A method for urban road risk assessment based on night light and time-series InSAR cooperation

Abstract

The present application relates to the technical field of remote sensing and intelligent transportation, and particularly relates to a city road risk assessment method based on night light and time series InSAR cooperation. The method comprises the following steps: acquiring night light data, synthetic aperture radar time series data, road vector and geological auxiliary data; constructing a load inversion model by preprocessing night light features and combining point of interest features to realize road traffic load grading; extracting road deformation rate and cumulative settlement by using time series InSAR technology; coupling multi-source data to road units by spatial segmentation rules, constructing a multi-factor driving model by using multiple regression and grey correlation analysis to quantify the contribution rate of each factor to road settlement; constructing a risk assessment model based on the contribution rate weighting, combining deformation, load and road importance to output road unit risk level. The present application realizes the cooperative inversion of traffic load and geological environment, and improves the fine degree and attribution analysis ability of city road risk assessment.

Description

Technical Field

[0001] This invention relates to the fields of remote sensing and intelligent transportation technology, and in particular to a method for urban road risk assessment based on the synergy of nighttime lighting and temporal InSAR. Background Technology

[0002] Currently, urban road safety monitoring mainly relies on traffic surveys, manual inspections, and fixed-point monitoring. In terms of road load assessment, traditional methods use traffic flow statistics and axle load weighing, which are highly accurate but depend on fixed stations and manual data collection. This results in limited monitoring range, long update cycles, and high costs, making it difficult to achieve large-scale dynamic monitoring at the urban scale.

[0003] In recent years, nighttime light remote sensing technology has been used to analyze traffic activity. Satellites such as DMSP-OLS, NPP-VIIRS, and SDGSAT-1 can acquire urban nighttime brightness, with SDGSAT-1 achieving a resolution of 10 meters and extracting fine light distribution at the road scale. In terms of surface deformation monitoring, InSAR technology has been widely used for urban subsidence monitoring. PS-InSAR and SBAS-InSAR are suitable for built-up areas and low-coherence areas, respectively, and can achieve millimeter-level subsidence monitoring. Road risk assessment commonly uses index system methods (such as AHP), statistical model methods (regression analysis), and machine learning methods (random forests, neural networks). However, existing technologies still have the following shortcomings: Road load is difficult to acquire dynamically. Traditional surveys rely on fixed stations and lack the ability for large-scale continuous observation. Deformation monitoring and traffic load are disconnected. InSAR can only reflect the location of settlement but cannot explain the cause. It lacks load information support and it is difficult to establish the coupling relationship between load and settlement. Risk assessment lacks a multi-source collaborative mechanism. Traditional methods mostly use single indicators or local data and cannot comprehensively consider multiple factors such as traffic load, settlement rate, geological conditions and underground engineering. There is a lack of a unified multi-source remote sensing collaborative analysis framework. High-precision monitoring is costly. Although corner reflectors and ground-based SAR can improve accuracy, they are expensive to deploy and have limited coverage. They are not suitable for urban-scale surveys and lack a layered monitoring mechanism of screening before fine measurement. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides an urban road risk assessment method based on the synergy of nighttime lights and temporal InSAR, aiming to improve the deficiencies of existing technologies such as insufficient dynamic perception capability of road load, separation between deformation monitoring and load information, imperfect multi-source data synergy mechanism, high cost of high-precision monitoring and lack of hierarchical strategy.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for urban road risk assessment based on the synergy of nighttime light and temporal InSAR includes: Acquire nighttime light data, synthetic aperture radar time-series data, road vector data, point-of-interest data, and geological environment auxiliary data of the area to be monitored; The nighttime light data is preprocessed to extract road light features, and a load inversion model is constructed by combining the road vector data and point of interest data to obtain the load classification results of road traffic. The synthetic aperture radar time series data are sequentially subjected to atmospheric correction, terrain phase removal, and time series InSAR processing to obtain road deformation monitoring results; Based on the road vector data, road units are divided according to preset spatial segmentation rules. Load classification results, deformation monitoring results and geological environment auxiliary data are coupled to the corresponding units. A multi-factor driving model is constructed using multiple regression and grey relational analysis to quantify the contribution rate of each factor to road settlement. Based on the contribution rate, the weights of each factor are determined. Combined with deformation monitoring results, load classification results, and road importance, a risk assessment model is constructed to determine and output the risk level of the road unit.

[0006] Furthermore, the preprocessing steps for the nighttime light data include: Radiometric calibration and atmospheric correction were performed on the raw nighttime light data, and time series normalization was performed in combination with multi-temporal images to eliminate the influence of sensor differences and environmental background noise, thereby obtaining standardized road brightness characteristic values.

[0007] Furthermore, the step of constructing the load inversion model includes: Statistical features of brightness of road units in nighttime light data are extracted, and combined with the urban functional distribution features and road grade information reflected by point of interest data, a mapping model from brightness features to traffic axle load index is established. The importance of roads is determined using the road grade information, and the automatic classification of road traffic loads in the entire area is achieved after training with known samples, which include measured axle load weighing data from traffic survey stations.

[0008] Furthermore, the steps of the time-series InSAR processing include: Register the synthetic aperture radar time series data and construct an interferometric pair network; Candidate points are extracted by phase decoherence suppression, phase unwrapping, and orbital error removal, and the deformed phase and residual phase are separated in the time and spatial domains.

[0009] Furthermore, the acquisition of road deformation monitoring results specifically includes: The deformation monitoring results include at least the deformation rate, cumulative settlement, and differential settlement. The first type of algorithm is used to obtain high-precision point deformation of stable scatterers in urban built-up areas; The second type of algorithm is used to improve the deformation inversion capability in low-coherence regions by constructing a small baseline interferometric network; For key monitoring areas, residual analysis is conducted by introducing auxiliary ground calibration equipment to identify abnormal deformation areas and construct a spatiotemporal evolution map of road deformation.

[0010] Furthermore, the preset spatial segmentation rules specifically include: Based on road vector topology and road centerlines, buffer analysis is performed to divide the road network into computational units with independent spatial attributes. Unstructured multi-source remote sensing data and structured industry management data are uniformly resampled or projected onto the computing unit, and time-domain alignment is performed based on a preset time reference.

[0011] Furthermore, the steps for constructing a multi-factor driven model using multiple regression and grey relational analysis include: Using road traffic load, geological environment auxiliary data, groundwater dynamic data, and underground engineering construction status as independent variables and road settlement as the dependent variable, we initially screened influencing factors through statistical regression analysis, and used grey relational analysis to quantify the coupling strength between each influencing factor and settlement characteristics to determine the contribution rate of each factor.

[0012] Furthermore, the geological environment auxiliary data includes: At least one of the following: groundwater level dynamic monitoring data, soft soil layer thickness distribution data, existing underground engineering spatial distribution data, surface cover type, and land use type; The multi-factor driving model establishes quantitative analysis relationships and, based on the geological environment auxiliary data and load classification results, separates natural geological settlement factors and traffic load-induced settlement factors to achieve attribution analysis of road settlement.

[0013] Furthermore, the steps for constructing the risk assessment model include: Based on the aforementioned contribution rate, the weights of each risk factor are determined using a weighting method that combines subjective and objective assessments. The contribution rate, deformation monitoring results, load classification results, and road importance are used as input variables. A risk level classification model is established using machine learning algorithms to output the comprehensive risk value and corresponding risk level of each road unit.

[0014] Furthermore, it also includes: Deploy ground-based corner reflectors or ground-based radar monitoring equipment in pre-defined deformation-sensitive areas or areas affected by key projects to obtain local deformation observation data with high sampling rates; The local deformation observation data is fused and analyzed with the time-series InSAR processing results. Abnormal subsidence areas are identified through residuals, and the deformation results retrieved from satellite remote sensing are locally corrected and their accuracy verified.

[0015] The present invention has the following beneficial effects: 1. In this invention, road traffic load is inverted by high-resolution nighttime light data, and combined with points of interest and road grade information, which breaks through the limitations of traditional traffic surveys that rely on fixed stations and manual collection, and improves the coverage and update timeliness of load monitoring.

[0016] 2. In this invention, by spatially coupling the load classification results obtained from nighttime light inversion with the deformation monitoring results obtained from time-series InSAR, a multi-factor driving model is constructed using multiple regression and grey relational analysis. This model quantifies the contribution rates of factors such as traffic load, geological environment, groundwater, and underground engineering construction to road settlement, effectively explaining the causes of settlement and overcoming the technical deficiency of single InSAR technology, which can only reflect the location of settlement but cannot explain the causes of settlement.

[0017] 3. In this invention, the weights of each factor are determined by the contribution rate, and a risk assessment model is constructed by combining subjective and objective weighting methods with machine learning algorithms. The risk level and comprehensive risk value of the road unit are output by integrating deformation monitoring results, load classification results and road importance, thereby improving the accuracy of risk assessment and spatial positioning capability.

[0018] 4. In this invention, local high-precision deformation data is acquired by corner reflectors or ground-based radar, and residual correction and accuracy verification are performed on satellite InSAR results. While ensuring the ability to conduct city-scale surveys, the monitoring accuracy of key areas is improved, and the overall deployment cost of high-precision monitoring equipment is reduced. Attached Figure Description

[0019] Figure 1 This is a flowchart of an urban road risk assessment method based on the synergy of nighttime lights and temporal InSAR proposed in this invention; Figure 2 This is a flowchart of the road traffic load inversion process proposed in this invention; Figure 3 This is a flowchart of the time-series InSAR deformation monitoring and correction process proposed in this invention; Figure 4 This is a flowchart of the multi-factor driving model and attribution analysis proposed in this invention; Figure 5 This is a flowchart of the local high-precision monitoring and residual correction proposed in this invention. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Taking Tianjin's road risk assessment as a specific example, this invention provides a method for urban road risk assessment based on the synergy of nighttime lights and temporal InSAR, such as... Figure 1 As shown, it includes the following steps: S100: Acquire nighttime light data, synthetic aperture radar time series data, road vector data, point of interest data, and geological environment auxiliary data of the area to be monitored; Specifically, in this embodiment, the area to be monitored is the main urban area of ​​Tianjin and the coverage area of ​​major traffic arteries. The acquired data includes: Nighttime light data was obtained from nighttime light imagery of the SDGSAT-1 satellite between 2021 and 2026. This satellite carries a high-sensitivity low-light imager with a spatial resolution of 10 meters, capable of capturing nighttime brightness information of urban roads. Data acquisition prioritized clear nighttime imagery with no clouds, low aerosol levels, and no moonlight, with at least one valid data set acquired each month, for a total of approximately 60 image sets.

[0022] Synthetic Aperture Radar (SAR) time-series data was used to acquire C-band SAR images from Sentinel-1A / B satellites from 2017 to 2026. An interferometric wide-swath mode was selected, with a spatial resolution of 5 m × 20 m and a revisit period of 12 days. Bidirectional acquisition, including both ascent and descent, was employed to reduce the impact of geometric distortion, resulting in approximately 260 images.

[0023] The road vector data uses the road centerline vector layer from the Tianjin Municipal Basic Geographic Information Data. This data includes attribute information such as road name, road class, number of lanes, and road width. The road classes are divided into expressways, arterial roads, secondary arterial roads, and local roads.

[0024] Points of Interest (POI) data, using POI data updated in 2024 within Tianjin, includes categories such as restaurants, shopping malls, schools, hospitals, logistics parks, parking lots, and gas stations, used to characterize the distribution of urban functional areas and traffic attraction intensity.

[0025] Geological environment auxiliary data includes multi-year groundwater level dynamic monitoring data in Tianjin, provided by the Tianjin Hydrological and Water Resources Monitoring Center, covering approximately 120 monitoring points and updated monthly; soft soil layer thickness distribution data, derived from Tianjin Engineering Geological Survey Report, at a scale of 1:50,000; spatial distribution data of existing underground engineering projects, covering subway lines, underground pipelines, and deep foundation pits; and data on surface cover type and land use type, obtained based on Gaofen-2 image interpretation.

[0026] All vector data were uniformly converted to the WGS84 coordinate system, and the raster data were uniformly converted to a spatial resolution of 10 meters using the nearest neighbor resampling method, and time-series aligned with January 1, 2021 as the time base.

[0027] By collaboratively acquiring multi-source remote sensing and geographic data and unifying the spatiotemporal benchmark, a consistent data foundation is provided for load inversion, deformation monitoring, and multi-factor analysis, overcoming the technical shortcomings of traditional methods such as single data, inconsistent spatiotemporal benchmarks, and difficulty in achieving large-scale dynamic perception.

[0028] S200: Preprocess the nighttime light data, extract road light features, and combine the road vector data and point of interest data to construct a load inversion model to obtain the load classification results of road traffic; Furthermore, the preprocessing steps for the nighttime light data include: Radiometric calibration and atmospheric correction were performed on the raw nighttime light data, and time series normalization was performed in combination with multi-temporal images to eliminate the influence of sensor differences and environmental background noise, thereby obtaining standardized road brightness characteristic values.

[0029] Specifically, the following preprocessing operations were performed sequentially on the 60 SDGSAT-1 nighttime light images acquired from 2021 to 2026. First, radiometric calibration was performed, converting the raw digital quantization values ​​of the images into top atmospheric radiance. The conversion formula is as follows: ; in, Radiance, unit: ; Image pixel value; The calibration gain coefficients are read from the image metadata. This is the calibration offset coefficient.

[0030] Then, atmospheric correction is performed, using the 6S radiative transfer model to remove the effects of atmospheric molecular scattering and aerosol absorption, yielding the true surface radiance. Input parameters include satellite observation geometry, aerosol optical thickness, and water vapor content. After atmospheric correction, the image brightness values ​​eliminate interference from atmospheric path radiation, more accurately reflecting the nighttime illumination intensity of the surface.

[0031] Finally, time series normalization processing was performed on the multi-temporal images. Land features showing no significant change within Tianjin were selected as pseudo-invariant feature points, including the central area of ​​Yuqiao Reservoir, the runway of Tianjin Binhai International Airport, and the Cultural Center Square. Using the image acquired in June 2023 as the reference image, a histogram matching method was employed to match the brightness distribution of other images to the radiometric baseline of the reference image, eliminating radiometric drift caused by sensor differences and atmospheric conditions at different time phases.

[0032] Furthermore, such as Figure 2 As shown, the steps for constructing the load inversion model include: Statistical features of brightness of road units in nighttime light data are extracted, and combined with the urban functional distribution features and road grade information reflected by point of interest data, a mapping model from brightness features to traffic axle load index is established. The importance of roads is determined using the road grade information, and the automatic classification of road traffic loads in the entire area is achieved after training with known samples, which include measured axle load weighing data from traffic survey stations.

[0033] Specifically, after data preprocessing, the lighting characteristics of each road unit are extracted using road vector data. Spatial statistics are then performed on the standardized nighttime lighting data using the buffer range of the road vector lines to calculate the average brightness. , brightness standard deviation and brightness gradient characteristics, among which , The spatial coordinates of the road unit. The time of data acquisition.

[0034] Simultaneously, POI data was introduced, and kernel density analysis was used to extract urban functional characteristics around the roads, including commercial agglomeration, industrial area proportion, and logistics distribution intensity. To quantitatively characterize the contribution of points of interest to traffic load, point of interest factors were defined. Using the geometric center of the road unit as the center and a buffer zone with a radius of 500 meters, the number of commercial, logistics, and transportation points of interest within the buffer zone is counted, and then weighted and summed according to a preset contribution coefficient. The calculation formula is as follows: ; in, The total number of point of interest categories. Indicates the first Categories of interest, such as business, logistics, and transportation; For the first The contribution coefficient of interest categories is set as follows: 0.5 for commercial categories, 1.0 for logistics categories, and 0.8 for transportation categories. For the first in the buffer The number of points of interest.

[0035] This is the sum of dimensionless contribution coefficients, reflecting the intensity of human activity in this road unit. Road grade information. Read directly from the road vector data attributes and assign values ​​of 4, 3, 2, and 1 for expressways, main roads, secondary roads, and local roads, respectively.

[0036] A mapping relationship is established from nighttime light intensity, POI characteristics, and road grade information to traffic load. This functional relationship can be expressed as: ; Where W represents road traffic load; L represents nighttime light intensity; P represents POI characteristics; and R represents road grade characteristics. This represents the mapping model.

[0037] In this embodiment, a random forest algorithm is preferably used to achieve nonlinear regression from traffic load to characteristics of lights, POIs, and road classification. Training samples are derived from measured axle load weighing data from 35 fixed traffic survey stations in Tianjin. Each survey station provides statistics for one week by vehicle type and axle load level. Based on the axle load conversion factor, different axle load levels are converted to standard axle load equivalents, yielding the measured traffic load for each survey station's road unit. The feature vectors of the road units corresponding to the survey stations. As input, actual measurement The training dataset is constructed using the output labels.

[0038] The hyperparameters of the random forest model were set as follows: 100 decision trees, the number of features randomly selected during node splitting is the square root of the total number of input features (2), and the minimum number of samples per leaf node is 5. Five-fold cross-validation was used to evaluate the model's performance. After training, the model's coefficient of determination on the validation set was calculated. The root mean square error reached 0.86, with a root mean square error of 182 axles per lane per day.

[0039] Using a trained random forest model, the traffic load prediction for each road unit in the entire Tianjin area was obtained. Based on the predicted values, road traffic load is divided into three levels; when Axle load per lane per day, heavy load; when At that time, the middle load; At that time, the load was light.

[0040] At the same time, utilize road grade information The importance of roads is directly determined: expressways are classified as Level I, arterial roads as Level II, secondary arterial roads as Level III, and local roads as Level IV. The load classification results and road importance will serve as input data for subsequent risk assessment models.

[0041] By performing radiometric calibration, atmospheric correction, and multi-temporal normalization preprocessing on nighttime light images, and combining road vector and point of interest (POI) data to extract brightness features, POI weight factors, and road grade information, and using the random forest algorithm to establish a nonlinear mapping model from light features to traffic axle load indicators, the automated classification of urban-scale road traffic load and determination of road importance were realized.

[0042] S300: Perform atmospheric correction, terrain phase removal, and time-series InSAR processing on the synthetic aperture radar time-series data in sequence to obtain the road deformation monitoring results; Furthermore, such as Figure 3 As shown, the steps of the time-series InSAR processing include: Register the synthetic aperture radar time series data and construct an interferometric pair network; Candidate points are extracted by phase decoherence suppression, phase unwrapping, and orbital error removal, and the deformed phase and residual phase are separated in the time and spatial domains.

[0043] Specifically, atmospheric correction, terrain phase removal, and time-series InSAR processing are performed on the acquired synthetic aperture radar time-series data to obtain road deformation monitoring results. Taking Tianjin as an example, 260 Sentinel-1A / B satellite C-band SAR images from 2017 to 2026 were acquired using both ascending and descending orbits.

[0044] The SAR images were registered, with all auxiliary images registered to the main image. The image dated January 1, 2017, was selected as the main image, achieving a registration accuracy within 0.1 pixels. Then, an interferometric pair network was constructed using a small baseline set strategy. The temporal baseline threshold was set to 180 days, and the spatial baseline threshold was set to 200 meters, generating approximately 1500 interferometric pairs.

[0045] Atmospheric correction and topographic phase removal were performed. Atmospheric correction utilized tropospheric delay data provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) and employed a linear regression method to estimate and remove atmospheric phase. Topographic phase removal used a 30-meter spatial resolution SRTM digital elevation model, simulating the topographic phase based on imaging geometry parameters and subtracting it from the interferometric phase.

[0046] Two algorithms, Permanent Scatterer InSAR (PS-InSAR) and Small Baseline Set InSAR (SBAS-InSAR), were used to process the data. For stable scatterers in urban built-up areas (such as buildings, bridges, and streetlights), the PS-InSAR algorithm was employed to obtain millimeter-level deformation by analyzing the long-term phase stability of permanent scatterers. For low-coherence areas such as green belts and vegetated areas, the SBAS-InSAR algorithm was used to improve deformation inversion capabilities by constructing a small baseline interferometric network.

[0047] Phase unwrapping is performed using a minimum cost flow algorithm, and orbital errors are removed using accurate orbital ephemeris. Candidate points are extracted, and the deformed phase and residual phase are separated in the time and spatial domains. Finally, the deformation rate is constructed. and cumulative settlement Database. Deformation rate The calculation uses the average rate formula in the disclosure document: ; in, This embodiment represents the total number of periods of SAR time-series images. ; Indicates the first The cumulative subsidence of the image relative to the reference time, in mm; This represents the time interval between two adjacent image periods.

[0048] Cumulative settlement The calculation formula is: ; in, For a moment The cumulative deformation obtained by deformation phase transformation; The value is 0, which is the deformation variable for the reference time (January 1, 2017).

[0049] Furthermore, the acquisition of road deformation monitoring results specifically includes: The deformation monitoring results include at least the deformation rate, cumulative settlement, and differential settlement. The first type of algorithm is used to obtain high-precision point deformation of stable scatterers in urban built-up areas; The second type of algorithm is used to improve the deformation inversion capability in low-coherence regions by constructing a small baseline interferometric network; For key monitoring areas, residual analysis is conducted by introducing auxiliary ground calibration equipment to identify abnormal deformation areas and construct a spatiotemporal evolution map of road deformation.

[0050] Specifically, deformation monitoring results should include at least the deformation rate, cumulative settlement, and differential settlement. Differential settlement... It is used to characterize the degree of uneven settlement within a road unit, and is defined as the difference between the maximum and minimum cumulative settlement within the same road unit. .

[0051] For key urban roads, bridges, and soft soil sensitive areas in Tianjin, further high-precision local monitoring will be conducted by combining corner reflector or ground-based SAR data. For example, several trihedral corner reflectors will be deployed along Metro Lines 1 and 6 and around bridges along the Haihe River, positioned on the road shoulders, and precise coordinates will be obtained through GNSS measurements. After acquiring local deformation observation data at a high sampling rate (e.g., once every 6 days), residual analysis will be performed. Defined as the difference between the corner reflector observation deformation and the InSAR inversion deformation. .

[0052] Abnormal settlement areas were identified using residuals, and local corrections were performed on the InSAR results. Finally, three databases were constructed: settlement rate, cumulative settlement, and differential settlement, and spatiotemporal evolution maps of deformation for key roads were generated.

[0053] By fusing permanent scatterer and small baseline set algorithms, and combining corner reflectors for local residual correction, collaborative monitoring of urban road deformation rate, cumulative settlement, and differential settlement was achieved, overcoming the limitation of insufficient deformation inversion capability of single InSAR under complex terrain conditions. S400: Based on the road vector data, divide the road into units according to the preset spatial segmentation rules, couple the load classification results, deformation monitoring results and geological environment auxiliary data to the corresponding units, and use multiple regression and grey relational analysis to construct a multi-factor driving model to quantify the contribution rate of each factor to road settlement. Furthermore, the preset spatial segmentation rules specifically include: Based on road vector topology and road centerlines, buffer analysis is performed to divide the road network into computational units with independent spatial attributes. Unstructured multi-source remote sensing data and structured industry management data are uniformly resampled or projected onto the computing unit, and time-domain alignment is performed based on a preset time reference.

[0054] Specifically, road centerlines are extracted based on road vector topology, and buffer analysis is performed by setting corresponding buffer radii according to road grade, dividing the complex road network into computational units with independent spatial attributes. For unstructured multi-source remote sensing data and structured industry management data, spatial projection transformation is used to unify all data to the same geographic coordinate system or projected coordinate system. Through spatial feature aggregation, the average deformation, average brightness, and geological parameters within each road unit are calculated. Simultaneously, based on a preset time reference, data acquired at different frequencies are time-domain aligned to ensure the matching of load characteristics and deformation characteristics over time.

[0055] Furthermore, such as Figure 4 As shown, the steps for constructing a multi-factor driven model using multiple regression and grey relational analysis include: Using road traffic load, geological environment auxiliary data, groundwater dynamic data, and underground engineering construction status as independent variables and road settlement as the dependent variable, we initially screened influencing factors through statistical regression analysis, and used grey relational analysis to quantify the coupling strength between each influencing factor and settlement characteristics to determine the contribution rate of each factor.

[0056] Specifically, a combination of multiple regression analysis and grey relational analysis was used to study the influence of different factors on road settlement. The multi-factor driving model expression is as follows: ; in, The average annual deformation rate of the road unit is expressed in millimeters per year. For geological environmental factors, the thickness of the soft soil layer is measured in meters; For groundwater variation factors, the annual variation of groundwater level is expressed in meters per year; Underground engineering construction status; These are the regression coefficients; It represents the residual.

[0057] In this embodiment, the least squares method is used to fit the multi-factor driven model and screen for significant factors. The fitting result of this embodiment is: , , , Quantify the coupling strength between each factor and sedimentation. First, perform dimensionless transformation. in This refers to the road unit number. The factor number is used; then the correlation coefficient is calculated, and the resolution coefficient is set to 0.5. ; Finally, the correlation degree is calculated. The contribution rate was obtained by normalizing the correlation between each factor. In this embodiment, the correlations calculated were 0.85 for traffic load, 0.62 for soft soil thickness, 0.41 for groundwater level, and 0.53 for construction status. The normalized contribution rates were 45%, 28%, 12%, and 15%, respectively.

[0058] Furthermore, the geological environment auxiliary data includes: At least one of the following: groundwater level dynamic monitoring data, soft soil layer thickness distribution data, existing underground engineering spatial distribution data, surface cover type, and land use type; The multi-factor driving model establishes quantitative analysis relationships and, based on the geological environment auxiliary data and load classification results, separates natural geological settlement factors and traffic load-induced settlement factors to achieve attribution analysis of road settlement.

[0059] Specifically, based on a multi-factor driven model, the natural geological subsidence component is defined. Traffic load-induced settlement component By comparing the ratios of the two factors, an attribution analysis of road settlement is achieved. In this embodiment, the core business district of Tianjin... The data indicates that the settlement is mainly caused by traffic loads; the ratio of soft soil in the Binhai New Area is about 0.6, indicating that natural geological settlement is dominant.

[0060] By dividing the road network into independent calculation units through preset spatial segmentation rules, and constructing a multi-factor driven model based on multi-source data coupling and multivariate regression-grey relational analysis, the contribution rate of factors such as traffic load, geological environment, groundwater and underground engineering construction to road settlement is quantitatively quantified and attributed. This effectively separates natural geological settlement from traffic load-induced settlement, providing a scientific basis for differentiated road maintenance and risk prevention.

[0061] S500: Based on the contribution rate, determine the weight of each factor, combine the deformation monitoring results, load classification results and road importance, construct a risk assessment model, and determine and output the risk level of the road unit.

[0062] Furthermore, the steps for constructing the risk assessment model include: Based on the aforementioned contribution rate, the weights of each risk factor are determined using a weighting method that combines subjective and objective assessments. The contribution rate, deformation monitoring results, load classification results, and road importance are used as input variables. A risk level classification model is established using machine learning algorithms to output the comprehensive risk value and corresponding risk level of each road unit.

[0063] Specifically, based on the contribution rate calculated by the multi-factor driving model, and combined with multi-source monitoring indicators, a quantitative evaluation and classification of road safety risks are completed.

[0064] The procedure for determining the weights of risk factors employs a combined subjective and objective weighting method. Objective weights are directly derived from the contribution rates of each factor to settlement output by the multi-factor driving model, reflecting the actual driving intensity of each physical quantity on pavement deformation. Subjective weights are based on the analytic hierarchy process (AHP), evaluating road importance, traffic function level, and traffic diversion function through the establishment of a road engineering expert knowledge base. The subjective and objective weights are then combined using a combined weighting method to obtain the final set of risk factor weights.

[0065] A risk assessment model was constructed, using deformation monitoring results, load classification results, road importance, and the contribution rates of various driving factors as input variables. The formula for calculating the comprehensive risk value of each road unit is as follows: ; in, For the first The overall risk value of each road unit; The total number of risk factors; For the first Weighting coefficients of risk factors; For the first The road unit in the first Raw observations under risk factors; This is a standardized mapping function.

[0066] For each risk factor, a min-max normalization process is performed, and then the result is entered into the formula to calculate the comprehensive risk value. In this embodiment, the deformation severity factor is a combination of deformation rate, cumulative settlement, and differential settlement, while the traffic load level factor, road importance factor, and geological environment sensitivity factor are normalized respectively.

[0067] Based on the obtained comprehensive risk value, a risk level classification model is established using the random forest algorithm. 400 samples from historical road defect inspection data are used as the training set, and the normalized multidimensional feature vectors are mapped to the risk level space. Random forest hyperparameters: 100 decision trees, 3 random features for node splitting, and a minimum number of samples per leaf node (5). Five-fold cross-validation achieves a classification accuracy of 87%.

[0068] Based on the classification model output, the entire road network is divided into three levels: high-risk, medium-risk, and low-risk, and a thematic map of road risk levels is generated. High-risk units are marked in red, approximately 850 in total, mainly distributed along subway construction lines, heavy-load expressways, and areas with thick soft soil; medium-risk units are marked in yellow, approximately 3200 in total; and low-risk units are marked in green, approximately 7950 in total. The assessment results are simultaneously stored in a risk database, which records road unit numbers, spatial coordinates, comprehensive risk scores, dominant risk factors, and recommended inspection cycles.

[0069] Through a closed-loop assessment mechanism of load, deformation, and risk, dynamic perception and precise classification management of road safety status at the urban scale can be achieved, providing auxiliary decision support for municipal maintenance departments.

[0070] Furthermore, it also includes: Deploy ground-based corner reflectors or ground-based radar monitoring equipment in pre-defined deformation-sensitive areas or areas affected by key projects to obtain local deformation observation data with high sampling rates; The local deformation observation data is fused and analyzed with the time-series InSAR processing results. Abnormal subsidence areas are identified through residuals, and the deformation results retrieved from satellite remote sensing are locally corrected and their accuracy verified.

[0071] Specifically, such as Figure 5 As shown, for the pre-defined deformation-sensitive areas and key project-affected areas in Tianjin, ground corner reflectors or ground-based radar monitoring equipment are further deployed to obtain local deformation observation data with high sampling rates. The satellite InSAR inversion results are then locally corrected and their accuracy verified through residual analysis.

[0072] Deformation-sensitive areas were selected from regions in Tianjin with soft soil thickness greater than 15 meters and annual groundwater level drop exceeding 0.5 meters, including Binhai New Area and Jinnan District. Key project impact areas were selected from the subway construction impact areas, with subway lines 1 and 6, and lines 4 and 8 (currently under construction) as the center, extending 200 meters to both sides as the key monitoring scope.

[0073] Several ground corner reflectors were deployed within the aforementioned area. The corner reflectors employed a trihedral corner reflector structure, were made of aluminum alloy, and had a high-reflectivity coating on their surface. Each corner reflector was fixed to the road shoulder or green belt via a reinforced concrete base, with the base buried 1.5 meters deep. The precise spatial coordinates of the corner reflectors were measured using a GNSS receiver, achieving a horizontal positioning accuracy better than 1 cm and an elevation accuracy better than 2 cm.

[0074] For road sections with high deformation rates and complex spatial distribution, such as intersections where subways pass under main roads, ground-based radar monitoring equipment is used for continuous area monitoring. The ground-based radar is a ground-based synthetic aperture radar system operating in the Ku-band, with a spatial resolution of 0.5 meters, a sampling frequency of once every 10 minutes, and a monitoring range covering 0.5 to 2 kilometers.

[0075] The corner reflector data acquisition frequency is synchronized with the Sentinel-1 satellite revisit cycle, set to once every 12 days, to obtain the line-of-sight deformation of the corner reflector position relative to a fixed reference point, and record it as follows. The units are millimeters. The ground-based radar uses a continuous scanning mode and extracts the deformation time series of each pixel in the monitoring area through permanent reflector technology, with a sampling interval of 10 minutes. In this embodiment, a high-risk section along Tianjin Metro Line 1 is selected as an example. Continuous monitoring for 12 months yielded 30 periods of corner reflector deformation data and approximately 43,200 periods of ground-based radar deformation data.

[0076] Using high-precision local deformation observation data acquired by corner reflectors or ground-based radar as the reference true value, and comparing it with the satellite InSAR inversion results at the same location and time, the formula for calculating the residual is: ; in, The value is the residual, in millimeters. This refers to the cumulative settlement observed by corner reflectors or ground-based radar. This represents the cumulative settlement retrieved from satellite InSAR.

[0077] Statistical analysis was performed on the residual sequences for each monitoring point, and the mean of the residuals was calculated. and standard deviation If the absolute value of the residual at a certain moment is greater than If the point is identified as an abnormal settlement area, then the spatial distribution of the abnormal settlement area is determined using the Kriging interpolation method to generate a residual surface.

[0078] The inverse distance-weighted interpolation method is used to extend the residual surface to surrounding road units, thereby locally correcting the deformation results retrieved from satellite InSAR. The corrected settlement calculation formula is as follows: ; in, This is the residual estimate obtained through spatial interpolation.

[0079] Using uncorrected corner reflectors as independent verification points, the root mean square error of the InSAR results before and after correction is calculated. in To determine the number of verification points, 10 points were used in this embodiment. Calculation results show that the RMSE of the InSAR results before correction was 5.2 mm, and after correction, the RMSE decreased to 2.1 mm, improving the accuracy by approximately 60%. This local correction and accuracy verification process effectively improves the reliability of road settlement monitoring in key areas.

[0080] By deploying corner reflectors or ground-based radar to acquire local high-precision deformation data, and combining it with time-series InSAR for residual analysis and spatial interpolation correction, a hierarchical monitoring mechanism of wide-area screening, local fine measurement, and residual correction is formed, which overcomes the deficiency of insufficient deformation inversion accuracy in key areas by the single InSAR technology.

[0081] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for urban road risk assessment based on the synergy of nighttime light and temporal InSAR, characterized in that, include: Acquire nighttime light data, synthetic aperture radar time-series data, road vector data, point-of-interest data, and geological environment auxiliary data of the area to be monitored; The nighttime light data is preprocessed to extract road light features, and a load inversion model is constructed by combining the road vector data and point of interest data to obtain the load classification results of road traffic. The synthetic aperture radar time series data are sequentially subjected to atmospheric correction, terrain phase removal, and time series InSAR processing to obtain road deformation monitoring results; Based on the road vector data, road units are divided according to preset spatial segmentation rules. Load classification results, deformation monitoring results and geological environment auxiliary data are coupled to the corresponding units. A multi-factor driving model is constructed using multiple regression and grey relational analysis to quantify the contribution rate of each factor to road settlement. Based on the contribution rate, the weights of each factor are determined. Combined with deformation monitoring results, load classification results, and road importance, a risk assessment model is constructed to determine and output the risk level of the road unit.

2. The urban road risk assessment method based on the synergy of nighttime lights and temporal InSAR as described in claim 1, characterized in that, The steps for preprocessing the nighttime light data include: Radiometric calibration and atmospheric correction were performed on the raw nighttime light data, and time series normalization was performed in combination with multi-temporal images to eliminate the influence of sensor differences and environmental background noise, thereby obtaining standardized road brightness characteristic values.

3. The urban road risk assessment method based on the synergy of nighttime lights and temporal InSAR as described in claim 1, characterized in that, The steps for constructing the load inversion model include: Statistical features of brightness of road units in nighttime light data are extracted, and combined with the urban functional distribution features and road grade information reflected by point of interest data, a mapping model from brightness features to traffic axle load index is established. The importance of roads is determined using the road grade information, and the automatic classification of road traffic loads in the entire area is achieved after training with known samples, which include measured axle load weighing data from traffic survey stations.

4. The urban road risk assessment method based on the synergy of nighttime lights and temporal InSAR as described in claim 1, characterized in that, The steps of the time-series InSAR processing include: Register the synthetic aperture radar time series data and construct an interferometric pair network; Candidate points are extracted by phase decoherence suppression, phase unwrapping, and orbital error removal, and the deformed phase and residual phase are separated in the time and spatial domains.

5. The urban road risk assessment method based on the synergy of nighttime lights and temporal InSAR as described in claim 1, characterized in that, The specific results of obtaining road deformation monitoring include: The deformation monitoring results include at least the deformation rate, cumulative settlement, and differential settlement. The first type of algorithm is used to obtain high-precision point deformation of stable scatterers in urban built-up areas; The second type of algorithm is used to improve the deformation inversion capability in low-coherence regions by constructing a small baseline interferometric network; For key monitoring areas, residual analysis is conducted by introducing auxiliary ground calibration equipment to identify abnormal deformation areas and construct a spatiotemporal evolution map of road deformation.

6. The urban road risk assessment method based on the synergy of nighttime lights and temporal InSAR as described in claim 1, characterized in that, The preset spatial segmentation rules specifically include: Based on road vector topology and road centerlines, buffer analysis is performed to divide the road network into computational units with independent spatial attributes. Unstructured multi-source remote sensing data and structured industry management data are uniformly resampled or projected onto the computing unit, and time-domain alignment is performed based on a preset time reference.

7. The urban road risk assessment method based on the synergy of nighttime lights and temporal InSAR as described in claim 1, characterized in that, The steps for constructing a multi-factor driven model using multiple regression and grey relational analysis include: Using road traffic load, geological environment auxiliary data, groundwater dynamic data, and underground engineering construction status as independent variables and road settlement as the dependent variable, we initially screened influencing factors through statistical regression analysis, and used grey relational analysis to quantify the coupling strength between each influencing factor and settlement characteristics to determine the contribution rate of each factor.

8. The urban road risk assessment method based on the synergy of nighttime lights and temporal InSAR as described in claim 7, characterized in that, The geological environment auxiliary data includes: At least one of the following: groundwater level dynamic monitoring data, soft soil layer thickness distribution data, existing underground engineering spatial distribution data, surface cover type, and land use type; The multi-factor driving model establishes quantitative analysis relationships and, based on the geological environment auxiliary data and load classification results, separates natural geological settlement factors and traffic load-induced settlement factors to achieve attribution analysis of road settlement.

9. The urban road risk assessment method based on the synergy of nighttime lights and temporal InSAR as described in claim 1, characterized in that, The steps for constructing the risk assessment model include: Based on the aforementioned contribution rate, the weights of each risk factor are determined using a weighting method that combines subjective and objective assessments. The contribution rate, deformation monitoring results, load classification results, and road importance are used as input variables. A risk level classification model is established using machine learning algorithms to output the comprehensive risk value and corresponding risk level of each road unit.

10. The urban road risk assessment method based on the synergy of nighttime lights and temporal InSAR as described in claim 1, characterized in that, Also includes: Deploy ground-based corner reflectors or ground-based radar monitoring equipment in pre-defined deformation-sensitive areas or areas affected by key projects to obtain local deformation observation data with high sampling rates; The local deformation observation data is fused and analyzed with the time-series InSAR processing results. Abnormal subsidence areas are identified through residuals, and the deformation results retrieved from satellite remote sensing are locally corrected and their accuracy verified.

Images