Method for identifying internal defects of pavement structure

Abstract

This invention discloses a method for identifying internal defects in pavement structures, comprising: constructing a defect identification dataset and obtaining a set of pavement defect types and their formation mechanisms based on the dataset; acquiring pavement defect maps, identifying defect locations in the maps, and determining the defect type at each location based on the set of pavement defect types; identifying water-related defects in the maps using deep learning, and identifying internal cavities in the pavement using ground-penetrating radar. This invention, by studying the mechanisms of asphalt pavement surface defect identification based on computer vision and pavement internal defect identification based on ground-penetrating radar, proposes a multi-technical method for perceiving and analyzing pavement structure defects, thereby improving the accuracy of pavement defect perception.

Description

Technical Field

[0001] This invention belongs to the field of road inspection, and in particular relates to a method for identifying internal defects in road surface structures. Background Technology

[0002] In recent years, highway construction has continued to grow. Due to the unique climate, hydrogeological and geomorphological conditions in the region, highways built in the early stages have developed increasingly serious early-stage defects. Among these defects, uneven subgrade settlement and low subgrade strength have led to a large number of pavement cracks. The continuous development of cracks seriously affects the integrity of the pavement and reduces its service life. Furthermore, it is difficult to accurately detect and measure internal structural defects of the pavement, making timely repair and treatment difficult. As a result, voids generated in the pavement structure under long-term external loads are difficult to detect and treat in time, leading to water leakage at these voids and various water-related defects. This seriously endangers the structural safety and service life of the pavement, greatly increases the difficulty of pavement management and maintenance, and the high maintenance costs have become a significant burden on traffic management departments, bringing considerable negative impacts to managers and builders.

[0003] Composite asphalt pavement is a typical form of highway tunnel pavement. Fatigue cracking and reflective cracking are the most typical and widespread defects in tunnel asphalt pavement. These cracks significantly reduce pavement integrity, have a substantial impact on pavement load-bearing capacity, and seriously threaten driving safety. Currently, traditional detection or sensing technologies for asphalt pavement cracks mainly include manual inspection, rapid detection vehicles, and various pre-embedded sensors. Most areas still rely on manual inspection, but manual inspection is inefficient and prone to errors. With the rapid development of image processing technology and laser scanning technology, rapid detection systems for pavements have developed rapidly. The application of rapid detection systems has significantly improved the speed of single-pass detection; however, existing technologies have low recognition accuracy and cannot meet the requirements for refined pavement management and maintenance. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying internal defects in road structures, so as to solve the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides a method for identifying internal defects in pavement structures, comprising the following steps:

[0006] Construct a disease identification dataset and obtain the disease type set and formation mechanism of road surface diseases based on the disease identification dataset;

[0007] Collect road surface distress maps, identify the locations of distresses in the road surface distress maps, and determine the distress types at the distress locations based on the obtained distress type set;

[0008] Deep learning is used to identify water-related defects in the road surface distress map, and ground-penetrating radar is used to identify internal cavities in the road surface.

[0009] Optionally, the process of constructing a disease identification dataset and obtaining a set of road surface disease types and their formation mechanisms based on the disease identification dataset includes:

[0010] A damage identification dataset is constructed based on the causes of road surface defects, the deterioration effect of road structure, and the degree of impact on driving safety.

[0011] Based on the aforementioned disease identification dataset and a three-dimensional convolutional neural network, a disease identification model is constructed to simulate the generation process of diseases, analyze the changes in pavement stress characteristics and driving safety under the action of diseases, as well as the causes, state types and distribution range of pavement damage, and simulate the process of internal pavement damage defects developing into diseases.

[0012] Optionally, the process of identifying the location of defects in the pavement defect map includes:

[0013] The road surface defect images are subjected to digital image enhancement, grayscale processing, and binarization preprocessing.

[0014] A YOLOv5 recognition model is constructed based on the YOLOv5 network. The preprocessed pavement distress map is then identified based on the YOLOv5 recognition model to obtain the distress location.

[0015] Optionally, the process of identifying water-related defects in the pavement distress map based on deep learning includes:

[0016] The response characteristics of electromagnetic waves in water-damaged areas were obtained based on the finite-difference time-domain method.

[0017] Feature extraction is performed on the road surface defect map to obtain a water-related defect dataset;

[0018] Water-related diseases are identified based on a two-dimensional convolutional neural network and the aforementioned water-related disease dataset.

[0019] Optionally, the process of identifying internal cavities in the road surface based on ground-penetrating radar includes:

[0020] A model of an air-filled and water-filled cavity at different structural layer locations was constructed based on the finite-time difference method.

[0021] Several hierarchical structure images of the road surface were obtained based on ground-penetrating radar and the inflatable and water-filled cavity model.

[0022] Offset imaging processing is performed on several of the hierarchical images to obtain a radar wave simulation image;

[0023] Based on the radar wave image, a forward modeling simulation of the road cavity is performed to obtain the radar wave feature image;

[0024] Based on the radar wave feature image, internal voids are identified using the YOLOv5 recognition model.

[0025] Optionally, the YOLOv5 network includes a Backbone part, a Neck part, and a Head part; the process of constructing the YOLOv5 recognition model includes:

[0026] A dataset is constructed based on the road surface defect map, and the dataset is divided into a test set, a training set, and a validation set.

[0027] The road surface distress images are input into the Backbone part of the YOLOv5 network to obtain distress feature maps at different scales;

[0028] The disease feature maps at different scales are input into the Neck part, and the disease feature maps are sampled and feature fused to obtain tensor data at different scales.

[0029] The tensor data is input into the Head part for gradient calculation, and the result is verified based on the validation set to obtain the YOLOv5 recognition model.

[0030] Optionally, the process of identifying water-related diseases based on a two-dimensional convolutional neural network and the water-related disease dataset includes:

[0031] A classifier integrating several two-dimensional convolutional neural networks is constructed to perform deep learning on the water-related disease dataset;

[0032] The weights of each of the two-dimensional convolutional neural network classifiers are controlled based on a voting scoring method, and the water disease identification results are output.

[0033] Optionally, the process of performing offset imaging processing on several of the hierarchical images includes:

[0034] The wave velocity range of electromagnetic waves in several hierarchical images is obtained based on commonly used dielectric constants.

[0035] Set the offset parameter based on the wave velocity range;

[0036] The feature vectors of the hierarchical images are extracted based on the wavelet transform method, and the wavelet entropy of each hierarchical image is calculated.

[0037] The wave speed corresponding to the minimum wavelet entropy in the wave speed range is taken as the optimal wave speed.

[0038] Based on the optimal wave velocity and the offset parameters, offset imaging processing is performed on several hierarchical images.

[0039] The technical effects of this invention are as follows:

[0040] This invention starts with the image features of cracks and water seepage, and establishes image extraction and area measurement techniques based on the YOLO machine learning algorithm, as well as analytical identification algorithms for road surface defects. Furthermore, it improves identification efficiency and accuracy based on machine learning algorithms. Addressing the challenge of detecting internal defects such as voids and water damage in road surfaces, this invention starts with the radar spectral characteristics of the road surface's internal structure. Based on the differences in peak intensity and electromagnetic wave response characteristics of radar spectra for different materials and structures, and the optimal wavelet entropy processing method, it establishes ground-penetrating radar signal processing and analysis methods for different internal road surface defects. This provides a theoretical basis for ground-penetrating radar-based identification models for internal road surface voids and water damage. Attached Figure Description

[0041] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0042] In the attached diagram:

[0043] Figure 1 This is a flowchart of the method for identifying internal defects in road structures in an embodiment of the present invention;

[0044] Figure 2 This is a diagram of the YOLOv5 network structure in an embodiment of the present invention. Detailed Implementation

[0045] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0046] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0047] Example 1

[0048] like Figure 1-2 As shown, this embodiment provides a method for identifying internal defects in pavement structures, specifically including the following steps:

[0049] Analysis of Pavement Defects (Cracks, Water Leakage, Cavities, etc.) and Their Formation Mechanisms: This study investigates and statistically analyzes the causes, types, and scale of cracks and water leakage defects in municipal roads, their deterioration effects on pavement structure, and their impact on driving safety. A crack and water leakage identification dataset is established for use in machine learning algorithms. Based on this dataset, a three-dimensional convolutional neural network is used to simulate the formation process of cracks and water leakage defects in municipal highways. The study analyzes the changing patterns of pavement stress characteristics and driving safety under various defects, providing validation for computer vision-based pavement damage perception and identification methods. The study also investigates and analyzes the causes, states, and distribution range of internal defects and damage such as pavement voids and water damage, their impact on driving safety and road service life, and simulates the development process of internal damage defects further developing into severe cracks, pavement subsidence, potholes, and other defects.

[0050] Image recognition and analysis of road surface cracks and water seepage: High-resolution images of road surface cracks and water seepage are acquired. The acquired digital images undergo preprocessing such as digital image enhancement, grayscale processing, and binarization. Binary segmentation, by continuously adjusting the segmentation threshold, ensures that the segmented image only displays the affected areas, thereby extracting the image feature values ​​needed for deep learning algorithms.

[0051] The YOLOv5 deep learning algorithm is used to perform machine learning on the preprocessed images: The YOLOv5 network in this embodiment includes a backbone part, a neck part, and a head part;

[0052] The specific process of building the recognition model and deep learning is as follows:

[0053] A dataset is constructed using collected road surface defect images, and this dataset is divided into a test set, a training set, and a validation set. The road surface defect images are input into the Backbone part of the YOLOv5 network to obtain defect feature maps at different scales. The defect feature maps at different scales are then input into the Neck part, where sampling and feature fusion are performed to obtain tensor data at different scales. The tensor data are then input into the Head part for gradient calculation, and validation is performed based on the validation set to obtain the YOLOv5 recognition model.

[0054] YOLOv5 utilizes the concept of regression, which has the advantages of fast recognition speed and easier generalization of learning targets, making road damage recognition faster and more accurate, ultimately achieving the analysis of road defects.

[0055] Ground-penetrating radar identification and analysis of internal voids and water-related defects in pavement structures: Through FDTD (Finite-Domain Differential) forward modeling and indoor inversion experiments, this study analyzes the electromagnetic wave energy attenuation under the coupled state of air and asphalt layers under different antenna suspension heights, pavement structural parameters (porosity, thickness, etc.), and pavement moisture content. The response of electromagnetic waves in water-related defect areas is investigated, and an effective identification method and theory for asphalt pavement water-related defects are established. Characteristic parameters representing water-related defects (crack severity and area, pothole size and depth) are extracted to establish an asphalt pavement water-related defect dataset. A recognition model based on a two-dimensional convolutional neural network is established and optimized, ultimately forming an intelligent evaluation method for asphalt pavement water-related defect areas.

[0056] The process of obtaining the identification model includes: constructing a classifier that integrates several two-dimensional convolutional neural networks, performing deep learning on the water disease dataset; using a voting scoring method to control the weights of each of the two-dimensional convolutional neural network classifiers, and outputting the water disease identification results.

[0057] The process for identifying cavities inside the road surface is as follows:

[0058] Based on FDTD, models of air-filled and water-filled cavities at different structural layer locations were established.

[0059] Several hierarchical images of the road surface were obtained based on ground-penetrating radar and an inflatable and water-filled cavity model. The obtained hierarchical images were processed by offset imaging to obtain radar wave simulation images. The road cavities were simulated using GprMax and Matlab programming to obtain radar wave feature images. Based on the radar wave feature images, the internal cavities were identified using the YOLOv5 identification model constructed in the previous steps.

[0060] The migration imaging steps include: obtaining the wave velocity range of electromagnetic waves in several hierarchical images based on common dielectric constants; setting migration parameters according to different wave velocities in the obtained wave velocity ranges; extracting feature vectors of the hierarchical images based on wavelet transform and calculating the wavelet entropy of each hierarchical image; taking the wave velocity in the wave velocity range corresponding to the minimum value of the wavelet entropy as the optimal wave velocity; and performing migration imaging on several hierarchical images according to the optimal wave velocity and the migration parameters.

[0061] This embodiment analyzes the radar wave image features of the forward modeling results of cavitation and proposes a ground-penetrating radar (GPR) method for identifying abnormal features of road structural layer defects. A radar image database for cavitation defects is constructed, image features are identified, and YOLOv5 technology is used for automatic waveform identification of interlayer voids, thus forming a cavity defect identification and diagnosis technology.

[0062] This invention proposes a multi-technical method for perceiving and analyzing pavement structural defects by studying the identification mechanisms of asphalt pavement surface defects (cracks, water seepage) based on computer vision and pavement internal defects (cavities, water defects) based on ground penetrating radar, thereby improving the accuracy of pavement defect perception.

[0063] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying internal defects in pavement structures, characterized in that, Includes the following steps: Construct a disease identification dataset and obtain the disease type set and formation mechanism of road surface diseases based on the disease identification dataset; Collect road surface distress maps, identify the locations of distresses in the road surface distress maps, and determine the distress types at the distress locations based on the obtained distress type set; Deep learning is used to identify water-related defects in the road surface distress map, and ground-penetrating radar is used to identify internal cavities in the road surface. The process of identifying water-related defects in the pavement distress map based on deep learning includes: The response characteristics of electromagnetic waves in water-damaged areas were obtained based on the finite-difference time-domain method. Feature extraction is performed on the road surface defect map to obtain a water-related defect dataset; Water-related diseases are identified based on a two-dimensional convolutional neural network and the aforementioned water-related disease dataset. The process of identifying internal cavities in road surfaces using ground-penetrating radar includes: A model of an air-filled and water-filled cavity at different structural layer locations was constructed based on the finite-time difference method. Several hierarchical structure images of the road surface were obtained based on ground-penetrating radar and the inflatable and water-filled cavity model. Offset imaging processing is performed on several of the hierarchical images to obtain a radar wave simulation image; Based on the radar wave simulation image, forward modeling is performed on the road cavity to obtain radar wave feature images; Based on the radar wave feature image, internal voids are identified using the YOLOv5 recognition model.

2. The method for identifying internal defects in pavement structures according to claim 1, characterized in that, The process of constructing a pavement disease identification dataset and obtaining a set of pavement disease types and their formation mechanisms based on the dataset includes: A damage identification dataset is constructed based on the causes of road surface defects, the deterioration effect of road structure, and the degree of impact on driving safety. Based on the aforementioned disease identification dataset and a three-dimensional convolutional neural network, a disease identification model is constructed to simulate the generation process of diseases, analyze the changes in pavement stress characteristics and driving safety under the action of diseases, as well as the causes, state types and distribution range of pavement damage, and simulate the process of internal pavement damage defects developing into diseases.

3. The method for identifying internal defects in pavement structures according to claim 1, characterized in that, The process of identifying the location of road surface defects in the aforementioned road surface defect map includes: The road surface defect images are subjected to digital image enhancement, grayscale processing, and binarization preprocessing. A YOLOv5 recognition model is constructed based on the YOLOv5 network. The preprocessed pavement distress map is then identified based on the YOLOv5 recognition model to obtain the distress location.

4. The method for identifying internal defects in pavement structures according to claim 3, characterized in that, The YOLOv5 network includes a Backbone part, a Neck part, and a Head part; the process of constructing the YOLOv5 recognition model includes: A dataset is constructed based on the road surface defect map, and the dataset is divided into a test set, a training set, and a validation set. The road surface distress images are input into the Backbone part of the YOLOv5 network to obtain distress feature maps at different scales; The disease feature maps at different scales are input into the Neck part, and the disease feature maps are sampled and feature fused to obtain tensor data at different scales. The tensor data is input into the Head part for gradient calculation, and the result is verified based on the validation set to obtain the YOLOv5 recognition model.

5. The method for identifying internal defects in pavement structures according to claim 1, characterized in that, The process of identifying water-related diseases based on a two-dimensional convolutional neural network and the aforementioned water-related disease dataset includes: A classifier integrating several two-dimensional convolutional neural networks is constructed to perform deep learning on the water-related disease dataset; The weights of each of the two-dimensional convolutional neural network classifiers are controlled based on a voting scoring method, and the water disease identification results are output.

6. The method for identifying internal defects in pavement structures according to claim 1, characterized in that, The process of performing offset imaging processing on several of the aforementioned hierarchical images includes: The wave velocity range of electromagnetic waves in several hierarchical images is obtained based on commonly used dielectric constants. Set the offset parameter based on the wave velocity range; The feature vectors of the hierarchical images are extracted based on the wavelet transform method, and the wavelet entropy of each hierarchical image is calculated. The wave velocity corresponding to the minimum wavelet entropy in the wave velocity range is taken as the optimal wave velocity. Based on the optimal wave velocity and the offset parameters, offset imaging processing is performed on several hierarchical images.

Images