Road area environment safety evaluation method and system in highway engineering construction period

Abstract

The invention provides a safety evaluation method and system for a road area environment in a highway engineering construction period. The method comprises the steps of obtaining multi-source sensing data of the road area environment in the highway engineering construction period; inputting the multi-source sensing data into a multi-scale time sensing coding module to obtain sequence features of a macro-scale satellite and a micro-scale unmanned aerial vehicle; the sequence features are input to a multi-modal fusion Transform module, and multi-scale fourth features are obtained; inputting the multi-scale fourth feature into a change detection module to obtain a change labeling graph and a first change event log; and based on the change labeling graph and the first change event log, constructing a risk early warning module to obtain a second change event log. Through multi-source data fusion of satellite remote sensing and unmanned aerial vehicle images, accurate perception and dynamic monitoring of the road area environment in the highway engineering construction period are realized, the accuracy and real-time response capability of environmental safety assessment are improved, and the method is suitable for the field of highway engineering construction and construction safety management.

Description

Technical Field

[0001] The present invention relates to the field of highway engineering environmental detection, and particularly to a method and system for evaluating the safety of the road area environment during the construction period of highway engineering. Background Art

[0002] With the continuous improvement of the requirements for highway engineering construction and environmental safety supervision, the traditional environmental safety evaluation method that relies on manual inspections and single perception means has been difficult to meet the needs of accurately perceiving and intelligently judging the dynamic changes of the road area environment in complex construction scenarios. How to construct a multi-scale road area environment perception and evaluation mechanism based on digital technology that covers the entire construction cycle and takes into account both macro trends and local details has become a key issue in ensuring the management efficiency of the road area environment safety during the construction period of highway engineering.

[0003] The Chinese patent application with the publication number CN114509051A provides a method for measuring and setting out by unmanned aerial vehicle in road engineering. Although it improves the surveying and mapping efficiency and accuracy, its application scenario is limited to static information extraction, lacking continuous monitoring and cross-scale analysis of the changes in the road area environment during the construction period of highway engineering, and only relying on a single microscopic perspective of the unmanned aerial vehicle without introducing macro dimensions such as satellite remote sensing data, resulting in the problem of limited safety vision.

[0004] The Chinese patent application with the publication number CN115204676A proposes a method for assessing natural disaster risks for highways. Although it enhances the data integration ability and the construction of the evaluation system, improving the comprehensiveness and accuracy of the evaluation, this method mainly focuses on static disaster scenarios, lacking the ability to continuously perceive environmental changes during the construction period and a dynamic risk warning mechanism, and it is difficult to meet the real-time safety supervision needs in construction scenarios.

[0005] However, the current technology still faces many challenges. On the one hand, there is a lack of spatio-temporal modeling ability to integrate multi-source data, making it difficult to accurately identify key change events in the process of environmental evolution; on the other hand, there is a lack of an evaluation system for sustainable monitoring and intelligent warning, making it difficult to respond promptly and effectively intervene in sudden environmental risks. Summary of the Invention

[0006] To overcome the above-mentioned defects of the prior art, the present invention provides a method and system for road area environmental safety assessment during the construction period of highway engineering. The method constructs a macro-micro collaborative multi-source information perception system by integrating satellite remote sensing and UAV images, breaking through the vision limitation of traditional single perception means; in terms of environmental state expression, a dual-channel time perception module is adopted to encode the time-series data information at different scales, extract the macro-evolution trend in the satellite sequence data at the macro scale and the local change details in the UAV sequence data at the micro scale, and realize the deep integration and spatio-temporal unified expression of heterogeneous features through the multi-modal fusion Transformer module MFTM, improving the accuracy of environmental state expression; in terms of change detection, the change detection module CDM is combined to achieve high-precision identification of environmental evolution during the construction period, and the risk warning module RWM outputs a safety level assessment based on the change event map, effectively supporting the rapid response to sudden risks and the intelligent management of the construction process; in short, the present invention constructs an environmental safety assessment system covering the whole construction process, taking into account both macro trends and local details, and having the capabilities of dynamic perception and intelligent judgment, solving the problems of narrow data dimension, rough evaluation mechanism and lagging response in the existing methods, and can be widely used for road area environmental monitoring and safety management in complex highway engineering construction scenarios in practical applications.

[0007] To achieve the above object, the present invention provides the following technical solutions: A method for road area environmental safety assessment during the construction period of highway engineering, comprising: Obtaining multi-source perception data of the road area environment during the construction period of highway engineering; Inputting the multi-source perception data of the road area environment during the construction period of highway engineering into the multi-scale time perception encoding module TAFM to obtain macro-scale satellite sequence features and micro-scale UAV sequence features; Inputting the macro-scale satellite sequence features and micro-scale UAV sequence features into the multi-modal fusion Transformer module MFTM to obtain multi-scale fourth features of the road area environment during the construction period of highway engineering; Inputting the multi-scale fourth features of the road area environment during the construction period of highway engineering into the change detection module CDM to obtain a change annotation map of the road area environment during the construction period of highway engineering and a first change event log of the road area environment during the construction period of highway engineering; Based on the change annotation map of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering, constructing a risk warning module RWM of the road area environment during the construction period of highway engineering to obtain a second change event log of the road area environment during the construction period of highway engineering.

[0008] Further, the obtaining of the multi-source perception data of the road area environment during the construction period of highway engineering includes: The multi-source perception data of the road area environment during the construction period of the highway project includes satellite sequence data at the macro scale of the road area environment during the construction period of the highway project and UAV sequence data at the micro scale of the road area environment during the construction period of the highway project; The satellite sequence data at the macro scale of the road area environment during the construction period of the highway project is obtained by satellite remote sensing equipment; The UAV sequence data at the micro scale of the road area environment during the construction period of the highway project is obtained by UAV-borne equipment.

[0009] Furthermore, the input of the multi-source perception data of the road area environment during the construction period of the highway project into the multi-scale time perception coding module TAFM to obtain the satellite sequence features at the macro scale and the UAV sequence features at the micro scale includes: The multi-scale time perception coding module TAFM adopts a dual-channel design, including a satellite time encoder at the macro scale and a UAV time encoder at the micro scale; Input the satellite sequence data at the macro scale of the road area environment during the construction period of the highway project into the satellite time encoder at the macro scale to obtain the satellite sequence features at the macro scale; Input the UAV sequence data at the micro scale of the road area environment during the construction period of the highway project into the UAV time encoder at the micro scale to obtain the UAV sequence features at the micro scale.

[0010] Furthermore, the input of the satellite sequence features at the macro scale and the UAV sequence features at the micro scale into the multi-modal fusion Transformer module MFTM to obtain the multi-scale fourth features of the road area environment during the construction period of the highway project includes: The multi-modal fusion Transformer module MFTM includes a dynamic modal gating module DMGM and a multi-scale attention fusion module CSAFM; Input the satellite feature sequence at the macro scale and the UAV feature sequence at the micro scale into the dynamic modal gating module DMGM to autonomously learn the weights of the satellite feature sequence at the macro scale and the UAV feature sequence at the micro scale, and perform dynamic weight fusion on the weights of the satellite feature sequence at the macro scale and the UAV feature sequence at the micro scale to obtain the multi-scale first features of the road area environment during the construction period of the highway project; Based on the multi-scale first features of the road area environment during the construction period of the highway project, introduce learnable embedding vectors to obtain the multi-scale second features of the road area environment during the construction period of the highway project; Input the multi-scale second features of the road area environment during the construction period of the highway project into the multi-scale attention fusion module CSAFM to obtain the multi-scale third features of the road area environment during the construction period of the highway project; Perform global average pooling GAP operation on the multi-scale third features of the road area environment during the construction period of the highway project to obtain the multi-scale fourth features of the road area environment during the construction period of the highway project.

[0011] Furthermore, inputting the multi-scale fourth feature of the road area environment during the construction period of the highway project into the change detection module CDM to obtain the change annotation atlas of the road area environment during the construction period of the highway project and the first change event log of the road area environment during the construction period of the highway project includes: The change detection module CDM includes a multi-scale feature difference amount encoding module TSDEM and a change decoder CD; Input the multi-scale fourth feature of the road area environment during the construction period of the highway project into the multi-scale feature difference amount encoding module TSDEM to obtain the multi-scale feature difference amount of the road area environment during the construction period of the highway project; The multi-scale feature difference amount encoding module TSDEM adopts a sliding window mechanism SW and a semantic residual mechanism SR; Furthermore, the method for obtaining the multi-scale feature difference amount of the road area environment during the construction period of the highway project includes: Adopt the sliding window mechanism SW for the multi-scale fourth feature of the road area environment during the construction period of the highway project to obtain the feature difference amount of the road area environment during the construction period of the highway project; Adopt the semantic residual mechanism SR for the multi-scale fourth feature of the road area environment during the construction period of the highway project to obtain the high-order feature difference amount of the road area environment during the construction period of the highway project; Based on the multi-scale fourth feature of the road area environment during the construction period of the highway project, the feature difference amount of the road area environment during the construction period of the highway project, and the high-order feature difference amount of the road area environment during the construction period of the highway project, obtain the multi-scale feature difference amount of the road area environment during the construction period of the highway project.

[0012] Input the multi-scale feature difference amount of the road area environment during the construction period of the highway project into the change decoder CD to obtain the final change type of the road area environment during the construction period of the highway project; The change decoder CD includes two layers of Transformer decoders and one layer of fully connected layer FC; Furthermore, the method for obtaining the final change type of the road area environment during the construction period of the highway project includes: Input the multi-scale feature difference amount of the road area environment during the construction period of the highway project into the two layers of Transformer decoders of the change decoder CD to obtain the first semantic feature of the road area environment during the construction period of the highway project; Input the first semantic feature of the road area environment during the construction period of the highway project into the fully connected layer FC of the change decoder CD to obtain the change classification probability distribution of the road area environment during the construction period of the highway project; Adopt the maximum probability term for the change classification probability distribution of the road area environment during the construction period of the highway project to obtain the final change type of the road area environment during the construction period of the highway project.

[0013] Based on the multi-scale feature differences of the road area environment during the construction period of highway engineering and the final change types of the road area environment during the construction period of highway engineering, generate a change annotation atlas of the road area environment during the construction period of highway engineering and a first change event log of the road area environment during the construction period of highway engineering.

[0014] Furthermore, based on the change annotation atlas of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering, construct a risk warning module RWM for the road area environment during the construction period of highway engineering, and the second change event log of the road area environment during the construction period of highway engineering obtained includes: The risk warning module RWM for the road area environment during the construction period of highway engineering consists of two major modules: a clustering analysis module CAM and a safety level evaluation module for the road area environment during the construction period of highway engineering; Input the change annotation atlas of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering into the clustering analysis module CAM to generate a first risk assessment value for the road area environment during the construction period of highway engineering; Based on the first risk assessment value of the road area environment during the construction period of highway engineering, establish a safety level evaluation module for the road area environment during the construction period of highway engineering; Based on the change annotation atlas of the road area environment during the construction period of highway engineering, the first risk assessment value of the road area environment during the construction period of highway engineering, and the safety level evaluation module for the road area environment during the construction period of highway engineering, generate a second change event log of the road area environment during the construction period of highway engineering.

[0015] A vehicle trajectory prediction system based on deep learning, which is used to implement the above-mentioned vehicle trajectory prediction method based on deep learning. The system includes: Data acquisition module: used to obtain macro-scale satellite sequence data of the road area environment during the construction period of highway engineering by using satellite remote sensing equipment and micro-scale drone sequence data of the road area environment during the construction period of highway engineering by using equipment carried by drones; Data preprocessing module: used to perform preprocessing methods based on the macro-scale satellite sequence data of the road area environment during the construction period of highway engineering and the micro-scale drone sequence data of the road area environment during the construction period of highway engineering; Data feature encoding module: used to input the multi-source perception data of the road area environment during the construction period of highway engineering into a multi-scale time perception encoding module TAFM to obtain macro-scale satellite sequence features and micro-scale drone sequence features; Cross-modal fusion module: used to input the macro-scale satellite sequence features and micro-scale drone sequence features into a multi-modal fusion Transformer module MFTM to obtain multi-scale fourth features of the road area environment during the construction period of highway engineering; Intelligent change detection module: It is used to input the multi-scale fourth feature of the road area environment during the construction period of highway engineering into the change detection module CDM to obtain the change annotation map of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering; Risk warning decision-making module: It is used to construct the risk warning module RWM of the road area environment during the construction period of highway engineering based on the change annotation map of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering, and obtain the second change event log of the road area environment during the construction period of highway engineering.

[0016] Each of the above modules is executed in sequence to achieve a complete closed-loop from environmental perception to trajectory prediction.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: A road area environment safety evaluation method and system during the construction period of highway engineering according to the present invention first comprehensively utilizes satellite remote sensing and unmanned aerial vehicle (UAV)-mounted equipment to construct an environmental perception system combining macro and micro scales. The satellite remote sensing equipment is used to obtain the macro-scale satellite sequence data of the road area environment during the construction period of highway engineering to realize the dynamic monitoring of large-scale environmental problems such as terrain changes and ecological damage, while the UAV-mounted equipment obtains the micro-scale UAV sequence data of the road area environment during the construction period of highway engineering, focusing on the key areas of the construction site to realize the fine perception of construction behaviors and local risks. The multi-source perception data of the road area environment during the construction period of highway engineering obtained above is preprocessed to effectively improve the spatial coverage breadth and time update frequency of environmental monitoring. Compared with the traditional method relying on manual inspections or single perception equipment, this method significantly improves the automation and timeliness of environmental risk identification, can quickly detect risk events such as slope collapses, spoil piles, and ecological disturbances, timely identify illegal construction and potential impacts on sensitive areas, support the dynamic analysis and visual expression of environmental change trends, and provide efficient and intelligent technical support for engineering environmental protection supervision and emergency response.

[0018] On this basis, a multi-scale time-aware encoding module TAFM with a "dual-channel" structure is adopted to achieve differential modeling for different scale information in multi-source perception data of the road area environment during the construction period of highway engineering. In the multi-scale time-aware encoding module TAFM, one channel is dedicated to processing satellite image sequences at the macro scale to extract the overall regional change trend; the other channel processes drone image sequences at the micro scale to focus on the detailed changes at the construction site. This dual-channel design effectively realizes the time modeling collaboration from the global trend to the local dynamics, enhancing the system's perception ability of complex spatio-temporal changes. During the construction of highway engineering, environmental problems such as soil erosion, vegetation damage, and waste accumulation often show "slow-changing" characteristics, which are not easily detectable in the early stage but may cause serious consequences. The TAFM module, through the synchronous perception and temporal modeling of macro and micro changes, can not only continuously track the environmental evolution at the regional level but also keenly capture micro anomalies such as equipment violations and local dust emissions, realizing more intelligent and refined environmental monitoring and early warning, and effectively improving the practicality and reliability of the system in dynamic construction scenarios.

[0019] Furthermore, the macro-scale satellite sequence features and micro-scale drone sequence features are input into the multi-modal fusion Transformer module MFTM to obtain the multi-scale fourth feature of the road area environment during the construction period of highway engineering. The multi-modal fusion Transformer module MFTM includes a dynamic modal gating module DMGM and a multi-scale attention fusion module CSAFM. Among them, the DMGM can dynamically adjust the weights of satellite and drone data according to the actual environmental changes, realizing more flexible and robust data fusion; further introducing learnable embedding vectors to enhance the expression ability of the semantics of multi-source data; then focusing on key areas and abnormal changes through the CSAFM module; finally, generating the multi-scale fourth feature through global average pooling operation for supporting subsequent environmental risk identification and safety level assessment. This method enables the system to have the ability of intelligent data source discrimination and weight allocation. For example, in cloudy and low visibility conditions, drone images are preferentially used, and satellite images are emphasized for reference during large-scale scheduling. At the same time, through the attention mechanism, the system can accurately locate high-risk areas, such as behaviors like abnormal material stacking, pollution diffusion, and boundary-crossing construction, significantly improving the timeliness and accuracy of problem discovery. This fusion mechanism is especially suitable for construction scenarios with sensitive environments and high regulatory requirements, such as highway construction projects passing through ecological protection areas, adjacent to schools or residential areas, with higher monitoring accuracy and early warning effectiveness.

[0020] Subsequently, a Change Detection Module (CDM) is constructed to obtain the change annotation map of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering. The Change Detection Module (CDM) includes a Multi-scale Feature Difference Encoding Module (TSDEM) and a Change Decoder (CD). Among them, the Multi-scale Feature Difference Encoding Module (TSDEM) extracts the basic differences between different time nodes through the Sliding Window (SW) mechanism, and at the same time uses the Semantic Residual (SR) mechanism to capture more complex semantic-level changes, so as to generate a high-resolution "feature difference quantity"; the Change Decoder (CD) decodes and classifies these differences, identifies the change type and outputs the final detection result. The system can automatically generate an intuitive change annotation map for displaying the spatial change area, and at the same time record and form a structured first change event log to clarify the content, type and occurrence time of each change, providing a key basis for subsequent environmental assessment and management. Compared with the traditional method that relies on manual comparison of images before and after, this module has higher automation and accuracy, and can simulate an intelligent monitor with "memory ability", automatically compare multi-temporal images, identify abnormal changes and accurately annotate. For example, when the system detects that a certain area has changed from an empty space to a large amount of piled soil within three days, it can quickly label this area as "construction material pile change" and automatically record it in the event log.

[0021] Finally, a Risk Warning Module (RWM) for the road area environment during the construction period of highway engineering is constructed to obtain the second change event log of the road area environment during the construction period of highway engineering. The Risk Warning Module (RWM) consists of a Clustering Analysis Module (CAM) and a Safety Level Evaluation Module. The Clustering Analysis Module (CAM) analyzes the features (such as location, frequency, scope, etc.) in the change map, clusters similar or related changes into one category, and gives the first risk assessment value. The Safety Level Evaluation Module judges the risk level of each change according to the first risk assessment value and the nature of the change, and finally generates the second change event log, marking the safety level and risk corresponding to each change. This method can effectively identify potential safety risks at the highway construction site, helping managers not only understand "what changes have occurred", but also accurately evaluate "whether the changes pose a safety threat and its severity". Through this intelligent risk warning system, managers can make timely safety prevention and control decisions based on scientific evaluations, thus significantly improving the risk perception ability and emergency response efficiency at the construction site. Description of the Drawings

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0023] Figure 1 is the principle flow chart of a method for evaluating the safety of the road area environment during the construction period of a highway project of the present invention; Figure 2 is the network structure framework diagram of a method for evaluating the safety of the road area environment during the construction period of a highway project of the present invention; Figure 3 is the example diagram of the first change event log of a method for evaluating the safety of the road area environment during the construction period of a highway project of the present invention; Figure 4 is the example diagram of the second change event log of a method for evaluating the safety of the road area environment during the construction period of a highway project of the present invention; Figure 5 is the functional module diagram of a system for evaluating the safety of the road area environment during the construction period of a highway project of the present invention. Detailed implementation manners

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0025] Embodiment 1 Please refer to Figure 1 as shown, this embodiment provides a method for evaluating the safety of the road area environment during the construction period of a highway project, including: Step S1000, collecting multi-source perception sequence data of the road area environment during the construction period of the highway project.

[0026] Specifically, the purpose of this step is to systematically collect environmental data covering different time scales and spatial scales during the construction period of the highway project by deploying multi-source perception devices, providing multi-dimensional and highly reliable data support for subsequent change analysis, risk factor identification, and safety decision-making of the road area environment during the construction period of the highway project.

[0027] The multi-source perception sequence data of the road area environment during the construction period of the highway project includes satellite sequence data at the macro scale and UAV sequence data at the micro scale. Among them, the satellite sequence data at the macro scale is mainly used to reflect the dynamic evolution trend of the road area environment during the construction period of the highway project from a global perspective; the UAV sequence data at the micro scale focuses on achieving high-resolution modeling of the road area environment during the construction period of the highway project and capturing the local environmental state.

[0028] The deployed multi-source perception devices include satellite remote sensing devices and UAV-mounted devices, which are respectively used to achieve large-scale continuous monitoring and high-precision modeling of key areas.

[0029] Furthermore, step S1000 includes: Step S1100, deploying satellite remote sensing devices to collect satellite sequence data at the macro scale.

[0030] Using Sentinel-2 MSI (resolution 10 - 60m) + Landsat-9 OLI-2 (resolution 30m) satellite remote sensing devices, perform periodic image acquisition on the road area environment during highway construction to obtain satellite sequence data at the macro scale.

[0031] Specifically, the acquisition parameter configuration of the satellite sequence data at the macro scale includes: acquisition period: the regular acquisition period is 15 days / time, and it is adjusted to 7 days / time during the rainy season or seasons prone to anomalies; band range: covering the visible light band (RGB), red edge band (RE1 - RE3), and shortwave infrared band (SWIR) to support analysis tasks such as vegetation vitality detection, soil moisture estimation, and land cover type identification; coverage range: centered on the construction area, set a monitoring coverage range of a 50-km radius area.

[0032] Step S1200, deploying UAV-mounted devices to collect UAV sequence data at the micro scale.

[0033] Specifically, the UAV-mounted device is equipped with a multi-spectral sensor and a lidar (LiDAR) device to perform high-precision remote sensing scanning and three-dimensional point cloud reconstruction on the highway project construction area and its surrounding areas, so as to achieve airspace perception and high-precision modeling of the construction environment, and provide accurate data support for subsequent safety evaluation and optimization.

[0034] The UAV platform device selects an industrial six-rotor UAV platform with RTK / PPK high-precision positioning capabilities, and preferably selects DJI Matrice 300 RTK as the flight vehicle. DJI Matrice 300 RTK supports multi-payload mounting, has a single flight endurance of 55 minutes, a wind resistance level of 6, and supports long-distance image transmission (OcuSync3.0 protocol) at a maximum distance of 15 km. It is compatible with synchronous mounting of multi-spectral cameras and lidar to ensure efficient and stable aerial photography tasks in complex terrain or large-area construction scenarios.

[0035] The sensing devices carried by the UAV include a multi-spectral sensor (such as MicaSence Altum-PT) and a lidar sensor (such as DJI L1). Specifically, the multi-spectral sensor covers 5 spectral bands including blue, green, red, red edge, and near-infrared, supports real-time irradiance compensation and image radiometric calibration, and outputs multi-dimensional spectral reflectance images for vegetation identification, surface object classification, and detection of abnormal environmental areas; the lidar sensor uses a 905nm laser wavelength, has a point cloud sampling frequency as high as 240,000 points / second, and is built with a high-precision IMU and GNSS module, which can collect high-density three-dimensional point cloud data in the construction area in real time for constructing a fine digital elevation model (DEM) of the road area environment during the construction period of highway engineering and extracting obstacle geometric information.

[0036] Specifically, the acquisition parameter configuration of the microscopic-scale UAV sequence data includes: flight network design: adopting a 500×500m network block sampling method, aligning with the boundary of the macroscopic-scale satellite sequence data to achieve spatial consistency of multi-scale data fusion; acquisition frequency: regular monitoring (full-area inspection once a week) and emergency monitoring (for abnormal areas found in the macroscopic-scale satellite sequence data monitoring, activate the emergency response mechanism and conduct key area acquisition once a day); operation conditions and metadata synchronization recording: operations should be carried out under the condition that the solar altitude angle is greater than 30°, and metadata such as wind speed (less than 8m / s) and light intensity (greater than 50000lux) should be recorded in real time during flight to ensure the quality of image data and the accuracy of point clouds.

[0037] Step S2000, preprocess the multi-source sensing data of the road area environment during the construction period of highway engineering.

[0038] Specifically, this step aims to standardize and normalize the multi-source sensing data of the road area environment during the construction period of highway engineering, ensure the consistency and fusibility of the data in terms of spatial reference, time dimension, and resolution, and at the same time improve the geometric accuracy and radiation reliability of the data, laying a solid foundation for subsequent environmental change analysis and risk identification of the road area environment during the construction period of highway engineering.

[0039] Further, step S2000 includes: Step S2100, performing unified spatio-temporal benchmark processing on multi-source perception data of the road area environment during the highway engineering construction period.

[0040] Based on the multi-source perception data of the road area environment during the highway engineering construction period, first, perform coordinate reference transformation to uniformly transform all multi-source perception data of the road area environment during the highway engineering construction period to the China Geodetic Coordinate System 2000 (CGCS2000); second, perform time alignment processing, uniformly using UTC timestamps for annotation with a precision reaching the millisecond level to ensure the time consistency and synchronization of the multi-source data of the road area environment during the highway engineering construction period; finally, perform spatial resolution matching. Among them, for the satellite sequence data at the macro scale, bicubic interpolation is used for downsampling to a unified 2m spatial resolution, and for the UAV sequence data at the micro scale, an adaptive grid aggregation method is adopted to maintain a high-precision 0.05m spatial resolution, realizing spatial scale alignment before data fusion.

[0041] Specifically, this step improves the accuracy and consistency of the multi-source perception data of the road area environment during the highway engineering construction period, provides a high-quality and fusible data basis for subsequent environmental change analysis and risk identification, and ensures accurate spatial positioning and time series analysis.

[0042] Step S2200, performing radiometric correction and geometric correction processing based on the multi-source perception data of the road area environment during the highway engineering construction period.

[0043] For the radiometric characteristics and geometric distortions of data from different sources in the road area environment during the highway engineering construction period, different correction strategies are adopted respectively.

[0044] Based on the satellite sequence data at the macro scale, the Sen2Cor atmospheric correction processor is used for radiometric correction, and the Shuttle Radar Topography Mission (SRTM) 1 arc-second Digital Elevation Model (DEM) is combined to complete terrain correction to eliminate the interference of terrain undulation on image projection and reflectivity.

[0045] Based on the UAV sequence data at the micro scale, multi-spectral radiometric calibration and lidar point cloud geometric correction are adopted. Among them, multi-spectral radiometric calibration realizes radiometric normalization processing of images through a reflectance standard panel (such as Labsphere Spectralon) in cooperation with an irradiance sensor; lidar point cloud geometric correction is based on the trajectory solution of the Inertial Measurement Unit (IMU) / Global Navigation Satellite System (GNSS) integrated navigation system, and combined with the Cloth Simulation Filtering (CSF) algorithm for classification of ground points and non-ground points, improving the point cloud accuracy and geometric expression ability.

[0046] Specifically, this step ensures the unity and reliability between different data sources, provides an accurate data basis for subsequent data fusion and environmental change analysis, and significantly improves the accuracy of monitoring and decision-making.

[0047] Step S2300: Conduct quality control and accuracy verification based on the multi-source perception data of the road area environment during the construction period of highway engineering.

[0048] Based on the multi-source perception data of the road area environment during the construction period of highway engineering, first, perform abnormal data filtering operations, including: eliminating low-quality satellite image data with cloud cover greater than 10% and excluding UAV data with acceleration exceeding 0.5g in the IMU records to ensure data stability and continuity. Secondly, perform accuracy verification operations. Among them, based on the satellite sequence data at the macro scale, control the error within 0.05 through cross-verification of multiple NDVI calculation results; based on the UAV sequence data at the micro scale, conduct accuracy assessment by arranging inspection points to ensure that the planar position error does not exceed 3 cm, thereby ensuring the reliability and accuracy of the input data for the subsequent analysis model.

[0049] Specifically, this step significantly improves the stability and credibility of the multi-source perception data of the road area environment during the construction period of highway engineering, effectively avoids misjudgment of changes or deviation in risk identification caused by input data errors, and thus enhances the perception accuracy and decision-making reliability of the system for the evolution of the road area environment during the construction period of highway engineering.

[0050] Further, please refer to Figure 2 as shown in Figure 2 which is the network structure framework diagram of a road area environment safety evaluation method for highway engineering during the construction period of the present invention.

[0051] Step S3000: Input the multi-source perception data of the road area environment during the construction period of highway engineering into the multi-scale time-aware encoding module (TAFM) to obtain the satellite sequence features at the macro scale and the UAV sequence features at the micro scale .

[0052] Specifically, based on the multi-source perception data of the road area environment during the highway engineering construction period, a multi-scale time perception coding module TAFM is constructed to effectively capture the dynamic characteristics of the road area environment during the highway engineering construction period at different time scales. The multi-source perception data of the road area environment during the highway engineering construction period includes the macro-scale satellite sequence data of the road area environment during the highway engineering construction period and the micro-scale drone sequence data of the road area environment during the highway engineering construction period, which have different time resolutions and perception levels respectively. Among them, the macro-scale satellite sequence data of the road area environment during the highway engineering construction period has a low time resolution but a wide coverage area, which is suitable for long-term change trend modeling; the micro-scale drone sequence data of the road area environment during the highway engineering construction period has a high time resolution and spatial details, which is suitable for capturing local short-term disturbances.

[0053] The multi-scale time perception coding module TAFM adopts a dual-channel design, including a macro-scale satellite time encoder and a micro-scale drone time encoder, which cooperate to process multi-source perception data with different time resolutions, and realize the collaborative analysis of the macro-trend perception and micro-time capture of the road area environment change during the highway engineering construction period. Among them, the macro-scale satellite time encoder is used to model the trend of environmental change within a long time period, and the micro-scale drone time encoder is used to perceive construction interference or sudden pollution events within a short period.

[0054] Specifically, during the highway engineering construction period, there are significant time-scale differences in the road area environment change: on the one hand, factors such as vegetation coverage and hydrogeomorphology usually evolve slowly over time; on the other hand, sudden construction activities or local disturbances will have an impact quickly within a short time. Therefore, data at a single time scale is difficult to comprehensively reflect these complex dynamics. Through the dual-channel design of the multi-scale time perception coding module TAFM, the collaborative perception of the macro-scale satellite sequence data and the micro-scale drone sequence data of the road area environment during the highway engineering construction period can be realized, and the evolution law of the road area environment can be mined from the two perspectives of macro-evolution and micro-disturbance respectively, providing high-quality characteristic time-series semantic input for subsequent analysis.

[0055] Furthermore, step S3000 includes: Step S3100, input the macro-scale satellite sequence data of the road area environment during the highway engineering construction period into the macro-scale satellite time encoder to obtain the macro-scale satellite sequence features .

[0056] Specifically, the construction of the macro-scale satellite time encoder is used to capture the regional environmental evolution trend within a long time span (such as long-term bare soil expansion, vegetation decline, etc.).

[0057] First, based on the macro-scale satellite sequence data in each frame of image Perform the Patch embedding operation and divide it into non-overlapping image patches, with the size of each image patch being , and map it through a linear transformation to obtain the first macro-scale satellite embedding vector . The process is as follows:

[0058] where T is the total number of time steps (e.g., T = 24 represents the macro-scale satellite sequence data for each month within two years), is the set of real numbers, is the number of patches, is the embedding dimension, is a -dimensional real vector.

[0059] Specifically, the Patch embedding operation refers to dividing each frame of the image into N non-overlapping image patches, and the size of each patch is pixels.

[0060] Subsequently, to capture the temporal relationship between image frames, based on the first macro-scale satellite embedding vector introduce a learnable positional encoding vector to obtain the second macro-scale satellite embedding vector , which is expressed as follows:

[0061] where, is the set of real numbers, is the embedding dimension, is a -dimensional real vector.

[0062] Next, input the second macro-scale satellite embedding vector into the standard multi-layer Transformer encoder to establish the context dependency across time steps and obtain the third macro-scale satellite embedding vector

[0063] where T is the total number of time steps, is the set of real numbers, is the embedding dimension, is a -dimensional real vector.

[0064] Finally, use a multi-layer perceptron (Multi-Layer Perception, MLP) for the third macro-scale satellite embedding vector Perform an embedding mapping to obtain a macro-scale satellite sequence feature representation with a fixed dimension as , where Stack means stacking (concatenating) multiple vectors along a new one-dimensional axis to form a matrix. Here, the results output by multiple MLPs are stacked into a matrix with a shape of dimensions, represents a multi-layer perceptron neural network that performs a non-linear transformation on each input third macro-scale satellite embedding vector and outputs a -dimensional vector, is the set of real numbers, T is the total number of time steps, is the embedding dimension, is a -dimensional real vector.

[0065] Exemplarily, taking a construction section of a mountain highway as an example, satellite remote sensing images are collected every 15 days to form macro-scale satellite sequence data for one year (T = 12). The TAFM module can identify long-term macro trends such as "continuous decline in vegetation coverage", "monthly expansion of bare ground areas", and "long-term construction disturbances appear next to the river channel". These trends are not easily detectable in a short period of time, but show clear change trajectories over the entire annual cycle, providing a global time context for subsequent disaster prediction and environmental degradation analysis.

[0066] Specifically, this step introduces the macro-scale time series modeling ability, effectively solving the problem of weak ability to capture "low-frequency, long-cycle change trends" in the environmental assessment process. The modeling of macro-scale satellite sequence features not only provides a global perspective for problems such as ecological degradation and disaster risks, but also helps subsequent models identify the cumulative effects of environmental disturbances brought about by large-scale construction, enhancing the practicality and interpretability of the system in scenarios such as policy evaluation, resource scheduling, and long-term early warning, and providing a reference for macro decision-making.

[0067] Step S3200: Input the micro-scale UAV sequence data of the road area environment during the highway construction period into the micro-scale UAV time encoder to obtain micro-scale UAV sequence features .

[0068] Specifically, the construction of the micro-scale UAV time encoder is used to capture regional environmental evolution trends within a short time span (such as water seepage marks appear on the slope, and the earthwork height in the highway construction area increases sharply, etc.).

[0069] First, the micro-scale UAV sequence data is sequentially input into the convolutional embedding network to form the first micro-scale UAV embedding vector , which is expressed as follows:

[0070] Among them represents the number of frames of the microscopic-scale UAV sequence data acquisition (for example, for the sequence data of 5 consecutive days, K = 5), represents each frame of microscopic-scale UAV image, is a set of real numbers, respectively represent the height and width of the first microscopic-scale UAV sequence embedding vector, represents the embedding dimension, is a dimensional real vector.

[0071] Next, the first microscopic-scale UAV embedding vector is input into the lightweight temporal convolutional ConvLSTM encoder to obtain the second microscopic-scale UAV embedding vector .

[0072] Finally, a perceptron MLP is used to perform an embedding mapping on the second representation of the microscopic-scale UAV sequence embedding vector to obtain a fixed-dimensional microscopic-scale UAV sequence feature representation as where , is a set of real numbers, represents the number of frames of the microscopic-scale UAV sequence data acquisition, represents the embedding dimension, is a dimensional real vector.

[0073] Exemplarily, taking a mountainous highway slope section construction area as an example, the project uses UAVs to conduct anti-aerial photography cruises every 3 days, constructs microscopic-scale UAV sequence data with a length of 5 frames (K = 5), captures sudden changes such as "sudden temporary drainage ditch excavation" and "the exposed earthwork area is covered by sediment due to rainfall, resulting in instantaneous occlusion of the surrounding vegetation growth area", etc. After being modeled by the microscopic-scale UAV time encoder of TAFM, it can accurately and timely capture local disturbance events and mark their time, form a microscopic disturbance representation with temporal semantics, and provide high-frequency and fine-grained input data for subsequent anomaly detection and environmental assessment.

[0074] Specifically, this step can process disturbance information within a high frequency and short time span, which helps to improve the system's response ability and discrimination accuracy to emergencies at the construction site. Compared with the traditional single-frame image recognition method, the proposed temporal modeling mechanism is more applicable to the complex and changeable construction dynamic environment and has important application value in ecological protection, safety management, and emergency warning during the construction period.

[0075] Step S4000, the macro-scale satellite sequence features and the sequence features of micro-scale drones Input them into the Multi-modal Fusion Transformer Module (MFTM) to obtain the multi-scale fourth feature of the road area environment during the construction period of highway engineering .

[0076] Specifically, the multi-modal fusion Transformer module MFTM includes a Dynamic Modal Gate Module (DMGM) and a Cross-Scale Attention Fusion Module (CSAFM), which realizes the deep alignment and feature interaction modeling of data at different scales between the macro-scale satellite sequence features and the micro-scale drone sequence features, and improves the detection accuracy and response speed of ecological disturbances and environmental risks in the road area environment during the construction period of highway engineering.

[0077] Furthermore, step S4000 includes: Step S4100, input the macro-scale satellite sequence features and the sequence features of micro-scale drones into the Dynamic Modal Gate Module DMGM to obtain the multi-scale first feature of the road area environment during the construction period of highway engineering .

[0078] Specifically, the Dynamic Modal Gate Module DMGM can automatically learn the weights of the macro-scale satellite sequence features and the sequence features of micro-scale drones . The process of the Dynamic Modal Gate Module DMGM learning weights is as follows:

[0079] wherein, represents the Sigmoid function, represents the learnable weight matrix, represents the learnable bias term, represents the selection weight of the Dynamic Modal Gate Module DMGM. Specifically, the learning parameters are automatically adjusted through backpropagation during the training phase, so that useful modalities can be dynamically selected under different environmental conditions.

[0080] Exemplarily, taking the peak construction period in spring as an example, construction machinery operates frequently, and dust pollution occurs frequently and changes violently. At this time, macro-scale satellite images are restricted by the shooting frequency and resolution, and it is difficult to capture the small-scale pollution diffusion process. In this scenario, micro-scale UAV thermal infrared images are more sensitive to the dust distribution. The dynamic modality gating module DMGM can automatically increase the weight of the micro-scale UAV sequence features and reduce the influence of the macro-scale satellite feature sequence.

[0081] Finally, based on the macro-scale satellite sequence features and the micro-scale UAV sequence features weights perform dynamic weighted fusion to obtain the multi-scale first feature of the road area environment during the construction period of the highway project , and the process is as follows:

[0082] Exemplarily, taking the scenario of potential landslide hazards occurring after continuous rainy weather in a certain mountainous highway section as an example. The reflectivity and texture changes of the slope in the macro-scale satellite data are weak, and it is difficult to accurately locate potential landslide points. In contrast, the high-frequency thermal infrared images in the micro-scale UAV data reflect the temperature anomalies in the landslide area. Finally, the dynamic modality gating module DMGM determines that the micro-scale UAV data is the dominant one, increases its weight, dynamically adjusts the fusion strategy, and highlights its time sensitivity.

[0083] Specifically, this step can highlight the key modality (such as thermal infrared or infrared data) according to different abnormal scenarios such as environmental pollution and terrain damage, and weaken the influence of the noise modality on the final discrimination result. For example, in long-term environmental monitoring, macro-scale satellite data may have a longer-term trend, while sudden construction disturbances within a short period rely on micro-scale UAV data.

[0084] Step S4200, based on the multi-scale first feature of the road area environment during the construction period of the highway project , introduce a learnable embedding vector to obtain the multi-scale second feature of the road area environment during the construction period of the highway project , and the process is as follows:

[0085] Among them, is a learnable embedding vector, which is used to explicitly encode the information of each frame from the macro-scale satellite sequence features or the micro-scale UAV sequence features, and helps the model capture the interaction between multi-scale information.

[0086] Exemplarily, taking the long-term monitoring of a certain construction site as an example, the observation data at different times alternately come from different modalities (for example, micro-scale UAV data cannot be obtained on cloudy days, and only macro-scale satellite data is available). By introducing learnable embedding vectors , the module can learn to maintain information consistency when modalities are missing or switched. For example, when predicting the diffusion trend of a polluted area, data filling and deduction are automatically performed according to historical patterns.

[0087] Step S4300: Input the multi-scale second features of the road area environment during the construction period of the highway project into the multi-scale attention fusion module CSAFM to obtain the multi-scale third features of the road area environment during the construction period of the highway project .

[0088] Specifically, the multi-scale attention fusion module CSAFM includes a multi-layer structure of multi-head attention, residual connection, LayerNorm, and feed-forward network. The specific calculation formula is as follows:

[0089] Among them, LayerNorm represents the normalization operation, Q represents the query matrix, K represents the key matrix, V represents the value matrix, and MHA represents the multi-head attention mechanism. Its specific formula is as follows:

[0090] Among them, h represents the number of heads, represents the output of the i-th head, represents the output transformation matrix. Each attention head The calculation formula is:

[0091] Among them is the learnable weight matrix. The core calculation of the self-attention mechanism is as follows:

[0092] Among them, represents the dimension of the key, and softmax is used for normalization.

[0093] Exemplarily, taking the construction of a new elevated road in an urban fringe section as an example, the macro-scale satellite data shows a continuous decrease in vegetation. At this time, the micro-scale UAV data finds that there are illegal soil dumps and turbid water bodies in some areas. The multi-scale attention fusion module CSAFM can discover the temporal correlation and spatial proximity relationship between the disappearance of vegetation and water pollution, infer the existence of an ecological disturbance chain at this construction site, and support subsequent risk intervention measures.

[0094] Specifically, this module can mine macro-scale satellite sequence features Micro-scale drone features Based on the deep interactive relationship in time, space and modal semantics, it is possible to accurately identify the ecological disturbance chain caused by the linkage of multiple factors such as vegetation degradation and water pollution, and provide more logical evidence support and judgment basis for subsequent change detection and risk intervention.

[0095] Step S4400: multi-scale third characteristics of the road environment during the highway construction period The global average pooling (GAP) operation is used to obtain the multi-scale fourth feature of the road environment during the highway construction period. , expressed as:

[0096] Specifically, this step makes the multi-scale fourth characteristic of the road environment during the highway construction period It has good semantic integrity and representation stability, and provides a reliable input basis for environmental safety evaluation during highway construction.

[0097] Step S5000: The multi-scale fourth characteristic of the road environment during the highway construction period Input to the Change Detection Module (CDM) to obtain the change annotation map of the road environment during the highway construction period The first change event log of the road environment during the highway construction period.

[0098] Specifically, the change detection module CDM consists of two modules: a multi-scale feature difference encoding module (Temporal-Semantic Difference Encoding Module, TSDEM) and a change decoder (ChangeDecoder, CD), which are used to identify environmental risk changes such as pollution source diffusion, land destruction, illegal construction, ecological degradation, etc. that may occur during the highway construction period, and output a change annotation map of the road environment during the highway construction period with timestamps and spatial location information. The first change event log of the road environment during the highway construction period provides a basic basis for subsequent risk warning and decision support.

[0099] Further step S5000 includes: Step S5100: The fourth multi-scale characteristic of the road environment during the highway construction period is , input the multi-scale feature difference quantity encoding module TSDEM to obtain the multi-scale feature difference quantity of the road area environment during the construction period of highway engineering .

[0100] Specifically, the multi-scale fourth feature of the road area environment during the construction period of highway engineering , where represents the time step of the multi-scale fourth feature of the road area environment during the construction period of highway engineering (for example, data for 24 months, T = 24), represents the th frame of the multi-scale fourth feature of the road area environment during the construction period of highway engineering.

[0101] Adopt the sliding window mechanism (Sliding Window, SW), and at any time calculate the feature difference quantity of the road area environment during the construction period of highway engineering , and the calculation process is as follows:

[0102] Among them, the calculation principle of the formula is Euclidean Distance (ED) measurement, which is used to reflect the feature change degree of the road area environment during the construction period of highway engineering at this t moment and the moment, represents the sliding window size.

[0103] Exemplarily, taking a mountain slope construction area in a certain highway engineering project as an example, high-resolution satellite remote sensing and unmanned aerial vehicle automatic inspection paths are arranged in the area, and macro-scale satellite data and micro-scale unmanned aerial vehicle data are continuously collected from August 2022 to July 2023. During the period from May 2023 to July 2023, the feature difference quantity of the road area environment during the construction period of highway engineering in this area suddenly increases from 0.2 to 1.5, then it is recognized that there is obvious surface disturbance in this area, and a warning of the "soil erosion" type is triggered.

[0104] Specifically, the feature difference quantity of the road area environment during the construction period of highway engineering, combined with the sliding window mechanism SW and the Euclidean distance ED measurement method, can dynamically model the change of feature states in the time series. Compared with the traditional static judgment method based on a single-frame image, this method fully explores the evolutionary trend information of multiple time series and effectively enhances the system's response ability to risks of the "cumulative slow change - mutation trigger" type. This step not only realizes the early warning of the "construction disturbance appearance period", but also improves the system's automatic recognition ability of weak trends and hidden risks, providing high support for subsequent risk assessment and hierarchical control.

[0105] In addition, to capture the high-order change trend of the road area environment during the construction period of highway engineering at the semantic level, a semantic residual mechanism (Semantic Residual, SR) is introduced to obtain the high-order feature difference of the road area environment during the construction period of highway engineering. , which is used to measure the deviation degree between the current frame feature state and the historical average semantic state. The calculation process is as follows:

[0106] Among them, represents the mean value of the features within the past W time windows, indicating the historical average semantic state of the road area environment during the construction period of highway engineering.

[0107] Exemplarily, taking an excavation section of a mountain highway as an example, satellite remote sensing and regular drone inspection points are set up in this area to continuously collect macro-scale satellite data and micro-scale drone data from July 2022 to July 2023. During the period from May 2023 to July 2023, there is no obvious visual change in the highway slope images at the micro level (the feature difference of the road area environment during the construction period of highway engineering ), but the high-order feature difference of the road area environment during the construction period of highway engineering continues to rise, exceeding 0.8 times the historical mean. Combining the change in water body reflectivity and the expansion of the earthwork accumulation area in the image, the system determines that there is a concealed construction disturbance behavior in this area and initiates the "ecological slow-variant risk" early warning process.

[0108] Specifically, introducing the residual mechanism between the current frame feature state and the historical average semantic state can effectively capture sudden anomalies (such as sudden pollution) or gradual deviations (such as the degradation of green plants), making up for the defect that the feature difference of the road area environment during the construction period of highway engineering by a simple sliding window is insensitive to semantic information, and is especially suitable for capturing potential risk signals such as construction behaviors or sudden changes in natural states during the concealed construction period of highway engineering.

[0109] Finally, based on the multi-scale fourth feature of the road area environment in each frame during the construction period of highway engineering , the feature difference of the road area environment during the construction period of highway engineering and the high-order feature difference of the road area environment during the construction period of highway engineering , the multi-scale feature difference of the road area environment during the construction period of highway engineering is obtained. The process is as follows:

[0110] Specifically, the multi-scale feature difference of the road area environment during the construction period of highway engineering Integrating triple information of the semantic state, local change intensity, and trend offset at the current moment is conducive to the model capturing the composite change signal of "slow change superimposed with mutation".

[0111] Step S5200: Input the multi-scale feature difference quantity of the road area environment during the construction period of the highway project into the change decoder CD to obtain the final change type of the road area environment during the construction period of the highway project .

[0112] Specifically, the change decoder CD includes two layers of Transformer decoders and one layer of fully connected layer (FC) to realize the recognition and spatio-temporal positioning of various change types in the road area environment during the construction period, and has strong semantic modeling and classification capabilities.

[0113] First, input the multi-scale time difference measurement features of the road area environment during the construction period of the highway project into two layers of standard Transformer decoding modules to obtain the first semantic feature of the road area environment during the construction period of the highway project ; Then, after inputting the first semantic feature of the road area environment during the construction period of the highway project into the fully connected classification layer FC, obtain the change classification probability distribution of the road area environment during the construction period of the highway project , and the process is as follows:

[0114] Among them, represents the number of change categories of the road area environment during the construction period of the highway project (such as "increase in bare soil", "decrease in water area", "vegetation damage", "pollutant diffusion", etc.), and are learnable parameters. Softmax represents a mathematical function for multi-classification problems, which converts any real number vector into a probability distribution.

[0115] Finally, through the maximum probability term , obtain the final change type of the road area environment during the construction period of the highway project .

[0116] Specifically, this step can not only accurately distinguish the semantic differences between "increase in bare soil" and "vegetation damage", but also realize the accurate positioning and type determination of key environmental changes such as pollution diffusion and decrease in water area, providing clear data support for environmental safety supervision and targeted treatment measures, and significantly improving the recognition accuracy and classification robustness of complex change types.

[0117] Step S5300: Based on the multi-scale feature difference quantity of the road area environment during the construction period of the highway project and the final change types of the road area environment during the construction period of highway engineering , generate the change annotation atlas of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering.

[0118] Specifically, this step converts the recognition results into the change annotation atlas of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering, aiming to achieve the structured expression of the road area environment during the construction period of highway engineering and the visual reconstruction of the dynamic evolution process, and provide visual support for the subsequent system.

[0119] Bind the position features of the corresponding image at each moment to the recognized semantic change categories to generate the change annotation atlas of the road area environment during the construction period of highway engineering , where represents the axis coordinate value of the image, represents the axis coordinate value of the image, represents the timestamp, represents the change category (such as vegetation degradation, water body reduction, increase in construction bare land, etc.), represents the confidence level.

[0120] Based on the change annotation atlas of the road area environment during the construction period of highway engineering , further output the standardized first change event log of the road area environment during the construction period of highway engineering. The first change event log of the road area environment during the construction period of highway engineering includes six elements: Time: the specific date when the event occurs (in the format of YYYY - MM - DD); Location: the longitude and latitude coordinates of the event space position; Type: the recognized change category; Probability: the confidence level of the change category given by the model; Source: the data modality (satellite / drone) on which the change event is based; Trend: the event change trend (such as rising / falling / sudden change / slow change, etc.).

[0121] Exemplarily, such as Figure 3As shown, it presents an example of the first change event log of the road area environment during the construction period of a specific highway project, facilitating an intuitive understanding of the spatio-temporal characteristics and practical reference value of the system output results. This log adopts a structured data format and precisely records two typical environmental change events monitored on March 12, 2025. The first event is the vegetation degradation located at 117.28°E, 31.86°N by GPS, with a probability confidence of 0.91. The data is obtained by an unmanned aerial vehicle, and the change trend is a mutation. The second event is the water body reduction located at 117.25°E, 31.83°N by GPS, with a probability confidence of 0.87, supported by satellite data, and the change trend is a gradual change. As an important data basis for the system to initially identify environmental anomalies, this log has high confidence, standardized expression, and trend modeling capabilities. It not only realizes the standardized archiving of sudden and gradual changes but also provides key support for the subsequent Risk Warning Module (RWM). Its multi-dimensional elements such as time, space, type, trend, and data source provide accurate input for the generation of the second change event log of the road area environment during the highway project construction period, significantly enhancing the system's capabilities in spatial clustering analysis, trend evolution prediction, and risk level judgment, and promoting the transformation of environmental monitoring from single-point identification to system warning.

[0122] Specifically, the comprehensive perception and structured expression of the road area environment changes during the highway project construction period in this step effectively solve the problems of scattered identification, lagging update, and difficult trend judgment of environmental changes during the construction period, significantly improving the intelligent level and response efficiency of the road area environment monitoring during the highway project construction period. It has significant advantages especially in the following aspects: First, the strong robustness enables the model to integrate data at different time scales, that is, it can identify slow changes (such as ecological degradation) and can also promptly capture sudden events (such as pollution diffusion); Second, by introducing semantic residuals and change probability modeling, the classification ability of the model for different change types is effectively improved; Finally, the generated change atlas and log provide a standardized basis for the subsequent visualization, warning decision-making, and environmental protection accountability of the system. In summary, this step not only constructs a graphical expression framework for the road area environment during the highway project construction period but also lays a solid foundation for the intelligent monitoring and decision support of the entire system, having important engineering practice value and promotion application prospects.

[0123] Step S6000: Based on the change annotation atlas of the road area environment during the highway project construction period And the first change event log of the road area environment during the highway project construction period, construct the Risk Warning Module (RWM) of the road area environment during the highway project construction period to obtain the second change event log of the road area environment during the highway project construction period.

[0124] Specifically, the risk warning module RWM for the road area environment during the construction period of highway engineering consists of two major modules: the cluster analysis module (Cluster Analysis Module, CAM) and the safety level evaluation module for the road area environment during the construction period of highway engineering.

[0125] The construction of the risk warning module RWM for the road area environment during the construction period of highway engineering aims to achieve a comprehensive judgment and early warning of potential environmental risks in the road area environment during the construction period of highway engineering. By introducing a multi-scale risk perception mechanism based on spatial aggregation, time evolution trend, and semantic mutation characteristics, the system is empowered to automatically identify high-risk areas, classify and evaluate risk levels, and output response suggestions.

[0126] The further step S6000 includes: Step S6100, input the change annotation map of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering into the cluster analysis module (Cluster Analysis Module, CAM) to generate the first risk assessment value of the road area environment during the construction period of highway engineering .

[0127] Specifically, the cluster analysis module CAM uses the density clustering algorithm DBSCAN (Density-Based Spatial Clustering of Application with Noise) to identify high-risk change areas that are spatially relatively concentrated during different time periods, such as "multi-day continuous water body reduction areas", "pollution diffusion areas", "green plant damage contiguous areas", etc., and at the same time excludes discrete noise points.

[0128] First, the density clustering algorithm DBSCAN can perform cluster analysis on the significant change nodes in the change annotation map of the road area environment during the construction period of highway engineering to divide into change clusters , where each change cluster represents a local area that may have environmental risks. For each change cluster extract features to obtain a spatial feature vector , where the spatial feature vector includes four pieces of information: the number of nodes : measuring the change intensity, the number of change nodes included in the change cluster, the more the number of nodes, the greater the change intensity; the average adjacency degree : measuring the local connectivity within the area; the map density : defined as the ratio of the number of edges to the number of node pairs, the greater the density, the stronger the cohesion of the change area; the spatial span : The maximum node distance within the change cluster, which measures the expansion degree of the change area.

[0129] Based on each change cluster of the spatial feature vector the number of nodes , the average adjacency degree , the spectral density and the spatial span , the first risk assessment value of the road area environment during the construction period of the highway project is obtained , and the process is as follows:

[0130] where is the weight, which reflects the contribution degree of each piece of information in the spatial feature vector of each change cluster to the risk assessment.

[0131] Step S6200, based on the first risk assessment value of the road area environment during the construction period of the highway project , establish a safety level evaluation module for the road area environment during the construction period of the highway project.

[0132] Specifically, according to the threshold setting, it is divided into four categories: Class I safety, Class II controllable, Class III warning, and Class IV high risk.

[0133] Class I safety: , the risk of the change area is extremely low, and there is almost no need for intervention; Class II controllable: , there are minor risks, which can be managed through conventional monitoring means; Class III warning: , there are relatively high potential risks, and observation and intervention measures should be taken immediately; Class IV high risk: , there are obvious environmental threats, and it is necessary to give priority to starting the emergency response mechanism.

[0134] The above thresholds , and can be adaptively adjusted based on the statistical distribution of historical case data or expert experience to adapt to the characteristic differences of different geographical regions and construction scenarios.

[0135] Specifically, in this step, by constructing a safety level evaluation module for the road area environment during the construction period of highway engineering, complex spatial risk information is converted into intuitive and operable safety level labels, and differential supervision at different levels is also supported, making the safety warning of the road area environment during the construction period of highway engineering more interpretable and operationally guiding, helping to improve the construction management efficiency and the accuracy of resource scheduling, so as to realize the dynamic perception and hierarchical control of environmental risks in diverse construction scenarios.

[0136] Step S6300, generating a second change event log for the road area environment during the construction period of highway engineering based on the change annotation map, the first risk assessment value of the road area environment during the construction period of highway engineering, and the safety level evaluation module for the road area environment during the construction period of highway engineering. and the first risk assessment value of the road area environment during the construction period of highway engineering and the safety level evaluation module for the road area environment during the construction period of highway engineering, to generate a second change event log for the road area environment during the construction period of highway engineering.

[0137] The first change event log of the road area environment during the construction period of highway engineering includes six elements: Time: the specific date when the event occurred (in the format of YYYY-MM-DD); Location: the longitude and latitude coordinates of the event space position; Type: the identified change category; Level: the safety level given by the model (from level I to level IV); Source: the data modality (satellite / drone) on which the change event is based; Trend: the event change trend (such as rising / falling / sudden change / slow change, etc.).

[0138] Exemplarily, as Figure 4 shown, an example of the second change event log of a specific road area environment during the construction period of highway engineering is presented. This log adopts a structured data format and accurately records two typical environmental change events monitored on March 12, 2025. The first event is the vegetation degradation located at 117.28°E, 31.86°N by GPS, with a safety level of IV high risk, data obtained by drone, and a change trend of sudden change; the second event is the water body reduction located at 117.25°E, 31.83°N by GPS, with a safety level of III warning, supported by satellite data, and a change trend of slow change. Such logs play an important role during the highway construction period. On the one hand, they can be used to promptly detect and track sudden or gradual changes in the road area environment; on the other hand, through the systematic integration of multi-dimensional information, quantitative assessment and hierarchical management of environmental risks can be achieved, providing scientific and reliable data support for the formulation of subsequent countermeasures and environmental protection decisions.

[0139] Specifically, the system can achieve dynamic visualization monitoring, hierarchical management, and targeted early warning of environmental risks, improving the response efficiency and management level for sudden or gradual environmental events. On the one hand, the system supports the all-element recording and standardized expression of environmental change information, effectively avoiding the problems of information loss and subjective deviation in traditional manual recording; on the other hand, based on automated recognition and risk classification output, the system can achieve automated detection and classification output of environmental changes, significantly improving the data processing efficiency and the response speed of scientific decision-making, and supporting the rapid interpretation of "change areas, change levels, and their evolution trends". In addition, through the introduction of the "trend" field, it is possible to record and quantify the evolution direction and change speed of change events, helping to identify in advance the exacerbation trend of potential risks and realizing "early warning"; at the same time, with the support of the "data source" field for multi-modal data fusion, the system's ability to judge data quality and credibility is enhanced, and it can intelligently schedule the optimal perception method according to on-site environmental conditions (such as complex meteorology). For example, in cloudy weather conditions, drone image data is preferentially used to ensure the continuity and accuracy of environmental change perception.

[0140] Embodiment 2 Based on Embodiment 1, this embodiment provides a road area environmental safety evaluation system during the construction period of highway engineering, as Figure 5 shown, including: Data acquisition module: used to obtain macro-scale satellite sequence data of the road area environment during the construction period of highway engineering by using satellite remote sensing equipment and micro-scale drone sequence data of the road area environment during the construction period of highway engineering by using equipment carried by drones; Data preprocessing module: used for preprocessing methods based on the macro-scale satellite sequence data of the road area environment during the construction period of highway engineering and the micro-scale drone sequence data of the road area environment during the construction period of highway engineering; Data feature encoding module: used to input the multi-source perception data of the road area environment during the construction period of highway engineering into the multi-scale time perception encoding module TAFM to obtain macro-scale satellite sequence features and micro-scale drone sequence features; Cross-modal fusion module: used to input the macro-scale satellite sequence features and micro-scale drone sequence features into the multi-modal fusion Transformer module MFTM to obtain multi-scale fourth features of the road area environment during the construction period of highway engineering; Intelligent change detection module: used to input the multi-scale fourth features of the road area environment during the construction period of highway engineering into the change detection module CDM to obtain a change annotation map of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering; Risk early warning decision-making module: used to construct the risk early warning module RWM of the road area environment during the construction period of highway engineering based on the change annotation atlas of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering, and obtain the second change event log of the road area environment during the construction period of highway engineering.

[0141] Each of the above modules is executed in sequence to achieve a complete closed loop from environmental perception to trajectory prediction.

[0142] In the above technical solutions provided in the embodiments of the present application, the parts that are the same as the corresponding technical solutions in the prior art in terms of implementation principles are not described in detail to avoid excessive elaboration.

[0143] As described above in the specific embodiments, the purpose, technical solutions, and beneficial effects of the present invention have been further described in detail. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for evaluating the safety of the road area environment during the construction period of highway engineering, characterized in that, Including: Obtain multi-source perception data of the road area environment during the construction period of highway engineering; Input the multi-source perception data of the road area environment during the construction period of highway engineering into the multi-scale time perception encoding module to obtain macro-scale satellite sequence features and micro-scale drone sequence features; Input the macro-scale satellite sequence features and micro-scale drone sequence features into the multi-modal fusion Transformer module to obtain the multi-scale fourth features of the road area environment during the construction period of highway engineering; Input the multi-scale fourth features of the road area environment during the construction period of highway engineering into the change detection module to obtain the change annotation map of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering; Based on the change annotation map of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering, construct a risk warning module for the road area environment during the construction period of highway engineering to obtain the second change event log of the road area environment during the construction period of highway engineering.

2. The method for evaluating the safety of the road area environment during the construction period of a highway project according to claim 1, wherein, The multi-source perception data of the road area environment during the construction period of highway engineering includes macro-scale satellite sequence data of the road area environment during the construction period of highway engineering and micro-scale drone sequence data of the road area environment during the construction period of highway engineering; The macro-scale satellite sequence data of the road area environment during the construction period of highway engineering is obtained by satellite remote sensing equipment; The micro-scale drone sequence data of the road area environment during the construction period of highway engineering is obtained by drone-mounted equipment.

3. The method for evaluating the safety of the road area environment during the construction period of a highway project according to claim 2, wherein, The multi-scale time perception encoding module adopts a dual-channel design and includes a macro-scale satellite time encoder and a micro-scale drone time encoder.

4. A method for evaluating the safety of the road area environment during the construction period of highway engineering according to claim 3, characterized in that, The multi-modal fusion Transformer module includes a dynamic modal gating module and a multi-scale attention fusion module.

5. A method for evaluating the safety of the road area environment during the construction period of highway engineering according to claim 4, characterized in that, The step of obtaining the multi-scale fourth features of the road area environment during the construction period of highway engineering includes: Input the macro-scale satellite feature sequence and the micro-scale drone feature sequence into the dynamic modal gating module to obtain the multi-scale first features of the road area environment during the construction period of highway engineering; Based on the multi-scale first features of the road area environment during the construction period of highway engineering, introduce a learnable embedding vector to obtain the multi-scale second features of the road area environment during the construction period of highway engineering; Input the multi-scale second features of the road area environment during the construction period of highway engineering into the multi-scale attention fusion module to obtain the multi-scale third features of the road area environment during the construction period of highway engineering; Perform global average pooling operation on the multi-scale third features of the road area environment during the construction period of highway engineering to obtain the multi-scale fourth features of the road area environment during the construction period of highway engineering.

6. The method for evaluating the safety of the road area environment during the construction period of a highway project according to claim 5, wherein, The step of obtaining the change annotation map of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering includes: The change detection module includes a multi-scale feature difference quantity encoding module and a change decoder; Input the multi-scale fourth features of the road area environment during the construction period of highway engineering into the multi-scale feature difference quantity encoding module to obtain the multi-scale feature difference quantity of the road area environment during the construction period of highway engineering; Input the multi-scale feature difference quantity of the road area environment during the construction period of highway engineering into the change decoder to obtain the final change type of the road area environment during the construction period of highway engineering; Generate the change annotation atlas of the road area environment during the highway engineering construction period and the first change event log of the road area environment during the highway engineering construction period based on the multi-scale feature difference quantity of the road area environment during the highway engineering construction period and the final change type of the road area environment during the highway engineering construction period.

7. A method for evaluating the safety of the road area environment during the construction period of a highway project according to claim 6, characterized in that, The steps to obtain the multi-scale feature difference quantity of the road area environment during the highway engineering construction period include: The multi-scale feature difference quantity coding module adopts a sliding window mechanism and a semantic residual mechanism; Use the sliding window mechanism for the multi-scale fourth feature of the road area environment during the highway engineering construction period to obtain the feature difference quantity of the road area environment during the highway engineering construction period; Use the semantic residual mechanism for the multi-scale fourth feature of the road area environment during the highway engineering construction period to obtain the high-order feature difference quantity of the road area environment during the highway engineering construction period; Based on the multi-scale fourth feature of the road area environment during the highway engineering construction period, the feature difference quantity of the road area environment during the highway engineering construction period, and the high-order feature difference quantity of the road area environment during the highway engineering construction period, obtain the multi-scale feature difference quantity of the road area environment during the highway engineering construction period.

8. A method for evaluating the safety of the road area environment during the construction period of highway engineering according to claim 7, characterized in that The steps to obtain the final change type of the road area environment during the highway engineering construction period include: The change decoder includes two layers of Transformer decoders and one layer of fully connected layer; Input the multi-scale feature difference quantity of the road area environment during the highway engineering construction period into the two layers of Transformer decoders of the change decoder to obtain the first semantic feature of the road area environment during the highway engineering construction period; Input the first semantic feature of the road area environment during the highway engineering construction period into the fully connected layer of the change decoder to obtain the change classification probability distribution of the road area environment during the highway engineering construction period; Adopt the maximum probability term for the change classification probability distribution of the road area environment during the highway engineering construction period to obtain the final change type of the road area environment during the highway engineering construction period.

9. A method for evaluating the safety of the road area environment during the construction period of highway engineering according to claim 8, characterized in that, The steps to obtain the second change event log of the road area environment during the highway engineering construction period include: The risk warning module of the road area environment during the highway engineering construction period consists of two major modules: a clustering analysis module and a safety level evaluation module of the road area environment during the highway engineering construction period; Input the change annotation atlas of the road area environment during the highway engineering construction period and the first change event log of the road area environment during the highway engineering construction period into the clustering analysis module to generate the first risk assessment value of the road area environment during the highway engineering construction period; Based on the first risk assessment value of the road area environment during the highway engineering construction period, establish a safety level evaluation module of the road area environment during the highway engineering construction period; Based on the change annotation atlas of the road area environment during the highway engineering construction period, the first risk assessment value of the road area environment during the highway engineering construction period, and the safety level evaluation module of the road area environment during the highway engineering construction period, generate the second change event log of the road area environment during the highway engineering construction period.

10. A road area environmental safety evaluation system during the construction period of a highway project, which is used to implement the road area environmental safety evaluation method described in any one of claims 1-9, and is characterized in that, The system includes: Data acquisition module: used to obtain the macro-scale satellite sequence data of the road area environment during the highway engineering construction period by using satellite remote sensing equipment and the micro-scale drone sequence data of the road area environment during the highway engineering construction period by using drone-mounted equipment; Data preprocessing module: used to perform preprocessing methods based on the macro-scale satellite sequence data of the road area environment during the highway engineering construction period and the micro-scale drone sequence data of the road area environment during the highway engineering construction period; Data Feature Encoding Module: It is used to input the multi-source perception data of the road area environment during the construction period of highway engineering into the multi-scale time perception encoding module to obtain the macro-scale satellite sequence features and the micro-scale drone sequence features; Cross-modal Fusion Module: It is used to input the macro-scale satellite sequence features and the micro-scale drone sequence features into the multi-modal fusion Transformer module to obtain the multi-scale fourth features of the road area environment during the construction period of highway engineering; Intelligent Change Detection Module: It is used to input the multi-scale fourth features of the road area environment during the construction period of highway engineering into the change detection module to obtain the change annotation map of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering; Risk Warning Decision Module: It is used to construct the risk warning module of the road area environment during the construction period of highway engineering based on the change annotation map of the road area environment during the construction period of highway engineering and the first change event log of the road area environment during the construction period of highway engineering to obtain the second change event log of the road area environment during the construction period of highway engineering; Each of the above modules is executed in sequence to achieve a complete closed loop from environmental perception to trajectory prediction.

Images