A highway roadbed pavement health state monitoring and evaluation system based on machine vision

Abstract

This invention discloses a machine vision-based highway subgrade and pavement health status monitoring and assessment system. Addressing the problems of traditional highway maintenance relying on manual inspections, which is inefficient, subjective in hazard identification, and unable to promptly detect hidden hazards or conduct risk assessments, this system comprises a vehicle-mounted or drone-based visual acquisition unit, an edge computing processing module, and a cloud-based health assessment platform. The system acquires highway subgrade and pavement image data using high-definition vision equipment, employs the U-Net semantic segmentation algorithm to automatically identify visible hazards such as cracks, potholes, grooves, and subsidence, and calculates hidden hazards such as subgrade settlement and slippage using visual texture features. Based on highway maintenance standards, a multi-dimensional health assessment system is constructed to achieve hazard classification and quantification, spatial positioning, and automatic assessment of the overall road health index. Simultaneously, the detection results are visualized, stored, and maintenance recommendations are generated.

Description

Technical Field

[0001] This invention relates to a machine vision-based system for monitoring and evaluating the health status of highway subgrade and pavement, specifically a machine vision-based system for monitoring and evaluating the health status of highway subgrade and pavement. Background Technology

[0002] As core components of transportation infrastructure, highway subgrades and pavements are prone to developing visible and hidden hazards such as cracks, potholes, ruts, subsidence, and slippage under the long-term effects of vehicle loads, high temperatures, environmental erosion, and geological changes. If these hazards are not detected and addressed in a timely manner, they will continue to exacerbate structural damage, seriously affecting vehicle travel and increasing subsequent maintenance costs.

[0003] Currently, highway maintenance in China still relies mainly on manual on-site inspections, which has significant drawbacks. Manual inspections are inefficient, have limited coverage, and often involve road closures, severely impacting vehicle movement and traffic flow. Furthermore, they cannot effectively detect hidden hazards. Traditional solutions for dealing with hidden hazards rely on conventional sensors, which are costly to deploy, complex to install, and can only acquire data from single points, failing to detect the distribution of hidden hazards over a wide area.

[0004] Moreover, most existing detection technologies are for identifying surface hazards of roadbeds and pavements, lacking the ability to invert hidden hazards of roadbeds and pavements. They also have weak anti-interference ability for pattern recognition and are easily affected by natural environmental factors such as sunlight, shadows, and stains, resulting in missed or false detections. Furthermore, there is no standardized multi-dimensional health assessment and management system, making it impossible to make scientific maintenance decisions based on the detected hidden hazards.

[0005] Existing detection technologies have a low level of intelligence and are unable to make scientific decisions regarding various visible and hidden hazards. Therefore, it is necessary to develop a system that can take into account both pavement hazard identification and subgrade condition assessment, has a high degree of automation, provides reliable assessment results, and can provide scientific maintenance decisions. Summary of the Invention

[0006] The purpose of this invention is to provide a machine vision-based highway subgrade and pavement health status monitoring and evaluation system to solve the problems of low efficiency of manual inspection, insufficient accuracy in identifying visible and hidden hazards, and inability to provide scientific maintenance for the detection and evaluation of hidden hazards in the existing technology.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a highway subgrade and pavement health status monitoring and evaluation system based on machine vision, including an image acquisition module, an edge preprocessing module, a hidden danger intelligent identification module, a subgrade and pavement status inversion and storage module, a comprehensive health assessment module, and a cloud management and scientific decision-making module.

[0008] The image acquisition module uses a vehicle-mounted high-definition camera and a drone aerial photography device to collect images of the roadbed condition and surrounding environment along the road surface, and simultaneously record GPS location information to form geographic coordinates for effective real-time observation and effective location data collection.

[0009] The edge preprocessing module is deployed on the acquisition terminal to complete image dehazing, desmearing, shadow correction, noise filtering, and vibration elimination, thereby minimizing the impact of environmental factors and reducing the impact of environmental influence on the accuracy of subsequent hidden danger identification. The preprocessing results are then transmitted to the backend platform for storage.

[0010] The aforementioned hazard intelligent identification module is based on the U-Net semantic segmentation algorithm, which automatically segments and identifies safety hazards such as road surface cracks, grooves, ruts, bulges, and subsidence. It also classifies the types of hazards caused by these safety hazards and calculates the area, length, and severity of each hazard type.

[0011] The aforementioned roadbed and pavement state inversion and storage module uses the image acquisition module to collect the texture deformation and settlement bulge contour features of the roadbed and pavement, and combines them with the roadbed and pavement mechanical experimental model to repeatedly deduce hidden dangers such as roadbed and pavement settlement, slippage, bulges, and voids. This enables non-contact state detection of the interior of the roadbed and pavement, and stores the detected hidden dangers and the deduced results.

[0012] The health assessment module described above, based on current highway maintenance technical specifications, constructs a multi-dimensional evaluation index that includes the damage rate of the pavement and subgrade, smoothness, stability coefficient of the subgrade, and attenuation coefficient of the structure. Through multi-level analysis of the weight of each index, it calculates the comprehensive health index of the pavement and subgrade of the entire road section and completes the stage-by-stage health classification treatment of the pavement and subgrade.

[0013] The cloud management and scientific decision-making module is used to evaluate, manage, and scientifically calculate the data information collected by the image acquisition module, the edge preprocessing module, and the algorithm of the hazard intelligent identification module. It produces a visual hazard distribution layer, classifies health levels, calculates priority recommendations, and realizes integrated management of monitoring, evaluation, and decision-making.

[0014] Furthermore, the aforementioned intelligent hazard identification module introduces an attention mechanism to optimize feature extraction, enhancing the ability to detect hazards on a smaller scale under complex natural environmental factors such as sunlight, shadows, rain, snow, and storms, effectively reducing the rate of missed and false detections.

[0015] Furthermore, the roadbed and pavement state inversion and storage module is based on the mapping relationship between visual deformation features and roadbed and pavement mechanical parameters, eliminating the need for additional traditional road detection sensors and significantly reducing system construction and operation costs.

[0016] Furthermore, the image acquisition module of this system can adaptively adjust the image acquisition frequency and recognition thresholds according to the roadbed and pavement hazards, based on two different acquisition modes: vehicle-mounted inspection and drone inspection, thus adapting to various scenarios such as daily inspections and special inspections.

[0017] Compared with the prior art, the present invention has the following beneficial effects: Using machine vision for non-contact inspection replaces traditional manual inspection, significantly improving inspection efficiency and reducing labor costs, thereby enabling routine inspection of the entire roadbed and pavement.

[0018] It takes into account both visible and hidden hazards in the roadbed and pavement, and repeatedly simulates the hazards to make the detection results more comprehensive. This solves the shortcomings of traditional detection methods that can only detect visible hazards, thereby achieving comprehensive maintenance and upkeep of the roadbed and pavement.

[0019] Based on cloud-based management and scientific decision-making modules, a standardized, multi-dimensional health assessment system is constructed. This system quantifies hazard identification results into a comprehensive health index, directly supporting maintenance decisions. The increased intelligence and efficiency reduce costs associated with different types of roadbed and pavement hazards, significantly enhancing the efficiency of scientific maintenance.

[0020] Based on the roadbed and pavement condition inversion and storage module, the model incorporates an attention mechanism, which enhances its resistance to environmental interference, improves the accuracy of hazard identification, and makes the overall system more reliable, thus making it impossible for roadbed and pavement hazards to go undetected.

[0021] This system can be widely used in the daily maintenance of highways, national roads, provincial trunk roads, and normal urban roads, as well as in scenarios such as pavement and roadbed hazard surveys and safety early warnings, significantly improving the level of intelligent operation and maintenance of various highway infrastructures. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0023] Figure 1 This is a system overall structure framework diagram provided for an embodiment of the present invention; Figure 2 This is a schematic diagram of the algorithm flow; Figure 3 This is a schematic diagram illustrating the data acquisition and edge preprocessing process. Figure 4 A schematic diagram illustrating the execution process for health assessment and visualized decision-making; Detailed Implementation

[0024] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0025] An embodiment of the present invention provides a highway subgrade and pavement health status monitoring and assessment system based on machine vision. The overall architecture consists of an image acquisition module, an edge preprocessing module, a hidden danger intelligent identification module, a subgrade and pavement status inversion module, a comprehensive health assessment module, and a cloud management and decision-making module. The modules work together to complete the output system of automated detection, identification, iterative simulation, assessment, and maintenance decision-making for the entire highway section.

[0026] Please refer to Figure 1 This invention provides a machine vision-based system for monitoring and evaluating the health status of highway subgrade and pavement, comprising the following steps: 1. Implementation of the image acquisition module: The inspection vehicle is mounted on a vehicle-mounted system, with four high-definition industrial cameras installed on the roof to cover the entire lane. The cameras have a resolution of 4K and a sampling frame rate of 25fps. It is also equipped with a GPS positioning module and an inertial measurement unit module to ensure that each frame of the image can be accurately bound to the map's latitude and longitude, thus ensuring the accuracy of the location. The inspection vehicle travels at a constant speed of 30-60 km / h, continuously collecting images of the road surface, roadbed, and shoulders, forming a time-series image with map latitude and longitude location labels.

[0027] 2. Implementation of the edge preprocessing module: The acquisition terminal has a built-in edge computing unit that preprocesses the images acquired by the image acquisition module in real time. First, it performs dehazing by using a dark channel verification algorithm, then performs shadow correction using adaptive filtering, and finally removes sound interference caused by road stains, debris, etc. by using Gaussian filtering. Finally, it extracts the region of interest of the road surface and roadbed, and removes invalid scenes such as sky, green belt, and indicator lights. The preprocessed results are transmitted to the cloud server in real time via the 5G network.

[0028] The dark channel dehazing method involved is as follows: (1) Calculate the dark channel of the input image: For the input hazy image, calculate its dark channel. The definition of the dark channel is: For each pixel in the image, calculate its minimum value in all color channels, and then take the minimum value in the local area of ​​the pixel to obtain the dark channel value of the pixel.

[0029] (2) Estimating atmospheric light value: Atmospheric light value refers to the light intensity value of the brightest pixel in the scene, which represents the fog concentration. It can be obtained by finding the brightest point in the dark channel, and then finding the corresponding pixel in the original image and using it as the atmospheric light value.

[0030] (3) Calculate the transmissivity map: The transmissivity map represents the proportion of light scattered by fog as it travels from the scene to the camera. The transmissivity map can be calculated using the dark channel and atmospheric light values.

[0031] (4) Restore the fog-free image: Use the transmission rate map and atmospheric light value to restore the foggy image and obtain a clear fog-free image.

[0032] The Gaussian filter involved is a linear smoothing filter technique widely used in image processing and computer vision. Its core function is to remove noise and smooth images, and it is often used in image preprocessing, background blurring effects, and noise reduction steps before edge detection.

[0033] 3. Implementation of the intelligent hazard identification module The cloud-based deployment, based on an improved attention mechanism and the U-Net semantic segmentation algorithm, inputs a pre-processed model. The system quickly identifies various hazard types, such as cracks, grooves, ruts, subsidence, and bulges, which are then submitted to U-Net for pixel segmentation. This allows for precise calculation of quantitative indicators such as the area and width of the hazard. According to the "Highway Technical Condition Assessment Standard," the system automatically classifies hazards into three levels: minor, moderate, and severe. It also performs precise GPS positioning on the coordinate images with map latitude and longitude acquired by the image acquisition module. The U-Net semantic segmentation consists of a left-side encoder (contraction path), a right-side decoder (expansion path), and skip connections connecting the two. The left-side path progressively reduces the spatial resolution of the image while increasing the number of feature channels to extract high-level semantic features. The right-side path progressively restores the low-resolution feature map to the original image size, achieving pixel-level precise positioning. However, the decoder may lose edge and texture details during size restoration. Skip connections transfer the high-resolution feature map of the corresponding layer from the encoder.

[0034] 4. Implementation of the Subgrade and Pavement Condition Inversion Module Based on the results of U-Net semantic segmentation, longitudinal and transverse settlement curves and uneven deformation contour maps of the roadbed and pavement, as well as local collapse texture features, are extracted. Then, the visual features are input into a pre-trained roadbed and pavement mechanical mapping model to infer the amount of settlement, slippage trend, probability of concrete missing voids and other hidden hazards of the roadbed and pavement. This enables the indirect assessment of the possibility of hidden hazards without contact or sensors, as well as the monitoring of the internal state of the roadbed and pavement.

[0035] 5. Implementation of the Comprehensive Health Assessment Module Based on my country's current highway maintenance standards, a multi-dimensional evaluation system is constructed by selecting factors such as roadbed and pavement damage rate, rut depth, crack width, size of depressions and bulges, settlement, and stability coefficient of roadbed and pavement. The comprehensive health index of road sections is calculated using a weighted summation method, and highways are divided into five health levels: excellent, good, medium, poor, and suboptimal. The health level classification of the entire road section is completed, and regional coordinates are divided for each level of road section to ensure that road maintenance can be carried out as soon as possible.

[0036] 6. Implementation of cloud-based management and decision-making modules The system stores all hazard data, location information, and health assessment results into a database. It generates a color-coded thermal distribution electronic map based on the corresponding health level classification and prioritizes scientific decision-making suggestions for poor and secondary road sections, forming a closed-loop management system from data collection to maintenance decision-making.

[0037] Please refer to Figure 2This invention provides a machine vision-based system for monitoring and evaluating the health status of highway subgrade and pavement, comprising the following steps: Step 1: Image Input This embodiment uses an industrial area array camera mounted on a detection vehicle to continuously acquire high-resolution visible light images of the road surface while traveling at a constant speed along the highway. At the same time, it combines a GPS module to synchronously acquire location and mileage information, forming a raw road surface image data stream with geographic coordinates. This completes the image input and provides raw visual data for subsequent identification of roadbed and pavement hazards.

[0038] Step 2: Pre-processing (defogging, calibration) Preprocessing is performed on the input raw image. First, the dark channel prior dehazing algorithm is used to eliminate the fogging and blurring caused by fog and dust. Then, lens distortion correction and perspective transformation correction are completed through camera calibration parameters. At the same time, Gaussian filtering is used to remove environmental noise. Finally, a standardized road image with clear texture and accurate geometry is output to eliminate the impact of external environmental interference on the subsequent segmentation accuracy.

[0039] Step 3: U-Net Segmentation Algorithm The preprocessed image is input into the trained improved U-Net semantic segmentation model, which is trained on a large number of road surface samples labeled with cracks, potholes, ruts, subsidence, and slippage marks. The model automatically performs pixel-level segmentation, accurately identifies and outputs masks, category labels and contour boundaries of various road surface hazard areas, and realizes automated segmentation and extraction of road surface hazards.

[0040] Step 4: Quantifying Hidden Dangers in Road Surfaces and Subgrades Based on the U-Net segmentation results, the conversion relationship between pixels and physical size is established by combining camera calibration parameters. The equivalent values ​​of length, width, area, and depth of various hazards, as well as the distribution density of hazards, are automatically quantified and calculated to complete the quantitative statistics of road surface damage hazards and form a multi-dimensional hazard quantitative dataset.

[0041] Step 5: Visual Deformation Feature Extraction Based on the quantification results of potential hazards, subpixel-level contour extraction is performed on the segmented areas to calculate local curvature changes, edge gradient deformation, regional tilt features and relative displacement features of the road surface. The apparent deformation features of the road surface caused by the instability of the roadbed are explored, and a visual deformation feature vector is established to characterize the potential instability trend of the roadbed.

[0042] Step Six: Subgrade and Pavement Mechanical Mapping Based on the theory of highway subgrade structural mechanics, a mapping model of pavement visual deformation characteristics and subgrade mechanical parameters is constructed. The extracted visual deformation characteristics are input into the mapping model to solve for key mechanical parameters such as subgrade elastic modulus, soil shear strength, and base stress distribution, and to establish the correlation between the pavement appearance state and the subgrade internal mechanical state.

[0043] Step 7: Subgrade Settlement / Slippage Inversion Based on the above mechanical mapping results, the finite element inversion algorithm is used in conjunction with the deformation change law of time series images to invert and calculate the vertical settlement, horizontal slip displacement, settlement rate and slip trend of the roadbed, and complete the quantitative inversion of the internal health state of the roadbed, realizing the intelligent inference from the visual image of the road surface to the internal instability state of the roadbed.

[0044] Through the top-down unidirectional data flow algorithm process in this embodiment, the entire process from machine vision image acquisition to road surface hazard segmentation, quantification, deformation feature extraction, and then to subgrade mechanical mapping and settlement and slip inversion is fully automated. It can accurately assess the overall health status of the subgrade and pavement, and provide a reliable quantitative basis for highway maintenance decisions.

[0045] Please refer to Figure 3 This invention provides a machine vision-based system for monitoring and evaluating the health status of highway subgrade and pavement, comprising the following steps: Step 1: Multi-dimensional visual image acquisition This embodiment uses a vehicle-mounted multi-view vision acquisition unit, which integrates a visible light industrial camera, an infrared thermal imaging camera, and a linear scan camera. While the detection vehicle is moving at a constant speed, it simultaneously captures images of the road surface, acquiring multi-dimensional visual images containing information on the road surface texture, temperature field distribution, and cross-sectional contour. This forms multi-modal raw image data, providing a multi-dimensional visual data source for subsequent detection of roadbed and pavement hazards and hidden defects.

[0046] Step 2: Location information synchronization and binding While acquiring images, the vehicle's location, latitude and longitude, and mileage information are obtained in real time through a high-precision GPS positioning module and mileage encoder. A hardware-triggered synchronization mechanism is adopted to match and bind each frame of multi-dimensional visual images with the corresponding geographic coordinates and mileage information with timestamps, generating image data packets with precise location markers, so as to achieve traceability and location of potential hazards.

[0047] Step 3: Removal of environmental interference in the image Environmental interference removal processing is performed on the multi-dimensional images after binding location information: for visible light images, dark channel dehazing, illumination equalization and distortion correction algorithms are used to eliminate the effects of dust, backlight and lens distortion; for infrared images, thermal noise filtering and temperature field normalization are performed to remove invalid interference caused by ambient temperature and road shadows, and output high-quality standardized image data.

[0048] Step 4: Extracting and cropping the region of interest Based on the width of the highway lane and the detection range, a region of interest (ROI) is preset. The preprocessed multi-dimensional image is automatically cropped to remove irrelevant background areas such as road edges, guardrails, and green belts, retaining only the effective road surface area within the lane. This reduces data redundancy while accurately locating the core detection area of ​​the roadbed and pavement, improving the efficiency of subsequent hazard identification.

[0049] Step 5: Edge Data Compression and Transmission The cropped effective image data is lightweight compressed and encoded by the vehicle-mounted edge computing terminal. A lossy compression algorithm based on the region of interest is used to prioritize the image clarity of the potential danger area and compress the background area appropriately. After compression, the data is transmitted to the cloud server at high speed through the 5G wireless communication module. This reduces the transmission bandwidth pressure and ensures that the cloud platform can quickly receive and carry out subsequent intelligent assessment of the roadbed and pavement health status.

[0050] This embodiment achieves end-to-end processing of multi-dimensional visual data of highways from acquisition, positioning, preprocessing to edge compression transmission through the above-described unidirectional data flow process. While ensuring data validity and location accuracy, it significantly reduces data transmission pressure and provides efficient and stable data support for cloud-based identification of roadbed and pavement hazards, mechanical inversion and health assessment.

[0051] Please refer to Figure 4 This invention provides a machine vision-based system for monitoring and evaluating the health status of highway subgrade and pavement, comprising the following steps: Step 1: Loading Multi-Dimensional Evaluation Indicators This embodiment is based on the results of front-end machine vision detection, hazard quantification, and subgrade mechanics inversion. It loads multi-dimensional subgrade and pavement health evaluation indicators. The indicator system is divided into two dimensions: pavement appearance state indicators (including crack density, pothole area, rut depth, and pavement damage index PCI) and subgrade internal state indicators (including subgrade settlement, horizontal slip displacement, soil equivalent elastic modulus, and base stress variation coefficient), forming a complete multi-dimensional evaluation indicator set, providing basic data for subsequent comprehensive evaluation.

[0052] Step 2: Assigning weights in the hierarchical analysis The Analytic Hierarchy Process (AHP) was used to construct a health status assessment matrix for roadbed and pavement. Combining the "Highway Technical Condition Assessment Standard" with the experience of road maintenance engineering experts, each evaluation indicator was compared and scored pairwise. After passing the consistency test, the weight coefficients corresponding to each indicator were calculated, highlighting the weight of indicators that affect structural safety, such as roadbed settlement and slippage, while weakening the weight of secondary apparent hazards, thus completing the scientific weight allocation of multiple indicators.

[0053] Step 3: Calculation of Road Section Health Index First, all evaluation indicators are standardized without dimensions to eliminate calculation bias caused by different dimensions. Then, combined with the weight coefficients obtained by the analytic hierarchy process, a weighted comprehensive scoring model is used to calculate the comprehensive health index of the roadbed and pavement of the corresponding road segment, realizing the conversion from multi-source detection data to a single quantitative health score.

[0054] Step 4: Automatic Health Level Determination The system presets the threshold ranges corresponding to the health index and divides the health status of the roadbed and pavement into five levels: excellent, good, medium, poor, and substandard, based on the highway maintenance specifications. The system automatically matches the corresponding threshold ranges according to the calculated road health index, completes the intelligent determination of the road health level, and simultaneously marks high-risk road sections with serious subsidence and slippage hazards.

[0055] Step 5: Generating Maintenance Priorities By combining three dimensions—road segment health level, traffic flow level, and hazard development rate—a maintenance priority ranking rule is constructed. All road segments are prioritized, and a maintenance priority list is generated, clearly defining emergency response road segments, planned maintenance road segments, and routine inspection road segments, providing a quantitative basis for highway maintenance decisions.

[0056] Step Six: Visualize and Store the Results The system generates visual results such as a health level distribution map containing road segment location information, a maintenance priority report, and a hazard heat map. It synchronously stores all detection data, health indices, judgment results, and maintenance plans in a cloud database. It also supports online querying, report export, and visualization on the maintenance management platform, enabling the practical application of roadbed and pavement health status assessment results.

[0057] This embodiment achieves intelligent evaluation of the entire process from machine vision detection data to roadbed and pavement health level and maintenance priority through the above-mentioned one-way data flow comprehensive evaluation process, which greatly improves the scientificity and accuracy of highway maintenance decisions.

Claims

1. A machine vision-based system for monitoring and evaluating the health status of highway subgrade and pavement, characterized in that, include: Image acquisition module, edge preprocessing module, intelligent hazard identification module, roadbed and pavement condition inversion module, comprehensive health assessment module, cloud management and scientific decision-making module; The image acquisition module is used to acquire multimodal visual images of the highway surface and simultaneously bind geographical location information to generate a raw image data stream with location tags. The edge preprocessing module is used to remove environmental interference, crop the region of interest, and compress and transmit edge data in the original image, and output standardized image data. The intelligent hazard identification module is used to identify road hazards in standardized images based on semantic segmentation algorithms, quantify hazards, and extract visual deformation features. The roadbed and pavement state inversion module is used to construct a mechanical mapping relationship based on visual deformation characteristics and invert the internal state parameters of roadbed settlement and slippage. The comprehensive health assessment module is used to load multi-dimensional evaluation indicators, use the analytic hierarchy process to assign weights, calculate the road segment health index, determine the health level, and generate maintenance priorities. The cloud management and scientific decision-making module is used to store the entire process of detection data, generate visual evaluation results, and output highway maintenance decision-making suggestions.

2. The machine vision-based highway subgrade and pavement health status monitoring and evaluation system according to claim 1, characterized in that... The image acquisition module includes: Multi-view vision acquisition unit, vehicle-mounted high-definition industrial camera and drone aerial photography are used to simultaneously acquire road surface texture, temperature field and cross-sectional contour information; The location synchronization binding unit uses a GPS positioning module and a mileage encoder to achieve one-to-one matching and binding of images with latitude, longitude, and mileage information based on timestamps.

3. The machine vision-based highway subgrade and pavement health status monitoring and evaluation system according to claim 1, characterized in that... The edge preprocessing module includes: The environmental interference removal unit performs dehazing, illumination equalization, and lens distortion correction on visible light images, and performs thermal noise filtering and temperature field normalization on infrared images. The region of interest cropping unit automatically removes irrelevant backgrounds such as road edges, guardrails, and green belts based on the preset lane detection range, retaining only the effective road surface detection area; The edge compression transmission unit uses a priority algorithm for hidden danger areas to encode images and transmits standardized data to the cloud via 5G wireless communication.

4. The machine vision-based highway subgrade and pavement health status monitoring and evaluation system according to claim 1, characterized in that... The intelligent hazard identification module includes: The U-Net semantic segmentation unit uses an improved U-Net network to achieve pixel-level segmentation of road surface cracks, potholes, ruts, subsidence, and skid marks, and outputs hazard masks and contour boundaries. The hazard quantification unit converts pixel size into physical size based on camera calibration parameters and calculates the equivalent values ​​of hazard length, area, and depth, as well as hazard density index. The visual deformation feature extraction unit performs sub-pixel-level contour analysis on the potential hazard area, extracts local curvature, edge gradient, and relative displacement features, and constructs a visual deformation feature vector.

5. A machine vision-based highway subgrade and pavement health status monitoring and evaluation system according to claim 1, characterized in that... The subgrade and pavement condition inversion module includes: The mechanical mapping unit establishes a mapping relationship between visual deformation characteristics and the elastic modulus of the subgrade, the shear strength of the soil, and the base stress, based on the theory of roadbed structural mechanics. The settlement and slip inversion unit, based on the mechanical mapping results and combined with the finite element inversion algorithm, calculates the vertical settlement and horizontal slip displacement of the roadbed, and quantitatively characterizes the degree of instability of the internal structure of the roadbed.

6. The machine vision-based highway subgrade and pavement health status monitoring and evaluation system according to claim 1, characterized in that... The comprehensive health assessment module includes: A multi-dimensional index loading unit is used to load pavement surface damage indexes and subgrade internal mechanical state indexes to construct a complete health evaluation index system. Hierarchical analysis weight assignment unit, construct judgment matrix and complete consistency test, determine the weight coefficient of each indicator; The health index calculation unit performs dimensionless standardization on each indicator and uses a weighted comprehensive scoring model to calculate the comprehensive health index of the road segment. The grade determination and priority generation unit divides the health status into five levels: excellent, good, medium, poor, and bad, based on preset thresholds, and generates a maintenance priority list by combining traffic flow and the rate of development of hidden dangers.

7. The machine vision-based highway subgrade and pavement health status monitoring and evaluation system according to claim 1, characterized in that... The cloud-based management and scientific decision-making module includes: The data storage unit uniformly stores data from the entire process of image acquisition, hazard quantification, roadbed inversion, and health assessment. The visualization output unit generates a heat map of potential hazards, a road section health level distribution map, and a maintenance priority report. The decision support unit outputs targeted maintenance recommendations based on the assessment results, enabling full life-cycle operation and maintenance management of highways.

Images