Multi-spectral imaging evaluation system for aging degree of asphalt pavement and method thereof

Abstract

The present application relates to the field of road engineering, in particular to an asphalt pavement aging degree multispectral imaging evaluation system and method thereof, comprising: a multispectral imaging module collects pavement spectral data; a topological feature extraction module performs multiscale feature decomposition on the spectral data and extracts topological structure features; a spectral line attention calculation module calculates the weight coefficients of different spectral lines and generates a weighted feature vector; a time series feature learning module processes the time series changes of the feature vector and captures the aging evolution law; a life prediction module determines the current aging state based on the state vector, generates multiple possible aging evolution paths, determines the pavement aging grade and the remaining life; and a decision support module generates maintenance decision suggestions according to the aging grade and the remaining life. The innovative combination of multispectral imaging technology and deep learning algorithm realizes high-precision evaluation of the aging degree of asphalt pavement and accurate prediction of the remaining life, significantly improves the scientificity of maintenance decisions and reduces maintenance costs.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of road engineering, in particular to an asphalt pavement aging degree multispectral imaging evaluation system and method thereof, and more particularly to a technical scheme for realizing asphalt pavement aging degree evaluation and residual life prediction by using multispectral imaging technology combined with deep learning. BACKGROUND

[0002] Asphalt pavement is gradually aged by various factors such as ultraviolet radiation, temperature change, moisture erosion and traffic load during use. Pavement aging not only affects driving comfort and safety, but also greatly shortens the service life of the pavement and increases the cost of road maintenance.

[0003] Existing asphalt pavement aging evaluation methods mainly include manual visual inspection, physical sampling analysis and single-band image analysis. Manual visual inspection is highly subjective and has low precision; physical sampling analysis is highly accurate but destructive, low in efficiency and high in cost; single-band image analysis is convenient but limited in information, making it difficult to fully reflect the pavement aging state. In addition, most existing evaluation methods are limited to static judgment of pavement aging degree, lacking dynamic analysis of aging evolution law and accurate residual life prediction capability.

[0004] With the development of multispectral imaging technology and artificial intelligence, it is possible to use rich spectral information combined with deep learning algorithms to evaluate the aging degree of asphalt pavement. However, there is currently a lack of comprehensive solutions that organically combine multispectral imaging, topological feature extraction, time series learning and multi-path prediction. SUMMARY

[0005] The main purpose of the present application is to provide an asphalt pavement aging degree multispectral imaging evaluation system and method thereof, which acquires rich spectral information through multispectral imaging technology, combines innovative technologies such as topological feature extraction, attention mechanism, time series learning and multi-path prediction, and realizes accurate evaluation of asphalt pavement aging degree and accurate prediction of residual life.

[0006] The present application provides an asphalt pavement aging degree multispectral imaging evaluation system, which comprises:

[0007] A multispectral imaging module for acquiring spectral data of asphalt pavement, the spectral data including multiple spectral lines, each spectral line including spectral intensity values of multiple wavelength ranges;

[0008] A topological feature extraction module connected to the multispectral imaging module for multiscale feature decomposition of the spectral data, generating a feature matrix and extracting topological structure features from the feature matrix;

[0009] The spectral line attention calculation module is connected with the topological feature extraction module and is configured to calculate weight coefficients of different spectral lines in the feature matrix, and perform weighted processing on the feature matrix based on the weight coefficients to generate a weighted feature vector.

[0010] The time sequence feature learning module is connected with the spectral line attention calculation module and is configured to process time sequence changes of the weighted feature vector, capture time sequence evolution rules of the road surface aging, and output a state vector.

[0011] The life prediction module is connected with the time sequence feature learning module and is configured to receive the state vector, determine a current aging state in a predefined feature space based on the state vector, generate a plurality of possible aging evolution paths, calculate probability distributions of the paths, and comprehensively determine a road surface aging grade and a remaining life.

[0012] The decision support module is connected with the life prediction module and is configured to generate a maintenance decision suggestion according to the aging grade and the remaining life.

[0013] Preferably, the multispectral imaging module comprises:

[0014] A hyperspectral camera configured to acquire spectral information of the asphalt pavement in a wavelength range of 380-800 nm.

[0015] A scanning mechanism connected with the hyperspectral camera and configured to control the hyperspectral camera to scan along the road surface.

[0016] An image preprocessing unit connected with the hyperspectral camera and configured to remove noise and perform spectral calibration on the acquired spectral data.

[0017] A mobile terminal APP connected with the image preprocessing unit and configured to control the hyperspectral camera and the scanning mechanism, and display the spectral data.

[0018] Preferably, the topological feature extraction module comprises:

[0019] A multiscale decomposition unit configured to decompose the spectral data according to different wavelength intervals to form low-frequency features, medium-frequency features and high-frequency features.

[0020] A sliding window analysis unit connected with the multiscale decomposition unit and configured to extract local features along the wavelength dimension by sliding.

[0021] A topological structure extraction unit connected with the sliding window analysis unit and configured to calculate topological features of the spectral line data to generate a topological feature vector.

[0022] A feature fusion unit is connected with the multi-scale decomposition unit and the topology extraction unit, configured to fuse features of different scales and topology features to generate a complete feature matrix.

[0023] As preferred, the spectral line attention calculation module comprises:

[0024] A frequency domain correlation evaluation unit is configured to analyze the richness of each waveband information to generate a frequency domain correlation weight.

[0025] A discriminability calculation unit is connected with the frequency domain correlation evaluation unit and configured to calculate the saliency of spectral line difference between different aging grades to generate a discriminability weight.

[0026] A time series stability evaluation unit is connected with the discriminability calculation unit and configured to evaluate the stability of spectral line features changing over time to generate a stability weight.

[0027] A weight fusion unit is connected with the frequency domain correlation evaluation unit, the discriminability calculation unit and the time series stability evaluation unit, and configured to integrate the three weights to generate a final weight vector.

[0028] A weighted processing unit is connected with the weight fusion unit and configured to weight the feature matrix according to the weight vector to output a weighted feature vector.

[0029] As preferred, the time series feature learning module comprises:

[0030] A memory-enhanced network unit is configured to process the time series change of the weighted feature vector and filter and store key information through a multi-gate structure.

[0031] A multi-scale memory pool is connected with the memory-enhanced network unit and configured to maintain short-term memory, medium-term memory and long-term memory at the same time.

[0032] A space-time feature interaction unit is connected with the multi-scale memory pool and configured to fuse spatial features and time features.

[0033] A feature pyramid structure is connected with the space-time feature interaction unit and configured to integrate feature information of different scales to generate a final state vector.

[0034] As preferred, the life prediction module comprises:

[0035] A manifold construction unit is configured to map the state vector to a predefined multi-dimensional feature space.

[0036] A multi-path generation unit is connected with the manifold construction unit and configured to generate multiple possible aging evolution paths based on the current state.

[0037] a path evaluation unit, connected with the multi-path generation unit, for calculating the probability weight of each path;

[0038] a degradation level judgment unit, connected with the path evaluation unit, for dividing the pavement degradation level into 0 level, 1 level, 2 level and 3 level, wherein the 3 level represents the most serious pavement degradation level;

[0039] a remaining life calculation unit, connected with the degradation level judgment unit, for calculating the remaining service life based on the degradation level and the path probability.

[0040] As preferred, the life prediction module further comprises:

[0041] an uncertainty quantification unit for analyzing the measurement uncertainty, model uncertainty and prediction uncertainty;

[0042] a risk assessment unit, connected with the uncertainty quantification unit, for dividing the high-risk area and the low-risk area, and providing the decision risk assessment.

[0043] As preferred, the decision support module comprises:

[0044] a maintenance strategy generation unit for matching the optimal maintenance measure according to the degradation level and the remaining life;

[0045] a cost-benefit analysis unit, connected with the maintenance strategy generation unit, for calculating the return on investment ratio of different maintenance schemes;

[0046] an implementation timing recommendation unit, connected with the cost-benefit analysis unit, for recommending the best maintenance timing based on the degradation rate and the budget constraint;

[0047] a feedback optimization unit, connected with the implementation timing recommendation unit, for collecting the maintenance effect data and adjusting the prediction model parameters.

[0048] As preferred, it further comprises:

[0049] a database management module, connected with the topology feature extraction module, the time series feature learning module and the life prediction module respectively, for storing the feature data, state data and prediction results;

[0050] a communication interface module, connected with the decision support module, for realizing the data exchange with external systems, supporting data export and remote access;

[0051] a user interface module, connected with the decision support module, for displaying the evaluation results and maintenance suggestions in the form of charts and reports.

[0052] The asphalt pavement degradation level multispectral imaging evaluation method comprises the following steps:

[0053] Collecting spectral data of the asphalt pavement, the spectral data comprising a plurality of spectral lines, each of the spectral lines comprising spectral intensity values of a plurality of wavelength ranges;

[0054] Performing multi-scale feature decomposition on the spectral data to generate a feature matrix, and extracting topological structure features from the feature matrix;

[0055] Calculating weight coefficients of different spectral lines in the feature matrix, and performing weighted processing on the feature matrix based on the weight coefficients to generate a weighted feature vector;

[0056] Processing time sequence changes of the weighted feature vector to capture time sequence evolution rules of pavement aging, and outputting a state vector;

[0057] Receiving the state vector, determining a current aging state in a predefined feature space based on the state vector, generating a plurality of possible aging evolution paths, calculating probability distributions of each path, and comprehensively determining a pavement aging grade and a remaining service life;

[0058] Generating a maintenance decision suggestion according to the aging grade and the remaining service life.

[0059] The present application has the following beneficial effects:

[0060] 1. Improve evaluation accuracy: rich spectral information in the wavelength range of 380-800 nm is obtained through multi-spectral imaging, and topological feature extraction and attention mechanism are used to capture the small changes of pavement aging, which significantly improves the evaluation accuracy of the aging degree.

[0061] 2. Realize dynamic prediction: the time sequence feature learning module is used to capture the aging evolution rules, and the multi-path prediction framework is combined to realize the technical leap from static evaluation to dynamic prediction, and provide a scientific basis for maintenance decision.

[0062] 3. Reduce maintenance cost: accurately identify the pavement aging degree and predict the remaining life, realize accurate maintenance, avoid over-maintenance or insufficient maintenance, optimize resource allocation, and reduce the maintenance cost by about 20%.

[0063] 4. Extend the service life: early signs of aging are found in time, preventive maintenance is implemented, the service life of the pavement is extended by 15%-25%, and the quality of road service is improved.

[0064] 5. Realize intelligent management: provide a complete pavement aging evaluation and maintenance decision solution, promote the transformation of road maintenance from experience-based decision to data-based decision, and improve the scientific level of maintenance management. BRIEF DESCRIPTION OF DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the present application, the drawings needed in the following embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0066] Figure 1 The overall architecture diagram of the asphalt pavement aging degree multispectral imaging evaluation system of the present application;

[0067] Figure 2 The structural schematic diagram of the multispectral imaging module of the present application;

[0068] Figure 3 The structural schematic diagram of the topological feature extraction module of the present application;

[0069] Figure 4 The structural schematic diagram of the spectral line attention calculation module of the present application;

[0070] Figure 5 The structural schematic diagram of the time sequence feature learning module of the present application;

[0071] Figure 6 The structural schematic diagram of the life prediction module of the present application;

[0072] Figure 7 The structural schematic diagram of the decision support module of the present application;

[0073] Figure 8 The flowchart of the asphalt pavement aging degree multispectral imaging evaluation method of the present application. DETAILED DESCRIPTION

[0074] Please refer to Figure 1 - Figure 8 The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0075] With reference to Figure 1 The asphalt pavement aging degree multispectral imaging evaluation system provided by the present application comprises a multispectral imaging module 1, a topological feature extraction module 2, a spectral line attention calculation module 3, a time sequence feature learning module 4, a life prediction module 5 and a decision support module 6.

[0076] The multispectral imaging module 1 is used to collect the spectral data of the asphalt pavement, which includes a plurality of spectral lines, each spectral line including a plurality of wavelength range spectral intensity values. Preferably, the multispectral imaging module 1 collects spectral data in the wavelength range of 380-800nm, including 400 different spectral lines.

[0077] The topological feature extraction module 2 is connected with the multispectral imaging module 1, and is used to perform multiscale feature decomposition on the spectral data, generate a feature matrix, and extract topological structure features from the feature matrix.

[0078] The spectral line attention calculation module 3 is connected with the topological feature extraction module 2, and is used to calculate weight coefficients of different spectral lines in the feature matrix, and perform weighted processing on the feature matrix based on the weight coefficients to generate a weighted feature vector.

[0079] The time sequence feature learning module 4 is connected with the spectral line attention calculation module 3, and is used to process the time sequence change of the weighted feature vector, capture the time sequence evolution law of the pavement aging, and output a state vector.

[0080] The life prediction module 5 is connected with the time sequence feature learning module 4, and is used to receive the state vector, and determine the current aging state in a predefined feature space based on the state vector, generate a plurality of possible aging evolution paths, calculate the probability distribution of each path, and comprehensively determine the pavement aging grade and the remaining life.

[0081] The decision support module 6 is connected with the life prediction module 5, and is used to generate maintenance decision suggestions according to the aging grade and the remaining life.

[0082] In addition, the system also includes a database management module 7, a communication interface module 8 and a user interface module 9, which are used to support the data management, communication and user interaction functions of the system.

[0083] As shown in Figure 2 The multispectral imaging module 1 includes a hyperspectral camera 11, a scanning mechanism 12, an image preprocessing unit 13 and a mobile terminal APP 14.

[0084] The hyperspectral camera 11 is used to obtain the spectral information of the asphalt pavement in the wavelength range of 380-800nm. In an embodiment of the present application, the resolution of the hyperspectral camera 11 is 2048x1536 pixels, and the sampling interval is 1nm, and 400 different spectral lines can be obtained simultaneously.

[0085] The scanning mechanism 12 is connected with the hyperspectral camera 11, and is used to control the hyperspectral camera 11 to scan along the pavement. Preferably, the scanning mechanism 12 is installed on a special vehicle to ensure the stability and comprehensiveness of the scanning. In an embodiment, the scanning vehicle travels at a speed of 30km / h, and can complete full-coverage scanning of a 10km road in 1 hour.

[0086] The image preprocessing unit 13 is connected with the hyperspectral camera 11, and is used for noise removal and spectral calibration of the collected spectral data. Specifically, the image preprocessing unit 13 calibrates the original spectral data by using a dark pixel correction method, and the correction formula is as follows:

[0087]

[0088] Wherein: is the corrected reflectivity, is the original reflectivity, is the blackbody brightness (i.e. the collected background spectrum), is the brightness of the road surface without blackbody (i.e. the collected reference plate reflection spectrum), is the brightness of the collected road surface.

[0089] The mobile terminal APP 14 is connected with the image preprocessing unit 13, and is used for controlling the hyperspectral camera 11 and the scanning mechanism 12, and displaying the spectral data. The mobile terminal APP 14 includes three main functional modules: camera for controlling the hyperspectral camera 11, taking a photo for starting the hyperspectral camera 11 to collect the road surface image, and a gel plate for displaying the aging spectrum.

[0090] As shown in Figure 3 , the topological feature extraction module 2 includes a multi-scale decomposition unit 21, a sliding window analysis unit 22, a topological structure extraction unit 23 and a feature fusion unit 24.

[0091] The multi-scale decomposition unit 21 is used for decomposing the spectral data according to different wavelength intervals to form low-frequency features, medium-frequency features and high-frequency features. In an embodiment of the present application, the multi-scale decomposition unit 21 divides the spectral range of 380-800nm into three levels:

[0092] Low-frequency features: 400-650nm range, corresponding to material matrix changes;

[0093] Medium-frequency features: 380-420nm and 650-800nm range, corresponding to oxidation degree;

[0094] High-frequency features: specific sensitive waveband, corresponding to microscopic structure damage;

[0095] The sliding window analysis unit 22 is connected with the multi-scale decomposition unit 21, and is used for sliding extraction of local features along the wavelength dimension. Preferably, the sliding window analysis unit 22 scans by using a window with a size of 7-11, and the window step is 2, which can effectively capture the spectral change features in the local wavelength range.

[0096] ​The topology extraction unit 23 is connected with the sliding window analysis unit 22, and is configured to calculate the topological features of the spectral line data and generate a topological feature vector. Specifically, the topology extraction unit 23 calculates the persistent homology groups of different dimensions by constructing a multi-scale simple complex, identifies the main topological structure in the spectral line, generates a persistence diagram, and extracts statistical features to form the topological feature vector.

[0097] The feature fusion unit 24 is connected with the multi-scale decomposition unit 21 and the topology extraction unit 23, and is configured to fuse the features of different scales and the topological features to generate a complete feature matrix. The feature fusion adopts a hierarchical strategy, that is, the horizontal fusion of features of the same level is performed first, and then the vertical fusion of features of different levels is performed. The dimension of the fused feature matrix is m x 200, where m is the number of sampling points, and 200 is the dimension of the features after dimension reduction.

[0098] As shown in FIG. 3, Figure 4 The spectral line attention calculation module 3 includes a frequency domain correlation evaluation unit 31, a discriminability calculation unit 32, a time sequence stability evaluation unit 33, a weight fusion unit 34, and a weighted processing unit 35.

[0099] The frequency domain correlation evaluation unit 31 is configured to analyze the richness of the information of each wave band and generate a frequency domain correlation weight. Specifically, the frequency domain correlation evaluation unit 31 evaluates the information richness by calculating the information entropy of the spectral line in different wave bands, and the weight of the wave band with a higher information entropy value is greater. In an embodiment, the frequency domain correlation weight is normalized so that the sum is 1.

[0100] The discriminability calculation unit 32 is connected with the frequency domain correlation evaluation unit 31, and is configured to calculate the significance of the difference between spectral lines of different aging grades and generate a discriminability weight. Specifically, the discriminability calculation unit 32 uses the ratio of the between-class variance to the within-class variance as the discriminability index, and the weight of the wave band with a higher discriminability is greater.

[0101] The time sequence stability evaluation unit 33 is connected with the discriminability calculation unit 32, and is configured to evaluate the stability of the spectral line features changing over time and generate a stability weight. Specifically, the time sequence stability evaluation unit 33 evaluates the stability by calculating the coefficient of variation of the multiple sampling results, and the weight of the wave band with a smaller coefficient of variation is greater.

[0102] The weight fusion unit 34 is connected with the frequency domain correlation evaluation unit 31, the discriminability calculation unit 32, and the time sequence stability evaluation unit 33, and is configured to integrate the three weights to generate a final weight vector. In an embodiment of the present application, the weight fusion adopts an adaptive weighting method, which automatically adjusts the proportion of the three weights according to the current road surface state and environmental conditions. Generally, the initial proportions of the frequency domain correlation weight, the discriminability weight, and the time sequence stability weight are 3:4:3.

[0103] The weighting processing unit 35 is connected with the weight fusion unit 34, and is used for weighting the feature matrix according to the weight vector, and outputting a weighted feature vector. The weighting processing is realized by matrix multiplication, which ensures that important features are strengthened and unimportant features are suppressed, thereby improving the efficiency and accuracy of subsequent processing.

[0104] As shown in Figure 5 The time sequence feature learning module 4 includes a memory-enhanced network unit 41, a multi-scale memory pool 42, a space-time feature interaction unit 43, and a feature pyramid structure 44.

[0105] The memory-enhanced network unit 41 is used for processing the time sequence change of the weighted feature vector, and screening and storing key information through a multi-gate structure. The memory-enhanced network unit 41 is based on the LSTM architecture, but has a special optimization design, which increases four gate structures, including an information entropy evaluation gate, a long-term dependence gate, a short-term change gate, and a feature fusion gate.

[0106] The multi-scale memory pool 42 is connected with the memory-enhanced network unit 41, and is used for simultaneously maintaining short-term memory, medium-term memory, and long-term memory. Specifically:

[0107] Short-term memory: storing the feature change of the last 10 measurements, capturing the sudden aging features;

[0108] Medium-term memory: storing seasonal change patterns, with a period of 3 months;

[0109] Long-term memory: storing annual change trends, with a period of 1 year;

[0110] The space-time feature interaction unit 43 is connected with the multi-scale memory pool 42, and is used for fusing spatial features and time features. The spatial features mainly reflect the distribution characteristics of the spectrum in the wavelength dimension, and the time features reflect the change law of the spectrum over time. The space-time feature interaction adopts a bidirectional attention mechanism to realize mutual enhancement of the spatial features and the time features.

[0111] The feature pyramid structure 44 is connected with the space-time feature interaction unit 43, and is used for integrating feature information of different scales to generate a final state vector. The feature pyramid structure 44 includes three levels: the bottom layer retains fine spectral features (wavelength granularity of 1 nm), the middle layer extracts regional features (wavelength interval of 10 nm), and the top layer captures full-spectrum features (full 380-800 nm range). The feature pyramid realizes interactive fusion of features of different levels through up-sampling and down-sampling operations, and finally outputs a 128-dimensional state vector.

[0112] As shown in Figure 6 The life prediction module 5 includes a manifold construction unit 51, a multi-path generation unit 52, a path evaluation unit 53, an aging grade judgment unit 54, a remaining life calculation unit 55, an uncertainty quantification unit 56, and a risk assessment unit 57.

[0113] The manifold construction unit 51 is configured to map the state vector into a predefined multi-dimensional feature space. Specifically, the manifold construction unit 51 employs a nonlinear dimensionality reduction technique to map the 128-dimensional state vector into a 5-dimensional Riemannian manifold. On this manifold, each point represents an aging state, and the geodesic distance between points represents the difficulty of state transition.

[0114] The multi-path generation unit 52 is connected to the manifold construction unit 51 and is configured to generate multiple possible aging evolution paths based on the current state. In one embodiment of the present application, the multi-path generation unit 52 identifies 8 typical evolution path patterns based on historical data and generates the corresponding evolution trajectories from the current state.

[0115] The path evaluation unit 53 is connected to the multi-path generation unit 52 and is configured to calculate the probability weight of each path. The path evaluation employs a Bayesian probability model, which takes into account the historical similarity, the current state stability, and external influencing factors (such as climate conditions, traffic load, etc.).

[0116] The aging level judgment unit 54 is connected to the path evaluation unit 53 and is configured to divide the pavement aging degree into 0, 1, 2, and 3 levels, where level 3 represents the most severe pavement aging degree. The aging level judgment is based on the comprehensive score of multiple indicators, including spectral feature similarity, topological structure change, and physical property prediction value. Specifically, when the score is in the 0-25 interval, it is judged to be level 0 (no obvious aging), in the 25-50 interval, it is judged to be level 1 (mild aging), in the 50-75 interval, it is judged to be level 2 (moderate aging), and in the 75-100 interval, it is judged to be level 3 (severe aging).

[0117] The remaining life calculation unit 55 is connected to the aging level judgment unit 54 and is configured to calculate the remaining service life based on the aging level and the path probability. The remaining life calculation employs a weighted average method, which takes into account the life prediction results of each possible path, and its calculation formula is:

[0118]

[0119] wherein: is the predicted remaining life, is the probability weight of the i-th path, is the predicted life along the i-th path, is the total number of paths.

[0120] ​Uncertainty quantification unit 56 is used to analyze measurement uncertainty, model uncertainty, and prediction uncertainty. Measurement uncertainty stems from instrument errors and environmental disturbances, model uncertainty from model structure and parameter selection, and prediction uncertainty from the inherent randomness of the aging process. Uncertainty quantification employs the Monte Carlo method, deriving the statistical distribution of the prediction results through multiple simulations.

[0121] Risk assessment unit 57 is connected to uncertainty quantification unit 56, used to divide high-risk and low-risk areas and provide decision-making risk assessment. Risk assessment uses a risk matrix method, with the horizontal axis representing the probability of occurrence (level 1-5) and the vertical axis representing the severity of impact (level 1-5), generating a 5×5 risk matrix. Areas with a total score greater than 16 are classified as high-risk areas, areas with a total score less than 9 are classified as low-risk areas, and the middle area is classified as a medium-risk area.

[0122] like Figure 7 As shown, the decision support module 6 includes a maintenance strategy generation unit 61, a cost-benefit analysis unit 62, an implementation timing recommendation unit 63, and a feedback optimization unit 64.

[0123] The maintenance strategy generation unit 61 is used to match optimal maintenance measures based on the aging level and remaining lifespan. Specifically, different maintenance strategies are adopted for different aging levels:

[0124] Level 0 (no obvious aging): Routine monitoring, no special measures required;

[0125] Level 1 (Mild Aging): Preventive maintenance, such as sealing layer, micro-surfacing;

[0126] Level 2 (Moderate Aging): Functional repair, such as thin-layer cover;

[0127] Level 3 (Severe Aging): Structural repair, such as milling and repaving;

[0128] The cost-benefit analysis unit 62 is connected to the maintenance strategy generation unit 61 and is used to calculate the return on investment (ROI) of different maintenance schemes. The ROI calculation formula is:

[0129] ,

[0130] Wherein, ROI is the return on investment, B is the benefit of maintenance (the value converted from life extension), and C is the maintenance cost.

[0131] The implementation timing recommendation unit 63 is connected with the cost-benefit analysis unit 62, and is used for recommending the best maintenance timing based on the aging rate and budget constraints. The implementation timing recommendation adopts a multi-objective optimization method to balance technical benefits and economic benefits. In an embodiment of the present application, when the aging rate exceeds a critical value (such as an increase of 0.5 levels in the aging grade per year) and the return on investment ratio is greater than 150%, immediate implementation of maintenance is recommended.

[0132] The feedback optimization unit 64 is connected with the implementation timing recommendation unit 63, and is used for collecting maintenance effect data and adjusting the prediction model parameters. The feedback optimization adopts a closed-loop control concept, regularly compares the prediction results with the actual effects, updates the model parameters through a back propagation algorithm, and realizes the continuous optimization of the model.

[0133] The database management module 7 is connected with the topological feature extraction module 2, the time series feature learning module 4, and the life prediction module 5 respectively, and is used for storing feature data, state data, and prediction results. The database management module 7 adopts a hybrid storage strategy, and stores recent data in a local SQLite database and archives historical data to a cloud PostgreSQL database.

[0134] The database management module 7 includes the following main data tables:

[0135] The spectrum data table stores original spectrum data and preprocessed spectrum data;

[0136] The feature data table stores extracted topological features and weighted features;

[0137] The aging grade table stores aging grade evaluation results of different road segments;

[0138] The life prediction table stores remaining life prediction results and uncertainty analysis;

[0139] The maintenance record table stores historical maintenance measures and their effects;

[0140] The communication interface module 8 is connected with the decision support module 6, and is used for realizing data exchange with external systems, supporting data export and remote access. The communication interface module 8 supports multiple communication protocols, including HTTP / HTTPS (Web service), MQTT (Internet of Things device communication), and WebSocket (real-time data transmission).

[0141] The main interfaces provided by the communication interface module 8 include:

[0142] The data acquisition interface receives field-collected spectrum data;

[0143] The evaluation result query interface provides aging grade and remaining life query functions;

[0144] Maintenance suggestion acquisition interface: acquire the maintenance decision suggestions generated by the system;

[0145] Data export interface: support data export in multiple formats such as CSV, Excel, and PDF;

[0146] The user interface module 9 is connected with the decision support module 6, and is used to display the evaluation results and maintenance suggestions in the form of charts and reports. The user interface module 9 adopts responsive design and supports PC and mobile terminal dual-platform operation.

[0147] The user interface module 9 mainly includes the following functional areas:

[0148] Pavement condition visualization area: display the pavement aging grade distribution in the form of a heat map;

[0149] Spectrum analysis area: display the characteristic spectrum and its comparison with the standard spectrum;

[0150] Life prediction area: display the remaining life prediction and uncertainty analysis in the form of charts;

[0151] Maintenance suggestion area: display the maintenance suggestions generated by the system and cost-benefit analysis;

[0152] As shown in Figure 8 The asphalt pavement aging degree multispectral imaging evaluation method of the present application comprises the following steps:

[0153] Step one, spectral data acquisition:

[0154] The spectral data of the asphalt pavement is collected by using a vehicle-mounted multispectral imaging system. First, the hyperspectral camera is controlled by the camera function of the mobile terminal APP to drive along the pavement at a stable speed of 30 km / h, and the spectral data in the wavelength range of 380-800 nm is collected. During the collection process, 100% pavement coverage is ensured, that is, there are at least 10 sampling points in any 1 square meter area.

[0155] Step two, data preprocessing and feature extraction:

[0156] The collected spectral data is preprocessed, including noise removal, spectral calibration and data standardization. Specifically, the dark pixel correction method is used for spectral calibration to eliminate the influence of ambient light, and then minimum-maximum normalization processing is performed to standardize all spectral data to the interval [0, 1].

[0157] Next, the preprocessed spectral data is subjected to multiscale feature decomposition to generate a feature matrix. The specific decomposition method includes:

[0158] Low-frequency feature extraction: for the range of 400-650 nm, the window size is 21 and the step size is 5;

[0159] Mid-frequency feature extraction: For the range of 380-420nm and 650-800nm, the window size is 11 and the step size is 3.

[0160] High-frequency feature extraction: For the sensitive waveband, the window size is 7 and the step size is 1.

[0161] Then, the topological features of the spectral lines are calculated by the topological structure extraction unit to generate a topological feature vector. Finally, the features of different scales and topological features are fused to generate a complete feature matrix with a dimension of m x 200.

[0162] Step three, spectral line attention calculation:

[0163] The weight coefficients of different spectral lines in the feature matrix are calculated. First, the richness of each waveband information is analyzed by the frequency domain correlation evaluation unit to generate a frequency domain correlation weight. Second, the significance of the differences between spectral lines in different aging grades is calculated by the discriminability calculation unit to generate a discriminability weight. Third, the stability of the spectral line features over time is evaluated by the temporal stability evaluation unit to generate a stability weight. Finally, the three weights are integrated by the weight fusion unit to generate a final weight vector, which is used to weight the feature matrix to output a weighted feature vector.

[0164] Step four, time series feature learning:

[0165] The time series changes of the weighted feature vector are processed to capture the time evolution law of pavement aging. Specifically, the time series data are processed by the memory-enhanced network unit, which maintains short-term, medium-term, and long-term memories simultaneously. The spatial and temporal feature interaction unit integrates spatial and temporal features, and the feature pyramid structure integrates information of different scales to finally output a 128-dimensional state vector.

[0166] Step five, aging assessment and life prediction:

[0167] The current aging state is determined based on the state vector and the future evolution is predicted. First, the state vector is mapped to a 5-dimensional Riemannian manifold by the manifold construction unit, and 8 possible evolution paths are generated by the multi-path generation unit. The path evaluation unit calculates the probability weight of each path. Then, the aging level judgment unit divides the pavement aging degree into 0-3 levels, and the remaining service life calculation unit calculates the remaining service life based on the aging level and path probability. At the same time, the uncertainty quantification unit and the risk assessment unit analyze the reliability and decision risk of the prediction results.

[0168] Step six, maintenance decision generation:

[0169] Maintenance decision recommendations are generated according to the aging grade and remaining service life. The maintenance strategy generation unit matches the optimal maintenance measures according to the aging grade, the cost-benefit analysis unit calculates the investment return ratio of each scheme, and the implementation opportunity recommendation unit determines the best maintenance opportunity. Finally, the feedback optimization unit collects actual maintenance effect data, optimizes the prediction model and decision strategy.

[0170] This embodiment is applied to the aging degree evaluation of expressway asphalt pavement. The system uses a vehicle-mounted hyperspectral camera to scan along the G95 Capital Expressway under sunny conditions. The resolution of the hyperspectral camera is 2048x1536, the wavelength range is 200-2500nm, and the actual analysis band used is 380-800nm.

[0171] The scanning vehicle travels at a speed of 30km / h, and a complete set of spectral data is collected every 10 meters of road section. After the raw spectral data is corrected and standardized, it is input into the topological feature extraction module. Multi-scale decomposition uses a three-layer structure, targeting low, medium and high frequency features respectively. Topological feature extraction is based on persistent homology theory, which constructs simple complexes and calculates persistent homology groups to generate feature vectors.

[0172] The spectral line attention calculation module analyzes the information richness, discriminability and stability of each band to generate a weight vector. In this example, the 450-500nm band (corresponding to the oxidation degree of asphalt) and the 650-700nm band (corresponding to surface cracks) obtained higher weights (0.25 and 0.22 respectively).

[0173] After time series feature learning and life prediction, the system evaluates the pavement aging degree of the K50+000 to K55+000 section of the G95 Capital Expressway as follows: K50+000 to K52+000 is grade 1 (mild aging), K52+000 to K53+500 is grade 2 (moderate aging), and K53+500 to K55+000 is grade 0 (no obvious aging). For the 2-grade aging section, the system predicts the remaining service life to be 2.5±0.3 years, and recommends functional repair during the next maintenance cycle, with an investment return ratio of about 210%.

[0174] This embodiment is applied to the aging evolution analysis of urban roads. The system has been regularly scanning the main urban road for 6 consecutive months, once a month, forming a complete time series data set.

[0175] The topological feature extraction module generates a feature matrix for each scan data. The spectral line attention module dynamically adjusts the weight coefficients based on 6 months of data, especially strengthening the weight of bands sensitive to seasonal changes. The time series feature learning module processes 6 months of time series data through a memory-enhanced LSTM network to capture the aging evolution pattern.

[0176] The analysis results show that the city trunk road significantly accelerated the aging rate during March-April, and the aging level rose from level 1 to level 2. Through multi-path prediction analysis, the system identified two main evolution paths: path A (probability 65%) predicted that the aging level would remain at level 2 before the end of August; path B (probability 35%) predicted that the aging level might rise to level 3 in September under the condition of continuous high temperature and heavy traffic.

[0177] Based on this prediction, the decision support module suggests preventive maintenance in early July to prevent further aging. The specific suggestion is to use micro-surfacing technology, which is expected to have a return on investment ratio of 180%, and can extend the remaining service life from the predicted 3.2 years to 5.8 years.

[0178] This embodiment is applied to asphalt pavement life prediction under complex climate and traffic conditions. The test area is a connecting line in a coastal city, which is affected by the marine environment and has large traffic volume changes, with complex aging conditions.

[0179] The system performs four complete scans within a year, in winter, spring, summer, and autumn. The topological feature extraction module extracts stable features and change features from the data of the four seasons. The spectral attention module dynamically adjusts the weights according to environmental factors, especially strengthening the weights of salt erosion sensitive bands.

[0180] The time series feature learning module constructs a complete model of four seasons change, and combines with historical meteorological data and traffic data. The life prediction module generates 12 possible evolution paths, covering different climate change and traffic growth scenarios. Through multi-scenario weighted analysis, the system predicts the remaining life of each part of the road section to be in the range of 1.8-4.5 years, and gives the 95% confidence interval.

[0181] The decision support module, based on the prediction results, suggests a segmented maintenance strategy: the road sections with aging level 3 (about 15%) are immediately repaired structurally; the road sections with aging level 2 (about 40%) are functionally repaired within half a year; the road sections with aging level 1 and 0 (about 45%) continue to be monitored. This maintenance strategy is expected to save 25% of the maintenance cost while ensuring the road service level does not decrease.

[0182] The asphalt pavement aging degree multispectral imaging evaluation system and method provided by the present application realizes accurate evaluation of asphalt pavement aging degree and accurate prediction of remaining life through deep integration of multispectral imaging technology and advanced algorithms, and has the following advantages:

[0183] 1. Multi-dimensional data acquisition: multispectral imaging technology is used to obtain rich spectral information in the wavelength range of 380-800 nm, forming a complete pavement aging feature spectrum.

[0184] 2. High-precision feature extraction: Extract key features from massive spectral data through topological feature extraction and attention mechanism, improve evaluation accuracy.

[0185] 3. Dynamic evolution prediction: Use time series feature learning and multi-path prediction to realize dynamic analysis and accurate prediction of pavement aging process.

[0186] 4. Scientific decision support: Based on accurate aging assessment and life prediction, provide scientific maintenance decision suggestions and optimize resource allocation.

[0187] The present application is suitable for aging assessment and maintenance management of various asphalt pavements such as expressways, urban roads and airport runways, which can significantly improve the scientificity and accuracy of maintenance decisions, reduce maintenance costs, prolong the service life of pavements and has broad application prospects.

[0188] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A multispectral imaging assessment system for the aging degree of asphalt pavement, characterized in that, include: A multispectral imaging module is used to acquire spectral data of asphalt pavement. The spectral data includes multiple spectral lines, and each spectral line includes spectral intensity values ​​for multiple wavelength ranges. A topological feature extraction module, connected to the multispectral imaging module, is used to perform multi-scale feature decomposition on the spectral data, generate a feature matrix, and extract topological structure features from the feature matrix. The spectral line attention calculation module, connected to the topological feature extraction module, is used to calculate the weight coefficients of different spectral lines in the feature matrix, and to perform weighted processing on the feature matrix based on the weight coefficients to generate a weighted feature vector. The temporal feature learning module, connected to the spectral line attention calculation module, is used to process the temporal series changes of the weighted feature vector, capture the temporal evolution law of road surface aging, and output the state vector. The life prediction module, connected to the time-series feature learning module, is used to receive the state vector, determine the current aging state in a predefined feature space based on the state vector, generate multiple possible aging evolution paths, calculate the probability distribution of each path, and comprehensively determine the road surface aging level and remaining life. A decision support module, connected to the lifespan prediction module, is used to generate maintenance decision recommendations based on the aging level and remaining lifespan. The topology feature extraction module includes: A multi-scale decomposition unit is used to decompose the spectral data according to different wavelength ranges to form low-frequency features, mid-frequency features and high-frequency features; A sliding window analysis unit, connected to the multi-scale decomposition unit, is used to extract local features by sliding along the wavelength dimension; The topology extraction unit, connected to the sliding window analysis unit, is used to calculate the topological features of the spectral line data and generate a topological feature vector. The feature fusion unit, connected to the multi-scale decomposition unit and the topology extraction unit, is used to fuse features and topological features of different scales to generate a complete feature matrix. The spectral line attention calculation module includes: The frequency domain correlation evaluation unit is used to analyze the richness of information in each band and generate frequency domain correlation weights. The discrimination calculation unit, connected to the frequency domain correlation evaluation unit, is used to calculate the significance of spectral line differences between different aging levels and generate discrimination weights. A time-series stability evaluation unit, connected to the discrimination calculation unit, is used to evaluate the stability of spectral features over time and generate stability weights. The weight fusion unit, connected to the frequency domain correlation evaluation unit, the discrimination calculation unit, and the time series stability evaluation unit, is used to integrate the three weights to generate the final weight vector. A weighted processing unit, connected to the weight fusion unit, is used to weight the feature matrix according to the weight vector and output a weighted feature vector; The temporal feature learning module includes: A memory-enhanced network unit is used to process the temporal changes of the weighted feature vector and to filter and store key information through a multi-gating structure; A multi-scale memory pool, connected to the memory-enhancing network unit, is used to simultaneously maintain short-term memory, medium-term memory, and long-term memory; The spatiotemporal feature interaction unit is connected to the multi-scale memory pool and is used to fuse spatial features and temporal features; The feature pyramid structure, connected to the spatiotemporal feature interaction unit, is used to integrate feature information at different scales to generate the final state vector.

2. The multispectral imaging assessment system for asphalt pavement aging degree according to claim 1, characterized in that, The multispectral imaging module includes: A hyperspectral camera is used to acquire spectral information of asphalt pavement in the wavelength range of 380-800nm. A scanning mechanism, connected to the hyperspectral camera, is used to control the hyperspectral camera to scan along the road surface; An image preprocessing unit, connected to the hyperspectral camera, is used to perform noise removal and spectral calibration on the acquired spectral data; A mobile terminal APP, connected to the image preprocessing unit, is used to control the hyperspectral camera and the scanning mechanism, and to display the spectral data.

3. The multispectral imaging assessment system for asphalt pavement aging degree according to claim 1, characterized in that, The lifetime prediction module includes: Manifold building units are used to map the state vector to a predefined multidimensional feature space; A multi-path generation unit, connected to the manifold building unit, is used to generate multiple possible aging evolution paths based on the current state; A path evaluation unit, connected to the multi-path generation unit, is used to calculate the probability weight of each path; An aging level determination unit, connected to the path evaluation unit, is used to classify the road surface aging degree into level 0, level 1, level 2 and level 3, where level 3 represents the most severe road surface aging degree. The remaining lifetime calculation unit is connected to the aging level determination unit and is used to calculate the remaining lifetime based on the aging level and path probability.

4. The multispectral imaging assessment system for asphalt pavement aging degree according to claim 3, characterized in that, The lifetime prediction module also includes: Uncertainty quantification unit, used to analyze measurement uncertainty, model uncertainty, and prediction uncertainty; The risk assessment unit, connected to the uncertainty quantification unit, is used to divide high-risk areas and low-risk areas, and to provide decision-making risk assessment.

5. The multispectral imaging assessment system for asphalt pavement aging according to claim 1, characterized in that, The decision support module includes: A maintenance strategy generation unit is used to match the optimal maintenance measures based on the aging level and remaining lifespan. The cost-benefit analysis unit, connected to the maintenance strategy generation unit, is used to calculate the return on investment ratio of different maintenance schemes; The timing recommendation unit is connected to the cost-benefit analysis unit and is used to recommend the best maintenance timing based on aging rate and budget constraints. The feedback optimization unit, connected to the implementation timing recommendation unit, is used to collect maintenance effect data and adjust the prediction model parameters.

6. The multispectral imaging assessment system for asphalt pavement aging degree according to claim 1, characterized in that, Also includes: The database management module is connected to the topology feature extraction module, the temporal feature learning module and the lifetime prediction module respectively, and is used to store feature data, state data and prediction results; The communication interface module, connected to the decision support module, is used to realize data exchange with external systems and supports data export and remote access. The user interface module, connected to the decision support module, is used to display assessment results and maintenance recommendations in the form of charts and reports.

7. A multispectral imaging assessment method for the aging degree of asphalt pavement, employing the multispectral imaging assessment system for the aging degree of asphalt pavement as described in any one of claims 1-6, characterized in that, Includes the following steps: Collect spectral data of asphalt pavement, the spectral data including multiple spectral lines, each spectral line including spectral intensity values ​​of multiple wavelength ranges; The spectral data is subjected to multi-scale feature decomposition to generate a feature matrix, and topological features are extracted from the feature matrix. Calculate the weight coefficients of different spectral lines in the feature matrix, and perform weighted processing on the feature matrix based on the weight coefficients to generate a weighted feature vector; The time series changes of the weighted feature vector are processed to capture the temporal evolution of road surface aging and output a state vector. The state vector is received, and the current aging state is determined in a predefined feature space based on the state vector. Multiple possible aging evolution paths are generated, the probability distribution of each path is calculated, and the pavement aging level and remaining life are determined comprehensively. Based on the aging level and remaining lifespan, maintenance decision recommendations are generated.

Images