parameters, combined with the preset subgrade type, a recommended repair scheme is obtained by matching from the preset crack feature-repair scheme mapping database; the recommended repair scheme includes material selection, construction technology and estimated construction period parameters; After the repair work is completed, a second inspection is conducted on the repaired area at a predetermined inspection period to obtain the comprehensive risk factor after repair. Based on the comprehensive risk coefficient after repair and the comprehensive risk value throughout the entire cycle, the risk coefficient reduction rate is obtained; If the risk reduction rate is lower than the preset standard threshold, a secondary repair plan adjustment instruction is generated and sent to the maintenance decision terminal; the secondary repair plan adjustment instruction is used to instruct the initiation of a repair plan re-matching process based on post-repair detection data.

8. A crack detection system for highway construction, characterized in that, The system includes: The data acquisition module is used to acquire a multimodal sensing dataset of the surface of the highway to be inspected; the multimodal sensing dataset includes multispectral data, three-dimensional point cloud data, and polarized light data. The fusion processing module is used to generate a multi-dimensional fused image based on the multimodal sensing dataset; The scene intelligent recognition module is used to classify the multi-dimensional fused image into scenes to obtain the current detected scene type; The threshold decision module is used to obtain interference judgment parameters based on the detection scene type and in combination with a preset scene-threshold mapping model. The interference filtering module is used to perform hierarchical interference filtering on the multi-dimensional fused image according to the interference determination parameters, so as to obtain an effective fused image. The analysis module is used to extract the geometric and thermodynamic parameters of the cracks from the effective fused images, and combine them with the working condition background data to obtain the comprehensive risk value and the remaining bearing capacity of the subgrade throughout the entire cycle; the working condition background data includes historical detection data, environmental data and traffic data; The report generation module is used to generate a graded inspection report for the highway to be inspected based on the comprehensive risk value throughout the entire cycle and the remaining bearing capacity of the roadbed.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images