An asphalt pavement health status monitoring system and method

Abstract

The present application discloses an asphalt pavement health status monitoring system and method, belonging to the technical field of traffic infrastructure monitoring. The system includes the following modules: a multi-source data acquisition module for realizing real-time acquisition of pavement state parameters; a data processing module for performing spatio-temporal alignment, quality assessment, anomaly processing and standardization on the acquired multi-source data; a multi-scale feature extraction module for extracting features from the micro-damage, section state and network levels respectively to construct a multi-scale feature representation system; a dynamic graph analysis module for constructing a hierarchical dynamic graph structure to realize unified modeling and analysis of the road network health status; a health assessment module for realizing real-time assessment of the pavement health status, prediction of deterioration trends and risk early warning; and a decision visualization module for displaying monitoring data, prediction results and early warning information, and automatically generating maintenance decision suggestions based on the assessment results.

Description

Technical Field

[0001] This application relates to the technical field of traffic infrastructure monitoring, and more specifically, to a system and method for monitoring the health status of asphalt pavements. Background Art

[0002] With the rapid development of the transportation industry, the wide application of asphalt pavements has greatly promoted social and economic progress. However, under the long-term high-load operation and complex environmental impacts, asphalt pavements are prone to various diseases such as aging, cracks, ruts, and looseness. These diseases not only reduce the service performance of the pavement but also pose a threat to driving safety and increase the risk of traffic accidents. At the same time, the repair and maintenance of pavement diseases usually require a large amount of manpower, material resources, and financial resources. How to achieve scientific maintenance management has become an urgent problem to be solved.

[0003] At present, traditional health monitoring of asphalt pavements mainly relies on manual inspections and regular detections. This method has defects such as untimely data acquisition, high disease missed detection rate, and low diagnosis efficiency, and it is difficult to meet the dynamic maintenance needs of modern road networks. In recent years, with the continuous development of sensor technology, artificial intelligence, and Internet of Things technology, pavement monitoring technology based on multi-source data collection has gradually emerged. Through static sensors and dynamic acquisition devices, multi-dimensional data such as static characteristics, dynamic responses, and environmental impacts of the pavement can be obtained in real time, providing rich data support for pavement health status assessment.

[0004] However, the existing monitoring systems and methods still have many deficiencies in practical applications. On the one hand, there are problems of heterogeneity, spatio-temporal misalignment, and unstable quality in the process of multi-source data collection, resulting in ineffective data fusion and affecting the accuracy of health status assessment. On the other hand, the existing analysis methods usually only focus on the single-scale characteristics of pavement diseases and lack the systematic modeling ability for multi-scale characteristics at the micro, section, and network levels. At the same time, for the dynamic change characteristics of pavement health status, traditional methods are difficult to effectively capture spatio-temporal correlations, limiting the comprehensive assessment of the global health status of the road network. In addition, the current monitoring systems are still insufficient in risk prediction and maintenance decision-making, and fail to effectively combine multi-dimensional data with assessment results to provide scientific maintenance suggestions for decision-makers.

[0005] In summary, how to construct a system and method for monitoring the health status of asphalt pavements that can efficiently collect multi-source data, accurately extract multi-scale characteristics, dynamically analyze the health status of the road network, and achieve real-time assessment, trend prediction, and scientific decision-making support has become an urgent technical problem to be solved. Summary of the Invention

[0006] To overcome a series of defects existing in the prior art, the purpose of this application is to provide an asphalt pavement health status monitoring system for the above problems, including the following modules.

[0007] Multi-source data acquisition module, which realizes the real-time acquisition of pavement state parameters.

[0008] Data processing module, which performs spatio-temporal alignment, quality assessment, anomaly processing and standardization on the acquired multi-source data.

[0009] Multi-scale feature extraction module, which extracts features from the micro-disease, section state and network levels respectively, and constructs a multi-scale feature representation system.

[0010] Dynamic graph analysis module, which constructs a hierarchical dynamic graph structure to realize the unified modeling and analysis of the road network health status.

[0011] Health assessment module, which realizes the real-time assessment of the pavement health status, prediction of the deterioration trend and risk warning.

[0012] Decision visualization module, which displays the monitoring data, prediction results and warning information, and automatically generates maintenance decision suggestions based on the evaluation results.

[0013] Furthermore, the multi-source data acquisition module includes the following components.

[0014] Static sensor unit, which continuously monitors the static changes of the pavement structure and materials through sensors installed on the road surface, and provides basic pavement health data.

[0015] Dynamic acquisition equipment unit, which uses mobile equipment to collect pavement state data in real time and obtains dynamic load and pavement deformation information.

[0016] Environmental monitoring equipment, which collects external environmental data related to the pavement state.

[0017] Vibration sensor unit, which captures the vibration of the road surface under different loads through vibration sensors.

[0018] Image acquisition unit, which uses high-definition cameras to take real-time pictures of the road surface.

[0019] Equipment operation monitoring unit, which monitors the operation status of each data acquisition device, discovers equipment failures in time and performs necessary equipment maintenance.

[0020] Furthermore, the data processing module includes the following components.

[0021] Spatio-temporal alignment unit, which is responsible for spatio-temporal alignment of the time-series data from different data sources to ensure that various data can accurately reflect the pavement state in the same time and space range.

[0022] The data quality assessment unit is responsible for quality assessment and self-calibration of various types of collected data, and identifying data integrity, accuracy, and validity.

[0023] The anomaly handling unit is responsible for correcting or removing abnormal data.

[0024] The data standardization unit is responsible for converting data from different sources into a unified standard format and dimension.

[0025] The data fusion unit combines data from multiple sources into a high-precision integrated dataset.

[0026] The data synchronization and update unit is used to achieve the consistency and real-time update of multi-source data at different processing stages.

[0027] Furthermore, the multi-scale feature extraction module includes the following components.

[0028] The microscopic feature extraction unit captures the detailed features of pavement microscopic diseases through high-resolution convolutional layers.

[0029] The section feature extraction unit extracts the state features at the section level using convolutional layers with medium receptive fields.

[0030] The network-level feature extraction unit uses a global pooling layer to extract the macroscopic features at the road network level.

[0031] The multi-scale feature fusion unit fuses microscopic, section, and network-level features to construct a unified multi-scale representation system for a comprehensive characterization of the pavement state.

[0032] The scale interaction unit uses a feature pyramid structure to promote information interaction between features of different scales.

[0033] The feature enhancement unit enhances the discriminability and robustness of features through feature normalization operations to improve the adaptability to complex pavement states.

[0034] Furthermore, fusing microscopic, section, and network-level features to construct a unified multi-scale representation system for a comprehensive characterization of the pavement state includes the following steps.

[0035] By designing a multi-branch neural network structure, features at the microscopic, section, and network levels are processed separately, and the attention mechanism is used to learn the importance weights of features at different scales, and the features at the three levels are fused.

[0036] The fused features are mapped and transformed into a vector representation of a fixed dimension, which includes both microscopic behavior patterns and mesoscopic section states and macroscopic network features, thus achieving a comprehensive characterization of the pavement state.

[0037] Furthermore, the dynamic graph analysis module includes the following components.

[0038] The graph structure construction unit abstracts the road network into a hierarchical dynamic graph structure, where nodes represent road segments and edges represent connection relationships, providing a basic topological structure for modeling the health state of the road network.

[0039] The node feature embedding unit embeds the output results of the multi-scale feature extraction module into the graph nodes to characterize the health state and attribute information of each road segment.

[0040] The edge weight learning unit learns the weights of the edges between nodes through an adaptive mechanism, dynamically reflecting the spatial correlation and influence intensity between road segments.

[0041] The spatio-temporal graph convolution unit uses the spatio-temporal graph convolution network to simultaneously capture the spatial dependence and temporal dynamic changes of the road network, realizing cross-scale feature propagation and fusion.

[0042] The spatio-temporal attention unit introduces an attention mechanism to dynamically assign importance weights to different nodes and time periods.

[0043] The graph pooling unit reduces the dimension of the graph structure through hierarchical pooling operations, extracts global features at the road network level, and realizes the aggregation of health states from local to global.

[0044] Furthermore, reducing the dimension of the graph structure through hierarchical pooling operations, extracting global features at the road network level, and realizing the aggregation of health states from local to global include the following steps.

[0045] Construct an adjacency matrix based on the physical connection relationships between road segments and define the initial neighborhood range as the local receptive field.

[0046] Within the predefined neighborhood range, design a dedicated pooling function to preliminarily aggregate the node features and map the state information of adjacent road segments into a unified feature representation.

[0047] By iteratively expanding the receptive field range of the pooling operation, aggregate and abstract the features layer by layer, and finally obtain a global feature representation with significantly reduced dimensions but retaining key semantic information.

[0048] Furthermore, the health assessment module includes the following components.

[0049] The real-time state analysis unit uses multi-scale feature characterization and dynamic graph analysis results to calculate the health score and key performance indicators of the road surface in real time.

[0050] The deterioration trend prediction unit predicts the future change trend of the road surface health state based on time series analysis and machine learning algorithms, and identifies potential deterioration risks.

[0051] A risk warning generation unit automatically generates warning information based on the health status and prediction trend, including the deterioration risk level and the possible scope of influence.

[0052] A dynamic comparison and analysis unit compares historical data with the current evaluation results to identify abnormal changes or long-term trends.

[0053] A maintenance requirement assessment unit evaluates the maintenance requirements and urgency levels of each region based on the health status and risk assessment results.

[0054] Furthermore, the decision visualization module includes the following components.

[0055] A data access unit accesses monitoring data, prediction results, and warning information in real time to provide a dynamically updated data source for visualization.

[0056] A map display unit intuitively displays the road network health status, disease distribution, and warning information in the form of a heat map or a layered map based on geographic information system technology.

[0057] A chart generation unit displays the pavement health index, deterioration trend, and risk assessment results through line charts or bar charts.

[0058] A warning prompt unit highlights high-risk sections or sudden diseases in the visualization interface, combined with color coding and pop-up reminders.

[0059] A decision recommendation unit automatically generates maintenance decision recommendations including regional priority rankings based on the health assessment results and maintenance requirement assessment, and presents them in the form of structured text or charts.

[0060] An interaction operation unit provides user interaction functions, supporting managers to view detailed data and adjust decision parameters as needed.

[0061] A report generation unit automatically generates a visualization report summarizing monitoring data, evaluation results, and decision recommendations.

[0062] The purpose of this application is also to provide a method for monitoring the health status of asphalt pavements, including the following steps.

[0063] Real-time collect static characteristic data and dynamic monitoring data of the asphalt pavement through a sensor network.

[0064] Perform spatio-temporal alignment processing on the collected heterogeneous data, and use wavelet transform for noise reduction and standardization to construct a multi-modal fusion data set.

[0065] Based on the feature pyramid network, achieve adaptive fusion of multi-scale features and construct a systematic pavement state characterization system.

[0066] Combining spatio-temporal graph convolutional network and self-attention mechanism, extract the dynamic change features of the road network, and realize the multi-level aggregation and representation of features through the hierarchical graph pooling method.

[0067] Based on historical data and real-time monitoring data, realize multi-dimensional risk identification, and determine the priority level of maintenance needs through quantitative comparative analysis of multi-dimensional factors.

[0068] Based on the geographic information system platform, realize the visualization presentation of monitoring data, evaluation results and early warning information, and provide intelligent suggestions for maintenance decision-making.

[0069] Compared with the prior art, the present application has the following beneficial effects.

[0070] The present application realizes the real-time monitoring, evaluation and prediction of the pavement condition through modules such as multi-source data collection, data processing, multi-scale feature extraction, dynamic graph analysis, health assessment and decision visualization. This system can not only capture the microscopic diseases and section states of the pavement, but also conduct unified modeling and analysis at the network level, providing intelligent decision support for the maintenance and management of the pavement. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is a schematic structural diagram of an asphalt pavement health condition monitoring system disclosed in an embodiment of the present application.

[0072] Figure 2 It is a schematic flow diagram of an asphalt pavement health condition monitoring method disclosed in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0073] To make the objectives, technical solutions and advantages of the implementation of the present invention clearer, the technical solutions in the embodiments of the present invention will be described in more detail below with reference to the accompanying drawings in the embodiments of the present invention. In the drawings, the same or similar reference numerals denote the same or similar elements or elements with the same or similar functions from beginning to end. The described embodiments are some, but not all, of the embodiments of the present invention.

[0074] All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0075] The embodiments described below with reference to the accompanying drawings and directional terms are exemplary only and are intended to explain the present invention and should not be construed as limiting the present invention.

[0076] As Figure 1 shown, an asphalt pavement health condition monitoring system includes the following modules.

[0077] Multi-source data acquisition module, which constructs a multi-source data acquisition network through static sensors, dynamic acquisition devices and environmental monitoring devices to achieve real-time acquisition of road surface state parameters.

[0078] Data processing module, which performs spatio-temporal alignment, quality assessment, anomaly processing and standardization on the acquired multi-source data.

[0079] Multi-scale feature extraction module, which extracts features from the micro-level diseases, section states and network levels respectively, constructs a multi-scale feature representation system, and realizes the comprehensive characterization of the road surface state.

[0080] Dynamic graph analysis module, which constructs a hierarchical dynamic graph structure and realizes the unified modeling and analysis of the road network health state through cross-scale feature fusion and spatio-temporal attention mechanism.

[0081] Health assessment module, which realizes the real-time assessment of the road surface health state, prediction of the deterioration trend and risk warning based on the multi-scale feature representation and the results of dynamic graph analysis.

[0082] Decision visualization module, which provides an intuitive visualization interface, displays the monitoring data, prediction results and warning information, and automatically generates maintenance decision suggestions based on the assessment results, providing a decision-making basis for managers.

[0083] In this embodiment, the multi-source data acquisition module is the basis of the asphalt pavement health state monitoring system. Its function is to construct a multi-level acquisition network through static sensors, dynamic acquisition devices and environmental monitoring devices to collect relevant data on the asphalt pavement health state in real time. Static sensors such as strain gauges, pressure sensors, temperature sensors, etc. can be stably installed at various positions on the road surface for long-term monitoring of changes at different positions. These sensors can acquire mechanical, temperature and other parameters on the road surface and inside in real time, providing accurate road surface state information. Dynamic acquisition devices such as sensors carried by vehicles, drones, road surface inspection vehicles, etc. can collect data during movement to obtain minute changes on the road surface. These dynamic data are crucial for capturing short-term changes and sudden situations that may occur during the actual use of the road surface. In addition, environmental monitoring devices can monitor influencing factors in the surrounding environment, such as humidity, precipitation, wind speed, etc. These factors have an important impact on the road surface state. Through the fusion of these multi-source data, it is possible to achieve all-round and real-time state monitoring of the asphalt pavement, providing accurate data support for subsequent analysis and decision-making.

[0084] In this embodiment, the core task of the data processing module is to process and optimize multi-source data to ensure the quality, integrity, and validity of the data. First, through spatio-temporal alignment technology, it ensures that data from different sensors and devices are accurately matched in terms of time and space dimensions. This is because there are differences in the acquisition frequencies and spatial coverage ranges of different devices, and the data needs to be effectively aligned for subsequent analysis. Secondly, the quality assessment module screens the collected data, removing noise and incomplete parts to ensure that subsequent analysis is not interfered by bad data. In this process, anomaly detection technology also plays an important role, being able to identify and process abnormal fluctuations in the data, such as data errors caused by equipment failures or environmental changes. Finally, data standardization processing converts data from different sources and different types into a unified format and scale, enabling all data to be fused and compared on the same platform, laying a foundation for subsequent feature extraction and health assessment.

[0085] In this embodiment, the design purpose of the multi-scale feature extraction module is to comprehensively reflect the health status of the road surface through feature extraction at different levels. From a microscopic perspective, this part focuses on extracting small-scale disease characteristics on the road surface, such as surface defects like cracks, potholes, and peeling. These microscopic diseases often have a long-term impact on the durability of the road surface, so their detection and analysis are crucial. Then, the extraction of section state characteristics focuses on larger-scale road surface areas, such as the flatness of the section, road surface bearing capacity, and damage accumulation. By analyzing the overall health status of the section, it can provide a basis for the repair and maintenance of the section. Finally, the network-level feature extraction is a comprehensive assessment of the health status of the entire road network, including factors such as the connectivity between sections, traffic capacity, and maintenance requirements. The characteristic of this module is that it can depict the health status of the road surface from multiple scales and multiple dimensions, providing all-round feature support for subsequent dynamic graph analysis.

[0086] In this embodiment, the dynamic graph analysis module realizes the unified modeling and analysis of the health status of the road network by constructing a hierarchical dynamic graph structure. The health status of the road surface is not only a time series problem but also a spatial structuring problem. The dynamic graph structure enables flexible modeling of each section in the road network in terms of time and space, and can effectively represent the state changes of different sections at different times. In this module, cross-scale feature fusion is one of the core technologies. By fusing microscopic, section, and network-level features, it can comprehensively understand the dynamic change process of road surface health. At the same time, the spatio-temporal attention mechanism helps the system dynamically adjust the focus of attention in different time and space dimensions during the analysis process, so as to better capture the key factors of road surface health changes. This module can not only analyze a single section but also conduct a comprehensive health status assessment of the entire road network, providing more accurate support for decision-making and management.

[0087] In this embodiment, the health assessment module utilizes multi-scale features and the results of dynamic graph analysis to conduct real-time assessment of the pavement health status, prediction of deterioration trends, and risk early warning. The real-time assessment function can help managers keep track of the pavement health condition at any time, promptly detect potential hazards, and make necessary maintenance decisions based on the assessment results. The deterioration trend prediction analyzes historical data and the current state to predict the future health change trend of the pavement, assisting relevant departments to take corresponding measures in advance. The risk early warning function issues warnings in advance when potential problems occur on the pavement, reminding managers to take measures in a timely manner. This module can conduct real-time assessment of pavement health through intelligent algorithms, and make more accurate predictions and decisions by considering the influencing factors of the external environment.

[0088] In this embodiment, the decision visualization module can convert complex pavement health data and analysis results into graphical information that is easy to understand and operate, such as heat maps, trend charts, warning prompts, etc. These information can help pavement managers quickly understand the current road conditions, predict future changes, and make corresponding maintenance decisions. Based on the assessment results, it can automatically generate maintenance decision suggestions, providing a scientific basis for managers, ensuring the optimal allocation of maintenance resources, reducing the occurrence of road damage, and extending the service life of the pavement. Through the visualization module, the entire system not only improves the transparency of data, but also makes pavement health management more intelligent and systematic.

[0089] The above-mentioned modules work together to jointly form a complete monitoring system for the health status of asphalt pavements. Through technologies such as multi-source data collection, precise data processing, multi-scale feature extraction, and dynamic graph analysis, it can comprehensively and real-time monitor and evaluate the health status of the pavement. Combining the health assessment and decision visualization modules not only provides managers with accurate health status assessment and deterioration trend prediction, but also optimizes the allocation of maintenance resources through intelligent decision suggestions.

[0090] Furthermore, the multi-source data collection module includes the following components.

[0091] The static sensor unit continuously monitors the static changes of the pavement structure and materials through sensors installed on the pavement, providing basic pavement health data.

[0092] The dynamic acquisition device unit uses mobile devices to collect pavement status data in real-time, obtaining dynamic load and pavement deformation information.

[0093] The environmental monitoring device collects external environmental data related to the pavement status, providing reference for external factors affecting the change of the pavement status.

[0094] The vibration sensor unit captures the vibration of the pavement under different loads through vibration sensors, and analyzes pavement damage and potential problems.

[0095] The image acquisition unit uses a high-definition camera to capture road surface images in real time to assist in determining surface damage or aging.

[0096] The equipment operation monitoring unit monitors the operating status of each data acquisition device, detects equipment failures in a timely manner and performs necessary equipment maintenance.

[0097] In this embodiment, the static sensor unit is one of the core components of the multi-source data acquisition module. Its main function is to monitor the static changes of the pavement structure and materials in real time through sensors installed on the pavement. These sensors are usually installed at different depths and positions inside the pavement, including strain gauges, temperature sensors, humidity sensors, etc. These devices can continuously collect basic status data of the pavement during long-term use, such as temperature changes, humidity fluctuations, pavement stress conditions, etc. These data are crucial to understanding the changing trends of the pavement health status, especially for problems such as road aging, settlement, and cracking under long-term use. The advantage of static sensors is that they can collect data stably and continuously without excessive manual intervention, thereby providing a reliable long-term data basis for subsequent analysis. In addition, static sensors can also realize real-time data upload, ensuring instant control of the pavement status. The application of this unit provides data guarantee for accurate assessment of the pavement health status and is the "eye" of the monitoring system.

[0098] In this embodiment, the dynamic acquisition device unit uses mobile equipment, such as inspection vehicles, drones, intelligent inspection vehicles, etc., to collect status data in real time on the road surface, especially the performance of the road surface under traffic load. Compared with static sensors, dynamic acquisition equipment can obtain information such as dynamic load, road deformation, rutting, cracks, etc. with the help of vehicle driving or other mobile devices. This information has a higher practical significance in reflecting the health status of the road surface, because the state change of the road surface under different loads and traffic conditions is usually the most critical evaluation indicator. For example, in sections with high traffic density, dynamic acquisition can accurately record the deformation of the road surface under the rolling of heavy vehicles, helping to determine the potential risk of road damage. The advantage of dynamic acquisition equipment is that it can cover a wider range of sections, obtain real-time and comprehensive health data, and can be compared with static data to help managers better identify and predict problems. Dynamic acquisition equipment can collect data efficiently and extensively, becoming an important supplement to the analysis of road health status.

[0099] In this embodiment, the environmental monitoring device plays a role in providing a reference for external environmental impacts in the multi-source data acquisition module. The health of the road surface is not only affected by changes in its own materials and structure, but also disturbed by external environmental factors such as temperature, humidity, precipitation, wind speed, and ultraviolet rays. For example, frequent precipitation or extreme temperature changes can accelerate the aging process of the road surface. Especially in asphalt road surfaces, an environment with a large temperature difference may cause cracking or deformation of the road surface. Therefore, by installing meteorological monitoring instruments (such as temperature sensors, humidity sensors, barometric pressure sensors, precipitation monitors, etc.), the environmental monitoring device can provide meteorological data related to changes in the road surface state. The combination of these external data with road surface health data can effectively explain the causes of road surface damage and provide more accurate background information for formulating maintenance strategies. Through the environmental monitoring device, external environmental factors can be closely combined with road surface health data, helping to analyze the interaction between road surface damage and environmental conditions and improving the comprehensiveness and accuracy of monitoring and analysis.

[0100] In this embodiment, the vibration sensor unit is used to capture the vibration conditions of the road surface under different loads, especially the vibration signals generated when traffic vehicles pass by. These vibration signals can reflect the force conditions of the road surface and possible structural problems. For example, when the road surface is subjected to a large load, the vibration sensor can record the corresponding vibration amplitude and frequency. By analyzing these data, it can be determined whether there are structural damages or potential cracks, settlements, etc. on the road surface. The vibration sensor unit can sensitively detect minute vibration changes, especially on sections with heavy traffic and frequent load changes, and can timely detect potential road surface problems. Through the collection and analysis of vibration data, potential safety hazards can be warned in advance, providing an accurate basis for maintenance decisions. In addition, the use of vibration sensors is not dependent on changes in the external environment, so it has strong stability and reliability.

[0101] In this embodiment, the image acquisition unit uses a high-definition camera or imaging system to capture road surface images in real time, which is an important technical means to assist in judging surface damage, aging, or other diseases of the road surface through visual means. The high-definition camera can be installed above the road surface or on vehicle-mounted equipment. Through automated image acquisition and analysis, surface damages such as cracks, potholes, peeling, and ruts can be detected. These surface damages directly affect the driving safety of the road surface, so accurate and efficient detection is required. The advantage of the image acquisition unit lies in its high resolution and real-time performance, which can intuitively reflect the subtle damages on the road surface and can automatically identify and classify different types of damages. In addition, by combining with data from other sensors, image data can be further used to verify or supplement the results of other data acquisition means, improving the overall accuracy. This unit can intuitively and real-time capture road surface problems, which is an effective supplement to traditional detection means.

[0102] In this embodiment, the main function of the device operation monitoring unit is to monitor the operation status of each data acquisition device in real time, ensure the normal operation of the devices, and be able to respond in a timely manner when a failure occurs. With the increasing use of sensors and dynamic acquisition devices, the operation status of the devices is crucial for the data quality and the accuracy of the monitoring results. This unit can ensure the normal operation of the devices at any time by integrating various monitoring technologies, such as battery power monitoring, temperature monitoring, signal transmission monitoring, etc. When a device failure or performance degradation is detected, an alarm will be automatically triggered and maintenance or replacement will be carried out in a timely manner. This unit improves the stability and reliability of data acquisition, ensuring the continuous and efficient operation of the monitoring system. The device operation monitoring unit provides guarantee for the long-term stable operation of the entire system, reducing data loss or errors caused by device failures.

[0103] In summary, through the collaborative work of the above-mentioned components, the multi-source data acquisition module can efficiently and accurately monitor the health status of asphalt pavements in real time. The static sensor unit provides basic pavement structure data, the dynamic acquisition device unit obtains the pavement performance under actual loads, the environmental monitoring device provides important references for external environmental factors, the vibration sensor unit captures potential structural problems by detecting vibrations, the image acquisition unit supplements surface disease information through visual analysis, and the device operation monitoring unit ensures the normal operation of data acquisition devices. Each component cooperates with each other to form a comprehensive, accurate and efficient pavement health monitoring system, providing strong data support for subsequent analysis and decision-making.

[0104] Furthermore, the data processing module includes the following components.

[0105] The spatio-temporal alignment unit is responsible for spatio-temporal alignment of time-series data from different data sources, ensuring that various data can accurately reflect the pavement state within the same time and space range.

[0106] The data quality assessment unit conducts quality assessment and self-calibration on various types of acquired data, identifies data integrity, accuracy and validity, and ensures data reliability.

[0107] The anomaly processing unit corrects or eliminates abnormal data to ensure the accuracy of the final data set and avoid interference with subsequent analysis.

[0108] The data standardization unit converts data from different sources into a unified standard format and dimension, ensuring seamless integration of multi-source data and compatibility with subsequent analysis.

[0109] The data fusion unit combines data from multiple sources into a high-precision comprehensive data set based on spatio-temporally aligned and standardized data.

[0110] A data synchronization and update unit for achieving the consistency and real-time update of multi-source data at different processing stages.

[0111] In this embodiment, the spatio-temporal alignment unit is one of the core components of the data processing module, mainly responsible for spatio-temporal alignment of the time-series data from different data sources to ensure that various data can accurately reflect the road surface conditions within the same time and space range. Different types of data adopt different timestamps and spatial positioning methods. Therefore, it is crucial to ensure the consistency of various data in the time and space dimensions. The spatio-temporal alignment unit precisely aligns the time-series data through various algorithms to ensure that the measurement data from different data sources can be synchronized in the time dimension and can be matched to the specific area of road surface monitoring in the spatial position. For example, the static sensor data and the dynamic acquisition device data may have different sampling frequencies, and the spatio-temporal alignment unit will synchronize and align the data through interpolation algorithms, time window methods or spatial interpolation methods. Through this process, the data can be analyzed within a unified spatio-temporal framework, providing an accurate basis for subsequent data processing and health status assessment. The technical effect of the spatio-temporal alignment unit is to eliminate the time differences and spatial errors from different data sources, improving the reliability of multi-source data fusion.

[0112] In this embodiment, during the multi-source data acquisition process, data anomalies may occur due to equipment failures, signal interference or other external factors. If not processed, it will affect the accuracy and stability of the entire monitoring system. The data quality assessment unit evaluates the integrity, accuracy and consistency of the data through various algorithms and makes automatic corrections or calibrations based on the evaluation results. For example, for missing data, interpolation algorithms can be used to fill it; for outliers, they can be automatically identified and excluded according to the distribution characteristics of historical data. In addition, the data quality assessment unit can also adjust the acquisition frequency and acquisition method in real time according to the specific patterns of the data and environmental conditions to optimize the data quality. This unit ensures the reliability of the final data set, avoids the adverse effects of invalid or distorted data on subsequent analysis, and further improves the accuracy of the system.

[0113] In this embodiment, the core function of the anomaly handling unit is to identify and handle the abnormal data that appears during the data acquisition process, ensuring the accuracy and consistency of the final dataset. In practical applications, sensors may generate incorrect data due to environmental interference, insufficient power, hardware failures, or communication problems. If these abnormal data are not eliminated or corrected in a timely manner, they may have a serious impact on subsequent analysis and decision-making. The anomaly handling unit can automatically identify and classify abnormal data through anomaly detection algorithms, combined with the historical patterns of the data and the currently acquired data. The processing methods for outliers usually include methods such as elimination, replacement, and correction. For example, for abnormal data far from the normal range, it can be directly eliminated; for outliers slightly deviating from the normal data, data smoothing algorithms can be used for correction. The anomaly handling unit can significantly improve the accuracy of the data, reduce the interference caused by incorrect data to the pavement health assessment, and ensure the effectiveness of subsequent analysis results.

[0114] In this embodiment, the data normalization unit is designed to address the heterogeneous data formats and dimensional differences from different data sources. Since different acquisition devices and sensors may use different units, dimensions, or data representation methods, the role of the data normalization unit is to convert all data from different sources into a unified standard format, so that the data can be seamlessly integrated and used for subsequent analysis. For example, pavement temperature data may be represented in degrees Celsius, while pressure data may be in pascals. Through the normalization unit, these data are converted into a unified unit and dimension to ensure that they are compatible and can be combined for analysis. Normalization is not limited to unit conversion, but may also include unifying the data scale, converting data types, etc., thus eliminating the differences between different data sources and ensuring the efficient integration of multi-source data. The data normalization unit provides a unified data processing framework, enabling data from various sensors and devices to be compared and analyzed on the same basis, improving the efficiency and accuracy of data processing.

[0115] In this embodiment, the data comes from multiple channels such as static sensors, dynamic acquisition devices, and environmental monitoring devices. The sources and formats of the data are diverse. Through the data fusion unit, these scattered data can be effectively integrated into a unified and high-precision dataset. For example, the structural data provided by static sensors and the traffic load data provided by dynamic acquisition devices can be merged through the data fusion unit. Through multi-level analysis and modeling, a complete pavement health portrait can be formed. Data fusion techniques include various methods such as weighted average method, Kalman filter, and Bayesian inference. The most suitable fusion method is selected according to the characteristics of the data and the analysis requirements. This unit improves the accuracy and reliability of the data through precise fusion of multi-source data, providing comprehensive data support for pavement health status assessment.

[0116] In this embodiment, the data synchronization and update unit plays an important supporting role in the entire data processing process. Due to the high real-time requirements of multi-source data, and there may be differences in the update frequency and timestamp of different data sources, the data synchronization and update unit is responsible for ensuring the synchronization and consistency of different data sources. This unit monitors the data collection process in real time, updates the data in the database in a timely manner, and performs real-time update and calibration on the data according to the preset synchronization mechanism. For example, after spatio-temporal alignment, the data synchronization and update unit is responsible for ensuring that the newly collected data is merged into the processing flow in a timely manner, and ensuring that historical data will not be overwritten or lost due to new data collection. In addition, with the continuous changes in the environment and road surface conditions, the data synchronization and update unit can also achieve dynamic real-time updates to ensure that the data always remains up-to-date and can adapt to the changing monitoring requirements. The technical effect of the data synchronization and update unit is to ensure the real-time, consistency, and accuracy of the data, avoid data lag or obsolescence problems, and improve the response speed of the monitoring system.

[0117] In summary, through the close cooperation of the above components, the data processing module effectively guarantees the quality, accuracy, and availability of multi-source data. The spatio-temporal alignment unit ensures the consistency of data from different sources in terms of time and space; the data quality assessment unit evaluates the reliability and effectiveness of the data; the anomaly processing unit effectively corrects or eliminates abnormal data; the data standardization unit achieves the consistency of data formats and units; the data fusion unit provides accurate road surface health data through the efficient integration of multi-source data; the data synchronization and update unit ensures the real-time and consistency of the data. The comprehensive application of these functions not only improves the accuracy of the data but also provides strong data support for the real-time assessment, early warning, and decision-making of the road surface health status.

[0118] Furthermore, the multi-scale feature extraction module includes the following components.

[0119] The microscopic feature extraction unit captures the detailed features of road surface microscopic diseases through high-resolution convolutional layers, providing a basis for local damage analysis.

[0120] The section feature extraction unit extracts the state features at the section level using convolutional layers with medium receptive fields, reflecting the overall condition of the local section.

[0121] The network-level feature extraction unit uses global pooling layers to extract the macroscopic features at the road network level, depicting the global road network state.

[0122] The multi-scale feature fusion unit fuses microscopic, section, and network-level features to construct a unified multi-scale representation system, achieving a comprehensive characterization of the road surface state.

[0123] A scale interaction unit uses a feature pyramid structure to promote information interaction between features of different scales and enhance the coherence of feature expression.

[0124] A feature enhancement unit enhances the distinctiveness and robustness of features through feature normalization operations to improve the adaptability to complex road surface conditions.

[0125] In this embodiment, the microscopic feature extraction unit focuses on capturing the detailed features of road surface microscopic diseases through a high-resolution convolutional layer, which is the basis for analyzing local road surface damage. In asphalt pavement health monitoring, microscopic diseases usually refer to structural damages within a small range such as cracks, potholes, and raveling. These problems are not easily detectable in the early stage but can lead to a rapid deterioration of road surface health in a short time. Through the high-resolution convolutional layer, the microscopic feature extraction unit can accurately capture the subtle damages on the road surface and identify detailed features such as tiny cracks, spalling, and rutting through detailed image analysis. These features provide important data support for subsequent local damage analysis, helping to quickly locate problem areas and take corresponding repair measures. Thus, the microscopic feature extraction unit greatly improves the recognition accuracy and speed of local diseases, enabling accurate diagnosis when the damage is still in the early stage, thereby avoiding the further expansion of road surface diseases, reducing maintenance costs, and increasing the service life of the road surface.

[0126] In this embodiment, the processing range of the section feature extraction unit is larger than that of the microscopic feature extraction unit and focuses on identifying the state within a longer section or area of the road. For example, section features can include the overall flatness of the road surface, the distribution of transverse and longitudinal cracks, rut depth, uneven compaction, etc. The convolutional layer with a medium receptive field can integrate the surrounding road surface information and extract the state features of the entire section, helping to detect structural problems in a larger range rather than just local subtle damages. Through the section feature extraction unit, common problems in the section can be identified, such as the effects of long-term water accumulation and continuous high-load traffic. This unit provides more comprehensive data support for the overall condition assessment at the section level, ensuring that the monitoring system not only focuses on individual problems but also can capture more complex potential risks that may affect a larger area, thereby achieving accurate judgment and early warning of the road surface health status.

[0127] In this embodiment, the network-level feature extraction unit conducts an overall assessment of the health status of the entire road network, rather than being limited to a specific section or area. Through the global pooling layer, the network-level feature extraction unit can extract macroscopic features that span multiple sections or even the entire road network, such as the distribution of structural damage across the entire road network, the degree of pavement aging, and the distribution of traffic loads. By compressing and summarizing the features within the entire road network, the global pooling layer can effectively capture the health status of the large-scale road network, provide global information to decision-makers, help them understand the differences in health status among different sections, and promptly identify potential problems that may pose a threat to the entire road network. In terms of technical effects, the network-level feature extraction unit can achieve dynamic monitoring and assessment of the entire road network, promote the intelligentization of pavement maintenance and management, enable decision-makers to formulate more scientific and accurate maintenance strategies, and effectively prevent large-scale systemic failures in the road network.

[0128] In this embodiment, since the pavement health status involves factors at multiple levels, from microscopic damage to the overall condition of a section and then to the macroscopic performance of the entire road network, features at different scales each reflect information at different levels, and single-scale features often cannot comprehensively describe pavement health. Through the multi-scale feature fusion unit, features at different levels can be effectively combined to form a comprehensive, cross-scale health assessment model. For example, when there is local microscopic damage in a certain section, combining the macroscopic features of the entire road network can further evaluate the potential impact of this damage on the health of the entire road network. This unit ensures the full utilization of health information from local to global through cross-scale fusion. In terms of technical effects, the feature fusion unit significantly improves the perception ability of complex pavement conditions, can accurately depict the overall and local health status of the pavement through the combination of multi-level information, and provides comprehensive data support for subsequent health assessment and decision-making.

[0129] In this embodiment, the scale interaction unit promotes the information interaction between features of different scales by using the feature pyramid structure, thereby enhancing the coherence of feature expression. Features of different scales contain different levels of road surface information, and this information is often interdependent and interactive. For example, microscopic features can reflect local early damage, while section features can reveal whether these local problems have spread to a larger area, and network-level features provide a global perspective on road surface health. The scale interaction unit enables the effective mutual transmission of information of different scales through the pyramid structure, optimizes the expression of features, and enables each layer of features to obtain the support of information of other scales. For example, through the interaction structure, section-level features can more accurately judge the impact of local diseases with the help of microscopic feature information, and global features can help optimize the interpretation of local features. In terms of technical effects, the scale interaction unit greatly enhances the correlation between features of different levels, making the entire feature expression system more coherent and comprehensive, thereby providing richer and more accurate basic data for subsequent analysis.

[0130] In this embodiment, the feature enhancement unit enhances the discriminability and robustness of features through feature normalization operations, thereby improving the adaptability to complex road surface conditions. In road surface health monitoring, data differences caused by different environmental conditions and acquisition methods are often faced. The feature enhancement unit makes features of different scales and different sources comparable and analyzable under the same standard through normalization operations. Feature normalization operations not only help reduce the deviation caused by different data sources, but also enhance the stability of features, enabling the system to have stronger recognition ability for different types of road surface damage. In addition, the enhanced features can better adapt to the changes in complex road surface conditions, and can maintain high accuracy and stability whether under different climate conditions or different load conditions. In terms of technical effects, the feature enhancement unit improves the adaptability of the road surface health status monitoring system to various complex conditions, thereby enabling efficient detection capabilities in various environments, and ensuring the accuracy and reliability of road surface health status assessment.

[0131] In summary, the multi-scale feature extraction module effectively improves the monitoring and analysis capabilities of road surface health status through multi-level feature extraction at the microscopic, section, and network levels, combined with scale interaction and feature enhancement technologies. The microscopic feature extraction unit finely captures early diseases, the section feature extraction unit provides a more comprehensive assessment of local problems, and the network-level feature extraction unit analyzes the health of the road network from a global perspective. The multi-scale feature fusion unit ensures the comprehensive utilization of information at different levels, and the scale interaction unit improves the circulation of information at each scale through the pyramid structure. Finally, the feature enhancement unit optimizes the discriminability and robustness of features, and improves the adaptability to complex road surface conditions.

[0132] Furthermore, the details of the microscopic road surface diseases are captured by a high-resolution convolutional layer, including the following steps.

[0133] Perform a convolution operation on the processed road surface image to generate multiple feature maps, which can be expressed by the formula: Y(i,j)=(I*K)(i,j)=∑ m ∑ n I(i+m,j+n)·K(m,n), where Y(i,j) represents the pixel value at position (i,j) in the output feature map after the convolution operation; (i,j) represents the coordinates of the pixel in the image, where i is the row coordinate and j is the column coordinate; I represents the input road surface image; K is the convolution kernel; * represents the convolution operation; (I*K) represents the result of the convolution operation between the input image I and the convolution kernel K, and the output is a feature map; m and n represent the row and column positions of the convolution kernel K, which are used to index each element of the convolution kernel K when calculating the convolution; I(i+m,j+n) represents a pixel in the input image I at position (i+m,j+n); K(m,n) represents an element in the convolution kernel K at position (m,n).

[0134] Use convolution kernels K of different sizes 1 ,K 2 ,...,K n to perform multi-scale convolution operations on the processed image, so as to extract multi-scale features, and fuse the multi-scale features through the following formula: Y multi-scale =[(I*K 1 ),(I*K 2 ),...,(I*K n )], where Y multi-scale represents the fusion result of the multi-scale feature map; (I*K 1 ) represents the result of the convolution operation between the input image I and the convolution kernel K 1 ; (I*K 2 ) represents the result of the convolution operation between the input image I and the convolution kernel K 2 ; (I*K n ) represents the result of the convolution operation between the input image I and the convolution kernel K n .

[0135] Adopt the ReLU activation function to perform activation operations on the convolution feature map, remove invalid features and strengthen the response of local damage to highlight the details of the road surface, which can be expressed by the formula: A(i,j)=ReLU(Y(i,j)), where A(i,j) represents the activation value at the (i,j) position in the feature map after being processed by the ReLU function; ReLU(·) represents the ReLU activation function.

[0136] Input the activated feature map into the classification network. By combining with the damage classification weight matrix W damage and the bias term b, use the Softmax function to obtain the classification probability distribution of the damage: P damage =Softmax(W damage ·A + b), where P damage represents the probability distribution of damage classification and is the result output by the classification layer of the neural network; A represents the feature map after convolution and activation function processing; Softmax(·) represents the Softmax function.

[0137] In summary, through steps such as high-resolution convolutional layers, convolutional operations at different scales, ReLU activation functions, and damage classification networks, a complete microscopic disease detection process is constructed. Convolutional operations can extract local features of pavement images to ensure the capture of detailed damages; multi-scale convolutional operations enhance the adaptability to diseases of different scales and improve the overall detection accuracy; the ReLU activation function effectively removes invalid features and strengthens the response of disease features; and damage classification provides accurate classification and probability distribution for pavement diseases through the Softmax function. The combination of these technical means makes the detection of pavement microscopic diseases more efficient, accurate, and reliable, providing strong data support for intelligent maintenance decision-making and pavement health management.

[0138] Furthermore, a global pooling layer is adopted to extract macroscopic features at the road network level and characterize the global road network state, including the following steps.

[0139] Design a multi-layer convolutional neural network. Starting from the features of the underlying road segments, gradually extract the road network structure features of increasing levels through layer-by-layer convolutional operations, and finally form a multi-scale representation of the road network.

[0140] Introduce global pooling operations after the convolutional layer to compress the multi-scale road network representation into a feature vector of a fixed dimension, realizing the conversion from local features to global features.

[0141] Perform non-linear transformation and dimensionality reduction on the pooled feature vector through a fully connected layer, synthesize the multi-dimensional features of the road network, and generate a low-dimensional compact representation that effectively characterizes the overall road network state.

[0142] In summary, by designing a multi-layer convolutional neural network, introducing global pooling operations and fully connected layers for non-linear transformation and dimensionality reduction, it is possible to effectively extract and characterize macroscopic features at the road network level. This process extracts rich road network structure information from local to global, providing strong data support for road network health condition assessment. The hierarchical feature extraction of the convolutional neural network can capture road network features at different levels, while global pooling ensures the effective retention of global information by compressing the feature map. The dimensionality reduction and non-linear transformation of the fully connected layer further enhance the abstraction ability of the features, and finally output a concise and effective low-dimensional representation, which is suitable for subsequent road network state assessment and prediction tasks.

[0143] Furthermore, by fusing microscopic, segment-level, and network-level features, a unified multi-scale representation system is constructed to comprehensively characterize the pavement state, including the following steps.

[0144] By designing a multi-branch neural network structure, features at the microscopic, segment, and network levels are processed separately, and the attention mechanism is used to learn the importance weights of features at different scales, and the features at the three levels are fused to achieve the optimal combination of features.

[0145] The fused features are converted into a vector representation with a fixed dimension through mapping, which includes both microscopic behavior patterns and mesoscopic segment states and macroscopic network features, so as to comprehensively characterize the pavement state.

[0146] In summary, by designing a multi-branch neural network, introducing the attention mechanism, and strategies of feature fusion and mapping transformation, it is possible to effectively integrate information from different scales, thus comprehensively characterizing the pavement state. The multi-branch network ensures the accurate extraction of features at different levels, and the attention mechanism further optimizes the feature fusion process by assigning dynamic importance weights to features at each scale. Finally, the fused features are compressed into a vector with a fixed dimension through mapping transformation, which not only retains the detailed information of the pavement state but also improves the computational efficiency. Overall, this multi-scale feature fusion method improves the accuracy and flexibility of pavement state assessment, providing strong technical support for pavement health monitoring in intelligent transportation systems.

[0147] Furthermore, the dynamic graph analysis module includes the following components.

[0148] A graph structure construction unit abstracts the road network into a hierarchical dynamic graph structure, where nodes represent road segments and edges represent connection relationships, providing a basic topological structure for modeling the road network health state.

[0149] A node feature embedding unit embeds the output results of the multi-scale feature extraction module into the graph nodes to represent the health state and attribute information of each road segment.

[0150] Edge weight learning unit, which learns the weights of the edges between nodes through an adaptive mechanism, and dynamically reflects the spatial correlation and influence intensity between road segments.

[0151] Spatio-temporal graph convolution unit, which uses a spatio-temporal graph convolution network to simultaneously capture the spatial dependence and temporal dynamic changes of the road network, and realizes cross-scale feature propagation and fusion.

[0152] Spatio-temporal attention unit, which introduces an attention mechanism to dynamically assign importance weights to different nodes and time periods, and focuses on the health state changes of key road segments and key time points.

[0153] Graph pooling unit, which reduces the dimension of the graph structure through hierarchical pooling operations, extracts global features at the road network level, and realizes the aggregation of health states from local to global.

[0154] In this embodiment, the goal of the graph structure construction unit is to abstract the complex road network into a hierarchical dynamic graph structure for subsequent health state analysis. Nodes in this graph structure represent each road segment of the road network, while edges represent the connection relationships between these road segments. In this way, the topological structure of the road network can be formally represented, facilitating the modeling and analysis of its health state. Specifically, the graph structure provides a basic framework for subsequent dynamic graph analysis, enabling us to model the road network in the spatial and temporal dimensions. For example, a certain road segment of the road network is affected by multiple factors such as traffic flow, environmental factors, and maintenance history, and these effects can be modeled through the edges between nodes. In terms of technical effects, the graph structure enables the health state of the road network to be comprehensively modeled at different levels, including the hierarchical structure from local road segments to the entire road network, improving the accuracy and flexibility of road network state analysis. By constructing a hierarchical dynamic graph, the analysis strategy can be dynamically adjusted according to the actual topological structure of the road network, thereby enhancing the self-adaptability and scalability of the model.

[0155] In this embodiment, the task of the node feature embedding unit is to embed the output results obtained from the multi-scale feature extraction module into each node of the graph, so as to provide health status and attribute information for each road segment. The core of this embedding process lies in learning to map high-dimensional features to low-dimensional representations of each node, facilitating subsequent graph convolution operations. These features can include various aspects of information such as microscopic damage of road segments, road segment-level health status, traffic flow, maintenance records, etc. In terms of technical effects, this unit enhances the richness of node representations, enabling each road segment to not only express its geometric features but also contain comprehensive information in multiple dimensions such as its health status and usage. Through this embedding mechanism, the influence of each road segment in the graph structure can be adaptively adjusted according to its characteristics, improving the accuracy and meticulousness of the overall road network health status analysis. This unit also lays the foundation for subsequent spatio-temporal analysis, enabling the dynamic graph to transform from a static topology to a dynamic state, better reflecting the actual operation and health changes of the road network.

[0156] In this embodiment, the edge weight learning unit uses an adaptive mechanism to learn the weights of edges according to the actual relationships and influence degrees between road segments. For example, the connection between two road segments is affected by various factors such as traffic flow, road conditions, traffic accidents, etc., which will change the intensity of their mutual influence. Therefore, the edge weight learning unit models these factors, enabling the edges of the graph to dynamically reflect the actual spatial relationships and interactions between road segments. In terms of technical effects, edge weight learning enables the graph structure to flexibly adapt to the complex dynamic behavior of the road network, providing accurate spatial structure information for spatio-temporal graph convolution and subsequent analysis. The weights of the edges can not only reflect the spatial compactness but also be adaptively adjusted to temporal changes through the learning mechanism, enhancing the dynamic modeling ability of the graph, thereby improving the comprehensiveness and accuracy of the road network health status analysis.

[0157] In this embodiment, the task of the spatio-temporal graph convolution unit is to simultaneously capture the spatial dependencies and temporal dynamic changes of the road network using a spatio-temporal graph convolution network (ST-GCN). In practical applications, the health status of the road network is not only affected by the interactions between road segments in space but also by temporal dynamic changes such as seasonal variations and day-night traffic flow changes. Therefore, the spatio-temporal graph convolution unit can capture the dependencies and change laws of the road network in both spatial and temporal dimensions through convolution operations on spatio-temporal data. This convolution operation enables the network to spread and fuse features across scales, thereby comprehensively representing the state of the road network. In terms of technical effects, spatio-temporal graph convolution enables the model to capture the spatial structure while also flexibly adapting to temporal changes, thus improving the dynamic modeling ability of the road network health status.

[0158] In this embodiment, in the road network health assessment, the influence degrees of different road segments and different time periods are different. Some road segments may show significant health changes during specific time periods, while other road segments may be relatively stable. Through the attention mechanism, it is possible to automatically identify which road segments are more critical in the current analysis task and which time period changes have a greater impact on the assessment of the overall health status. This dynamic weighting method can adaptively adjust the attention of features, improving the accuracy and pertinence of the analysis. In terms of technical effects, the spatio-temporal attention mechanism enhances the flexibility of the network, enabling it to dynamically adjust the priority of features for different analysis tasks. This mechanism not only improves the robustness and generalization ability of the model but also makes the road network health status assessment more refined and accurate.

[0159] In this embodiment, the graph pooling unit reduces the dimension of the graph structure through hierarchical pooling operations, extracts the global features at the road network level, and finally realizes the aggregation of the health status from local to global. The graph pooling operation gradually merges adjacent nodes and edges, reduces the scale of the graph layer by layer, and maintains the main information of the graph structure at each layer. Finally, the pooled graph structure will extract the low-dimensional representation representing the overall road network health status. The core of this operation lies in aggregating node information, enabling the model to better extract global information from local features. In terms of technical effects, the graph pooling unit enables the model to process large-scale road network data and effectively extract global features, avoiding the problem of information loss caused by the overly large scale of the road network. Through the pooling operation, important information can be maintained while reducing the dimension, providing strong support for the final health status assessment and prediction.

[0160] The above components work together in different ways, enhancing the model's modeling ability in the spatial and temporal dimensions. From graph structure construction, node feature embedding to edge weight learning, and then to spatio-temporal graph convolution, spatio-temporal attention, and graph pooling operations, all provide strong technical support in their specific fields, capable of comprehensively and accurately depicting the health status changes of the road network. Finally, through this multi-dimensional and dynamic analysis framework, it is possible to achieve real-time monitoring and prediction of the road network health status, providing an important basis for traffic management and decision-making.

[0161] Furthermore, cross-scale feature propagation and fusion are achieved through the following formula: h r (t)=σ(∑ k∈N(r) W rk ·h k (t)+W t ·h r (t - 1)+∑ s=1 S α s ·H s ), where h r(t) represents the feature vector of node r at time t; σ is the activation function used for non-linearly transforming the result after the convolution operation; N(r) represents the set of neighbor nodes of node r; W rk is the spatial relationship weight between node r and node k; h k (t) represents the feature vector of node k at time t; W t is the weight matrix of temporal convolution, reflecting the influence intensity of the state of node r at time t-1 on the current state; h r (t-1) represents the feature vector of node r at time t-1; S represents the number of scales for cross-scale feature fusion; α s is the weighting coefficient of scale s; H s represents the feature vector of the s-th scale.

[0162] The above cross-scale feature propagation and fusion formula successfully combines the advantages of spatio-temporal graph convolutional networks by introducing mechanisms such as spatial relationship weights, temporal convolution weights, and cross-scale feature weighted fusion, improving the accuracy and flexibility of the model in processing the health status of dynamic road networks. The introduction of the spatial relationship weight W rk and the temporal convolution weight W t enables the model to effectively capture spatial and temporal dependencies, while cross-scale feature fusion enables the network to simultaneously consider the mutual influence between local road segment features and global road network features through multi-scale information aggregation. Ultimately, this cross-scale feature propagation and fusion framework not only improves the performance of the model but also provides a new solution idea for the future health assessment of complex road networks.

[0163] Furthermore, the graph structure is dimensionally reduced through hierarchical pooling operations to extract global features at the road network level, realizing the aggregation of health status from local to global, including the following steps.

[0164] Construct an adjacency matrix based on the physical connection relationship between road segments and define the initial neighborhood range as the local receptive field.

[0165] Within the predefined neighborhood range, design a dedicated pooling function to initially aggregate node features, mapping the state information of adjacent road segments into a unified feature representation.

[0166] By iteratively expanding the receptive field range of the pooling operation, aggregate and abstract the features layer by layer, finally obtaining a global feature representation with significantly reduced dimensions but retaining key semantic information.

[0167] In this embodiment, the hierarchical pooling operation provides the graph neural network with powerful information aggregation and feature extraction capabilities. Through the construction of the adjacency matrix and the definition of the local receptive field, it is possible to focus on the important features of the local area; while the pooling function further simplifies and refines the features by summarizing the node features. During the process of gradually expanding the receptive field of the pooling operation layer by layer, it is not only possible to capture local information more precisely, but also to gradually abstract the global road network features, and finally obtain a low-dimensional global feature representation. Generally speaking, the hierarchical pooling operation improves the efficiency and performance of the model through the feature aggregation process from local to global, enabling it to efficiently process complex road network health status assessment tasks and providing strong support for the optimization and decision-making of intelligent transportation systems.

[0168] Furthermore, the health assessment module includes the following components.

[0169] The real-time status analysis unit uses multi-scale feature representations and the results of dynamic graph analysis to calculate the health score and key performance indicators of the road surface in real time, providing a current status assessment.

[0170] The deterioration trend prediction unit predicts the future change trend of the road surface health status based on time series analysis and machine learning algorithms, and identifies potential deterioration risks.

[0171] The risk warning generation unit automatically generates warning information according to the health status and prediction trend, including the deterioration risk level and the possible influence range, supporting rapid response decision-making.

[0172] The dynamic comparison analysis unit compares historical data and current assessment results to identify abnormal changes or long-term trends, providing data support for diagnosing the causes of deterioration.

[0173] The maintenance requirement assessment unit assesses the maintenance requirements and urgency levels of each region based on the health status and risk assessment results.

[0174] In this embodiment, the core function of the real-time status analysis unit is to calculate the health score and key performance indicators of the road surface in real time. By combining multi-scale feature characterization and the results of dynamic graph analysis, it is able to capture the health status of the road network at different levels and different time scales. This unit first processes the data from different perception layers, combines local features, section features, and network-level global features, and provides a comprehensive health assessment for each section, road network area, and even the entire network. In terms of technical effects, it can not only evaluate the immediate health status of the current section, but also take into account the dynamic changes in the road network structure, such as the impact of traffic flow, weather changes, accidents, etc. on the road surface status. Based on these multi-dimensional information inputs, the real-time status analysis unit can generate high-precision health scores and provide data support for subsequent decision-making. In addition, through dynamic graph analysis, this unit can capture the spatial correlation and time dynamic changes between sections, thus ensuring the timeliness and dynamic adaptability of the health status assessment. Finally, it provides accurate and timely assessment results for road maintenance, traffic management, and decision-making.

[0175] In this embodiment, the deterioration trend prediction unit can not only effectively identify potential health deterioration risks, but also give early warnings of potential problems. For example, the machine learning model can capture the trend of the gradual aggravation of the surface damage of the section and predict the health status at a certain future point in time based on historical data and existing evaluation results. In this way, it can give timely warnings before the problems become serious and provide forward-looking guidance for maintenance work. More importantly, the prediction algorithm can optimize itself and continuously adjust the prediction model according to new data to ensure the accuracy and timeliness of the prediction results.

[0176] In this embodiment, through the comprehensive analysis of the real-time health status and prediction trends, the risk warning generation unit can timely identify the risk points in the road network and provide clear decision-making bases. In terms of technical effects, the risk warning generation unit enables intelligent traffic management to shift from passive response to active warning. Through the fusion and analysis of multi-dimensional data, it can output risk warnings of different levels and generate corresponding countermeasures according to the risk levels, such as temporarily closing sections, repair plans, etc. At the same time, the risk warning generation unit has high flexibility and can adjust the warning criteria according to various factors such as specific section conditions, traffic flow, and climate change to ensure the accuracy and pertinence of the warning information.

[0177] In this embodiment, the dynamic comparison and analysis unit first collects historical health assessment data, compares it with the current health status of the road network, and uses difference analysis techniques to identify potential health anomalies. For example, it is found that a certain section of the road surface has shown an obvious deterioration trend in the past period, and this trend has not appeared in the historical data. In terms of technical effects, the role of this unit is to provide data support for the causes of road surface damage, and reveal potential factors affecting the health status of the road network, such as sudden natural disasters, excessive traffic load, and inadequate maintenance. Through comparative analysis, potential long-term trends can also be discovered. For example, some areas may gradually show obvious health decline due to long-term neglect of maintenance, which provides a basis for the management agency to prioritize maintenance. Ultimately, the dynamic comparison and analysis unit not only improves the accuracy of road surface health management, but also helps maintenance decision-makers discover the root causes of problems.

[0178] In this embodiment, the maintenance requirement assessment unit can achieve precise scheduling of the road network maintenance work, optimize resource allocation, and improve maintenance efficiency. Compared with traditional maintenance methods, this unit can dynamically adjust the maintenance plan to ensure timely response to changes in the road surface health status. In addition, the maintenance requirement assessment unit can also comprehensively evaluate the urgency of maintenance according to factors such as the usage frequency and traffic flow of the road section. For example, some urban arterial roads with high usage frequency may show more serious health problems due to long-term traffic load accumulation and need to be repaired preferentially in a short time. This data-driven maintenance assessment method can greatly improve the efficiency and effect of maintenance work and reduce resource waste.

[0179] In summary, the components of the health assessment module cooperate closely to jointly build a comprehensive and efficient road network health management system. The real-time status analysis unit ensures the accurate assessment of the health status, the deterioration trend prediction unit provides a forward-looking warning for future risks, and the risk warning generation unit can timely provide warning information for decision-making. The dynamic comparison and analysis unit can provide data support for the reasons of deterioration, while the maintenance requirement assessment unit helps to optimize the allocation of maintenance resources. This comprehensive assessment system not only improves the accuracy of road surface health management, but also greatly improves the efficiency of road network maintenance, providing strong data support and decision-making basis for the traffic management department. Through the collaborative work of these components, it is possible to achieve all-round and dynamic management from problem detection, risk prediction to resource optimization scheduling.

[0180] Furthermore, the future change trend of the road surface health status is predicted by the following formula: Y(τ)=μ+∑ g=1 p δ g Y(τ-g)+∑ e=1 q θ eε(τ - e) + ε(τ), where Y(τ) represents the pavement health status score at time τ; μ is a constant term representing the offset of the time series; p is the order of the autoregressive part; δ g represents the influence degree of the health status Y(τ - g) at the g-th moment on the current health status Y(τ); Y(τ - g) represents the pavement health status data at time τ - g, that is, the health status value lagged by g steps; q is the order of the moving average part; θ e represents the influence degree of the error term ε(τ - e) at the e-th moment on the current health status Y(τ); ε(τ - e) represents the error term at time τ - e; ε(τ) represents the error term at time τ.

[0181] Identify potential deterioration risks through the following formula: R(τ) = 1 / [1 + exp(-γ·(Y(τ) - Y threshold ))], where R(τ) represents the pavement deterioration risk index at time τ, which is used to quantify potential deterioration risks, and the larger the value, the higher the risk; Y threshold represents the threshold of the health status, which is used to distinguish between healthy and deteriorated states; γ is the sensitivity parameter of the risk index, which is used to control the influence degree of the difference between the pavement health status Y(τ) and the threshold Y threshold on the risk index R(τ).

[0182] In summary, the autoregressive and moving average combined model makes the prediction of the health status not only have historical dependence but also be able to flexibly respond to changes in environmental factors. At the same time, the deterioration risk identification method provides an intuitive risk assessment tool for managers by dynamically calculating the risk index, which helps them identify potential pavement problems at an early stage and make timely responses. In addition, it can also achieve all-round intelligent management from health status assessment, trend prediction to risk identification, thereby providing scientific decision-making support for pavement maintenance work, reducing maintenance costs, and improving road safety and operation efficiency.

[0183] Furthermore, the decision visualization module includes the following components.

[0184] The data access unit accesses monitoring data, prediction results, and warning information in real time, providing a dynamically updated data source for visualization.

[0185] The map display unit, based on geographic information system technology, intuitively displays the road network health status, disease distribution, and warning information in the form of a heat map or a layered map.

[0186] The chart generation unit displays the pavement health index, deterioration trend, and risk assessment results through line charts or bar charts, facilitating managers to quickly grasp the overall situation.

[0187] Early warning prompt unit, which highlights high-risk sections or sudden diseases in the visualization interface, combines color coding and pop-up reminders to ensure timely transmission of early warning information.

[0188] Decision-making suggestion unit, which automatically generates maintenance decision-making suggestions including regional priority ranking based on health assessment results and maintenance demand assessment, and presents them in the form of structured text or charts.

[0189] Interactive operation unit, which provides user interaction functions and supports managers to view detailed data and adjust decision parameters as needed.

[0190] Report generation unit, which automatically generates a visualization report, summarizes monitoring data, assessment results and decision-making suggestions, and is convenient for archiving and sharing.

[0191] In this embodiment, the core role of the data access unit is to ensure the dynamic update and timeliness of data to support the real-time monitoring and evaluation functions of the decision-making visualization platform. Specifically, the data access unit can summarize and integrate information from different sensors (such as road surface state sensors, meteorological sensors, etc.) and external data sources (such as traffic flow data, weather forecasts, etc.). This unit ensures that decisions can be made in real time according to the latest data, preventing decision-making errors caused by data delay or distortion. For example, if a sudden damage or weather change occurs on a certain section of the road, this unit can promptly transmit this information to the decision-makers to ensure that response suggestions can be provided in the shortest time and reduce the potential impact of emergencies. In addition, this unit can also ensure cross-platform and cross-device data synchronization, ensuring the flexibility and scalability of data access.

[0192] In this embodiment, the core technical effect of the map display unit is to transform complex health status information into an intuitive and easy-to-understand visual display in a graphical way, greatly enhancing the user's perception ability of the road network health status. The heat map uses the depth of color to display the changes in the health status, thus helping managers quickly identify sections with poor health status or greater potential risks, which is convenient for formulating targeted maintenance plans. The layered map can display different information through different layers. For example, one layer shows the health index of the road section, and another layer shows the disease type and location, further enhancing the readability of the data and the accuracy of decision-making. This unit also supports dynamic update and can automatically refresh the map when new data is received, maintaining the real-time and accuracy of the data. In terms of technical effects, this method of geographical information display not only improves the interactivity of the data, but also helps managers quickly grasp the overall situation and make more timely and effective decisions.

[0193] In this embodiment, through the combined display of historical data and predicted data, the line chart can clearly present the trend of the road surface health index changing over time, which helps to identify potential health problems in a timely manner. The bar chart is suitable for displaying the health assessment results and risk assessments of different road sections or regions. By comparing the data of different regions, it can help managers clearly know which road sections or regions have poor health conditions and need to be processed preferentially. In terms of technical effects, the chart generation unit can convert a large amount of complex data information into simple and clear charts, significantly improving the efficiency of data analysis and the accuracy of decision-making. In addition, the chart generation unit supports multiple graphic display formats and personalized customization, meeting the diverse requirements of different management levels and decision-making needs.

[0194] In this embodiment, the core function of the early warning unit is to promptly display high-risk road sections or sudden disease conditions to ensure that managers can quickly respond to potential dangers. For example, when a certain road section has a disease or the road surface health index is lower than a predetermined threshold, an early warning will be automatically issued and prominently reminded through methods such as red highlighting, pop-up windows, or sound prompts to ensure that managers notice key issues in the first place. In terms of technical effects, the early warning unit helps managers distinguish different types of risks through different colors and visual effects, and can automatically generate early warnings based on the set thresholds, improving the response speed and accuracy of the entire system.

[0195] In this embodiment, the core technical effect of the decision-making and recommendation unit is to automatically generate maintenance decision-making recommendations through data analysis and model prediction, reducing manual intervention and improving the scientific nature and efficiency of decision-making. This unit combines multiple factors, such as road surface health conditions, potential risk levels, traffic flow, historical maintenance records, etc., calculates the priorities of different road sections, and generates an operable maintenance plan. The generated decision-making recommendations can be presented in the form of charts or reports to help managers clearly understand the maintenance needs of each region and be able to allocate resources and arrange maintenance according to priorities. In terms of technical effects, this automated decision-making support greatly improves the accuracy and planning of maintenance work, helps to optimize resource allocation, and avoid resource waste. At the same time, the decision-making and recommendation unit can flexibly adjust decision-making recommendations according to real-time updated data, enabling maintenance work to keep up with the changes in the road network health status in a timely manner.

[0196] In this embodiment, the interaction operation unit allows users to interact with the system through the interface, flexibly select the area of interest, time period, or data dimension, and perform customized queries and analyses. For example, the manager can select a certain section of the road surface for detailed viewing, analyze its historical health status, predict future trends, or view relevant warning information. At the same time, the interaction operation unit also allows users to adjust some parameters, such as the warning threshold, maintenance priority criteria, etc., to meet different management requirements. In terms of technical effects, the interaction operation unit improves the flexibility of decision-making and the user experience, enabling the manager to customize the operation process according to their own needs and obtain personalized data analysis results.

[0197] In this embodiment, the report generation unit can generate reports in different formats according to management requirements, including PDF reports, Excel spreadsheets, or PowerPoint presentations, summarize various types of data, and present them clearly. Through the generation of reports, it is convenient to view, print, or share the generated analysis results, providing strong data support for subsequent work or reporting to superiors. In terms of technical effects, the report generation unit greatly improves the efficiency and accuracy of report generation through automated operations, avoiding errors or omissions that may occur in the process of manual report writing. At the same time, the standardized format of the report also improves the clarity and consistency of information transmission, helping to enhance the transparency of the decision-making process.

[0198] In summary, through the collaborative work of each component, the decision visualization module realizes the visual display of functions such as road surface health assessment, early warning, and maintenance suggestions, significantly enhancing the manager's control over the health status of the road network. The data access unit ensures the real-time update and accuracy of data, the map display unit improves the readability of information through graphical means, and the chart generation unit makes data analysis more intuitive and easy to understand. The early warning prompt unit and the decision-making suggestion unit reduce errors and delays in manual operations through automated early warning and decision support, while the interaction operation unit provides a flexible customized query function, and the report generation unit improves the efficiency of information archiving and sharing. Through the collaborative effect of this series of components, the decision visualization module provides efficient and accurate intelligent decision support for the health management of the road network, helping managers make timely and effective maintenance decisions, optimize resource allocation, and improve the safety and economy of road operation.

[0199] As Figure 2 shown, a method for monitoring the health status of asphalt pavements includes the following steps.

[0200] Real-time collect static characteristic data and dynamic monitoring data of the asphalt pavement through a sensor network.

[0201] Perform spatio-temporal alignment processing on the collected heterogeneous data, and use wavelet transform for noise reduction and standardization to construct a multi-modal fusion data set.

[0202] Based on the Feature Pyramid Network, an adaptive fusion of multi-scale features is realized to construct a systematic pavement condition representation system.

[0203] Combined with the spatio-temporal graph convolutional network and the self-attention mechanism, the dynamic change features of the road network are extracted, and the multi-level aggregation and representation of features are realized through the hierarchical graph pooling method.

[0204] Multi-dimensional risk identification is realized based on historical data and real-time monitoring data, and the priority level of maintenance requirements is determined through the quantitative comparative analysis of multi-dimensional factors.

[0205] Based on the geographic information system platform, the visualization of monitoring data, evaluation results and warning information is realized, and intelligent suggestions are provided for maintenance decision-making.

[0206] In this embodiment, the sensor network can achieve all-time and full-coverage dynamic monitoring. The collected data includes not only conventional static feature data (such as temperature, humidity, pavement structure characteristics, etc.), but also dynamic data related to traffic flow (such as load changes, pavement vibrations, etc.), which provides comprehensive data support for comprehensively understanding the pavement health condition. Since the pavement condition changes with factors such as weather and traffic flow, the collection of real-time data can timely reflect these changes, providing higher timeliness and accuracy for subsequent analysis. In addition, the distributed characteristics of the sensor network make the monitoring system highly scalable, and the monitoring points can be added or adjusted at any time according to needs to meet the monitoring requirements of road networks of different scales.

[0207] In this embodiment, in order to achieve effective data fusion, it is necessary to perform spatio-temporal alignment processing on these heterogeneous data. Spatio-temporal alignment mainly unifies the time and space dimensions of the collected data for subsequent analysis and fusion. For example, different types of sensors have different sampling frequencies and timestamps, which requires synchronizing the data through algorithms so that the states of data from different sensors at the same moment and the same position can accurately correspond. In addition, data denoising and standardization are to improve data quality, reduce possible noise interference in the collection process, and ensure the availability of data. As a powerful signal processing method, wavelet transform can effectively perform data denoising. Through wavelet transform, the signal can be decomposed into multiple scales, and the noise at different scales can be removed respectively, thereby retaining the effective information in the original signal. The standardization processing of data ensures the dimensional consistency of different data sources, enabling different types of data to be effectively compared and fused.

[0208] In this embodiment, the adaptive fusion of multi-scale features is achieved based on the Feature Pyramid Network (FPN), which is one of the key steps in the health status monitoring of asphalt pavements. The assessment of the health status of asphalt pavements requires comprehensive consideration of feature information at different scales, and this feature information may involve different temporal and spatial scales. For example, there may be small-scale pavement cracks in the short term, while in the long term, there may be more serious settlement or structural damage. The Feature Pyramid Network can adaptively extract valuable information from features at different scales, effectively fuse the features at different scales, and construct a systematic pavement state representation system. Specifically, through multi-level feature extraction and fusion, FPN can jointly model the pavement damage or health status at different scales and achieve an all-round representation of complex pavement structures. In terms of technical effects, this process can improve the sensitivity and accuracy of the pavement health status monitoring system. Especially in the face of complex and changeable pavement health status, FPN can reasonably extract features and perform weighted fusion at different scale levels, thereby improving the accuracy of health assessment and ensuring that pavement damage can be effectively identified at an early stage.

[0209] In this embodiment, the health status of the road network is not only related to a single section, but there is often spatio-temporal correlation, that is, the health status of a section is affected by factors such as the surrounding environment and traffic flow. The spatio-temporal graph convolutional network can effectively capture this spatio-temporal dependence relationship, and through graph convolution, it can perform information propagation and feature extraction on the spatio-temporal graph structure, thereby exploring the interaction and influence between sections in the road network. At the same time, the self-attention mechanism can automatically adjust the weights according to the importance of the data, strengthen the expression of key features, and suppress the influence of irrelevant features. This technical combination can not only extract the dynamic change features of the pavement health status, but also predict the health status of different regions and different time points according to the characteristics of the spatio-temporal graph network. In terms of technical effects, through the combination of the spatio-temporal graph convolutional network and the self-attention mechanism, the correlation and dynamic change trend between sections in the road network can be more accurately identified, providing more accurate feature input for subsequent health assessment and risk prediction, and improving the prediction ability and real-time response ability of the system.

[0210] In this embodiment, through the comprehensive analysis of historical data and real-time monitoring data, the risk status of different road sections can be comprehensively identified. The multi-dimensional risk identification method quantifies the pavement health by considering multiple factors such as traffic flow, weather changes, load conditions, etc., so as to accurately identify potential high-risk road sections. In terms of technical effects, this step can determine the priority level of maintenance requirements through the quantitative comparison and analysis of different dimensional factors. For example, some road sections may need to be repaired first due to heavy traffic flow and heavy load, while other road sections may only require routine maintenance. Through this precise risk identification, it can provide strong support for maintenance decision-making, ensure that maintenance resources can be reasonably allocated according to the actual needs of road sections, and avoid waste of resources.

[0211] In this embodiment, based on the Geographic Information System (GIS) platform, the complex pavement health status information can be visually presented in the form of a map, enabling managers to clearly see the health status, potential disease distribution, and warning information of different road sections. This geographical information display method allows managers to directly view the specific conditions of road sections on the map for spatial analysis and decision support. In addition, the GIS platform can also combine the system analysis results to provide intelligent maintenance decision-making suggestions for managers. In terms of technical effects, through the visual presentation of the GIS platform, not only can the intuitiveness and usability of information be improved, but also the transparency of data analysis and decision-making can be enhanced. Combining real-time monitoring data and historical analysis results, it can provide accurate maintenance suggestions and warning information for managers, helping them optimize the maintenance plan according to the actual situation and improve the efficiency and economy of pavement maintenance.

[0212] In summary, this asphalt pavement health status monitoring method realizes the full-process monitoring from data collection to health assessment, risk identification, and decision support by combining an advanced sensor network, data processing technology, and deep learning model. The sensor network provides comprehensive and real-time data support, data preprocessing ensures data quality, the combination of the Feature Pyramid Network and the Spatio-Temporal Graph Convolutional Network improves the accuracy and dynamic response ability of the system, risk identification and maintenance requirement assessment ensure the optimal allocation of resources, and the visual display of the GIS platform improves the transparency and efficiency of decision-making. Through this systematic method, asphalt pavement health status monitoring can provide accurate and intelligent maintenance decision support to ensure the safety and reliability of the pavement during its service life cycle.

[0213] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An asphalt pavement health status monitoring system, characterized in that: Includes the following modules: Multi-source data acquisition module, used to obtain road surface status parameters in real time; Data processing module, used for spatiotemporal alignment, quality assessment, exception handling and standardization of collected multi-source data; Multi-scale feature extraction module, which is used to extract features from microscopic damage, road section status and network level, and build a multi-scale feature representation system; Dynamic graph analysis module, used to build a hierarchical dynamic graph structure to achieve unified modeling and analysis of road network health status; Health assessment module, used to achieve real-time assessment of pavement health status, deterioration trend prediction and risk warning; Decision visualization module, which is used to display monitoring data, prediction results and warning information, and automatically generate maintenance decision suggestions based on the evaluation results; The multi-scale feature extraction module includes the following components: Microscopic feature extraction unit, which captures the detailed features of pavement microscopic diseases through high-resolution convolutional layers; The road segment feature extraction unit uses a convolutional layer with a medium receptive field to extract state features at the road segment level; The network-level feature extraction unit uses a global pooling layer to extract macro features at the road network level; Multi-scale feature fusion unit, which integrates microscopic, section-level and network-level features to build a unified multi-scale representation system and achieve a comprehensive description of the road surface status; Scale interaction unit, using feature pyramid structure to promote information interaction between features of different scales; The feature enhancement unit enhances the distinguishability and robustness of features through feature normalization operations to improve the adaptability to complex road conditions; The microscopic, segment and network-level features are integrated to build a unified multi-scale characterization system to achieve a comprehensive characterization of the road surface state, including the following steps: By designing a multi-branch neural network structure, the features at the micro, road segment and network levels are processed respectively, and the importance weights of features at different scales are learned using the attention mechanism, and the features at the three levels are integrated; The fused features are converted into a fixed-dimensional vector representation through mapping, so that it contains both microscopic behavior patterns and mesoscopic road segment states and macroscopic network features, thereby achieving a comprehensive description of the road surface state. The detailed features of pavement microscopic diseases are captured through a high-resolution convolutional layer, including the following steps: 对处理后的路面图像执行卷积操作，生成多个特征图，用公式表示为：Y(i,j)=(I＊K)(i,j)=∑m∑nI(i+m,j+n)·K(m,n)，其中，Y(i,j)表示卷积操作后输出 特征图中位置为(i,j)的像素值；(i,j)表示图像中像素的坐标，其中i是行坐标，j是列坐标；I表示输入的路面图像；K为卷积核；＊表示卷积操作；(I＊K)表示输入图像I与卷 积核K进行卷积运算后的结果，输出的是一个特征图；m和n表示卷积核K的行和列位置，在计算卷积时，用于索引卷积核K的各个元素；I(i+m,j+n)表示输入图像I中的一个像素 ，位于坐标(i+m,j+n)位置；K(m,n)表示卷积核K中的一个元素，位于位置(m,n)； 使用不同大小的卷积核K1,K2,...,Kn，对处理后的图像进行多尺度卷积操作，从而提取多尺度特征，并通过如下公式对多尺度特征进行融合：Ymulti-scale=[(I ＊K1),(I＊K2),...,(I＊Kn)]，其中，Ymulti-scale表示多尺度特征图的融合结果；(I＊K1)表示输入图像I与卷积核K1进行卷积运算后的结果；(I＊K2)表示输入 图像I与卷积核K2进行卷积运算后的结果；(I＊Kn)表示输入图像I与卷积核Kn进行卷积运算后的结果； The ReLU activation function is used to activate the convolution feature map, remove invalid features and strengthen the response of local damage to highlight the surface details of the road surface. The formula is: A(i,j)=ReLU(Y(i,j)), where A(i,j) represents the activation value of the (i,j)th position in the feature map after being processed by the ReLU function; ReLU(·) represents the ReLU activation function; 将激活后的特征图输入分类网络，通过与损伤分类权重矩阵Wdamage和偏置项b的结合，利用Softmax函数得到损伤的分类概率分布：Pdamage=Softmax(Wdamag e·A+b)，其中，Pdamage表示损伤分类的概率分布，是通过神经网络的分类层输出的结果；A表示经过卷积和激活函数处理后的特征图；Softmax(·)表示Softmax函数； The dynamic graph analysis module includes the following components: The graph structure building unit abstracts the road network into a hierarchical dynamic graph structure, where nodes represent road sections and edges represent connection relationships, providing a basic topological structure for modeling the health status of the road network; The node feature embedding unit embeds the output results of the multi-scale feature extraction module into the graph nodes to represent the health status and attribute information of each road segment; The edge weight learning unit learns the weights of the edges between nodes through an adaptive mechanism, dynamically reflecting the spatial correlation and influence intensity between road sections; The spatiotemporal graph convolution unit uses the spatiotemporal graph convolution network to simultaneously capture the spatial dependency and temporal dynamic changes of the road network, achieving cross-scale feature propagation and fusion; The spatiotemporal attention unit introduces an attention mechanism to dynamically assign importance weights to different nodes and time periods; The graph pooling unit reduces the dimension of the graph structure through hierarchical pooling operations, extracts global features at the road network level, and realizes the aggregation of health status from local to global; The graph structure is reduced in dimension through hierarchical pooling operations, global features at the road network level are extracted, and health status aggregation from local to global is achieved, including the following steps: Construct an adjacency matrix based on the physical connection relationship between road segments, and define the initial neighborhood range as the local receptive field; In the predefined neighborhood, a dedicated pooling function is designed to perform preliminary aggregation of node features and map the state information of adjacent road sections into a unified feature representation. By iteratively expanding the receptive field of the pooling operation, the features are aggregated and abstracted layer by layer, and finally a global feature representation with greatly reduced dimensions but retaining key semantic information is obtained; The health assessment module includes the following components: Real-time status analysis unit, which uses multi-scale feature representation and dynamic graph analysis results to calculate the health score and key performance indicators of the road surface in real time; Deterioration trend prediction unit, based on time series analysis and machine learning algorithms, predicts future trends in pavement health and identifies potential deterioration risks; The risk warning generation unit automatically generates warning information based on the health status and predicted trends, including the degradation risk level and possible impact range; Dynamic comparative analysis unit, which compares historical data with current assessment results to identify abnormal changes or long-term trends; The maintenance needs assessment unit assesses the maintenance needs and urgency of each area based on the health status and risk assessment results.

2. The asphalt pavement health status monitoring system according to claim 1, characterized in that: The multi-source data acquisition module includes the following components: Static sensor unit, which continuously monitors the static changes of pavement structure and materials through sensors installed on the pavement, providing basic pavement health data; Dynamic collection equipment unit, which uses mobile equipment to collect road surface status data in real time and obtain dynamic load and road surface deformation information; Environmental monitoring equipment, which collects external environmental data related to the road surface condition; A vibration sensor unit, which uses vibration sensors to capture the vibration of the road surface under different loads; An image acquisition unit uses a high-definition camera to capture road surface images in real time; The equipment operation monitoring unit monitors the operating status of each data acquisition device, detects equipment failures in a timely manner and performs necessary equipment maintenance.

3. The asphalt pavement health status monitoring system according to claim 1, characterized in that: The data processing module includes the following components: The spatiotemporal alignment unit is responsible for spatiotemporal alignment of time series data from different data sources to ensure that all types of data can accurately reflect the road surface status in the same time and space range; Data quality assessment unit, responsible for quality assessment and self-calibration of various types of collected data, identifying data integrity, accuracy and validity; The exception handling unit is responsible for correcting or eliminating abnormal data; Data standardization unit, responsible for converting data from different sources into a unified standard format and dimension; A data fusion unit that combines data from multiple sources into a high-precision comprehensive dataset; The data synchronization and update unit is used to achieve consistency and real-time update of multi-source data at different processing stages.

4. The asphalt pavement health status monitoring system according to claim 1, characterized in that: The decision visualization module includes the following components: Data access unit, which accesses monitoring data, prediction results and warning information in real time, and provides a dynamically updated data source for visualization; The map display unit, based on geographic information system technology, visually displays the health status of the road network, disease distribution and warning information in the form of heat maps or layered maps; A chart generation unit that displays the pavement health index, deterioration trend, and risk assessment results through line charts or bar charts; Early warning unit, which highlights high-risk sections or sudden damage in a visual interface, combining color coding and pop-up reminders; The decision-making recommendation unit automatically generates maintenance decision recommendations including regional priority rankings based on health assessment results and maintenance needs assessment, and presents them in the form of structured text or charts; Interactive operation unit, which provides user interaction functions and supports managers to view detailed data and adjust decision parameters as needed; The report generation unit automatically generates visual reports summarizing monitoring data, evaluation results and decision-making recommendations.

5. A method for monitoring the health status of an asphalt pavement, implemented based on an asphalt pavement health status monitoring system according to any one of claims 1 to 4, characterized in that: The following steps are involved: Collect static characteristic data and dynamic monitoring data of asphalt pavement in real time through sensor network; The collected heterogeneous data are aligned in time and space, and the wavelet transform is used for noise reduction and standardization to construct a multimodal fusion data set; Adaptive fusion of multi-scale features is achieved based on feature pyramid network, and a systematic road condition representation system is constructed; Combine the spatiotemporal graph convolutional network and the self-attention mechanism to extract the dynamic change characteristics of the road network, and realize the multi-level aggregation and representation of the features through the layered graph pooling method; Achieve multi-dimensional risk identification based on historical data and real-time monitoring data, and determine the priority of maintenance needs through quantitative comparative analysis of multi-dimensional factors; Based on the geographic information system platform, the monitoring data, assessment results and early warning information can be visualized, and intelligent suggestions can be provided for maintenance decisions.

Images