A multi-source remote sensing data fusion geological disaster intelligent interpretation method and system

Abstract

The application relates to the field of image analysis, and discloses a multi-source remote sensing data fusion geological disaster intelligent interpretation method and system. The method comprises the following steps: acquiring multi-temporal unmanned aerial vehicle images, laser radar point clouds and optical satellite images; fusing the same phase data to generate a first fused terrain model of fused real terrain and high-resolution texture; performing time sequence alignment on the fused models of different phases and satellite images to construct a multi-temporal three-dimensional terrain model sequence; extracting a deformation area and an evolution boundary through elevation difference based on the sequence; calculating the ratio of the cumulative deformation area of each phase to the total collapse and slide area to form a collapse and slide area ratio sequence; and automatically determining a disaster evolution stage and generating early warning information according to the comparison between the sequence and a preset creep threshold. The application realizes the automation and quantitative interpretation of geological disasters from multi-source data fusion, dynamic process inversion to intelligent early warning.

Description

Technical Field

[0001] This application relates to the field of image analysis, and in particular to a method and system for intelligent interpretation of geological disasters by fusing multi-source remote sensing data. Background Technology

[0002] Accurate monitoring and early warning of geological disasters are crucial for disaster prevention and mitigation. Currently, remote sensing technology has become an important means of large-scale, non-contact geological disaster investigation. However, existing technologies still have significant shortcomings in achieving intelligent interpretation of disaster bodies, especially in accurately retrospectively analyzing and warning of the dynamic evolution of disasters in vegetated areas.

[0003] First, the monitoring data sources are limited, with most methods relying on single types of optical remote sensing imagery or lidar data. While optical imagery offers rich textures, it struggles to penetrate dense vegetation to capture accurate topography, significantly reducing its monitoring effectiveness in mountainous areas. Lidar, though capable of acquiring topography beneath vegetation, lacks corresponding surface spectral texture information, and its data acquisition costs are high with limited coverage frequency. This limitation leads to blind spots or biases in the understanding of the spatial morphology and structure of disaster bodies. Second, the multi-source data fusion is superficial. Existing technologies largely remain at the level of simple overlaying or visual comparison of data from different sources, failing to achieve deep data-level fusion at the pixel or feature level. This prevents the generation of a unified three-dimensional data base that simultaneously possesses high-precision accurate topography and high-resolution surface texture, thus restricting the accuracy of subsequent quantitative deformation analysis. Third, the dynamic process monitoring capability is weak. Traditional methods are limited by the data acquisition cycle, often only obtaining static snapshots of individual time points. They cannot construct long-term series and three-dimensional observation sequences with unified spatial benchmarks, making it difficult to depict the complete creep evolution process of disasters from gestation and development to outbreak, let alone make trend predictions. Finally, the entire interpretation process has a low degree of automation and intelligence. From deformation identification to evolution stage judgment, it relies heavily on the experience of professionals for visual interpretation and subjective classification, which is inefficient and inconsistent, making it difficult to meet the practical needs of rapid, batch, and objective early warning. Summary of the Invention

[0004] In order to overcome the above-mentioned deficiencies of the prior art, the embodiments of this application provide a method for intelligent interpretation of geological disasters by fusing multi-source remote sensing data to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this application provides a method for intelligent interpretation of geological hazards through multi-source remote sensing data fusion, comprising:

[0006] Acquire UAV aerial survey images, airborne lidar point cloud data, and multi-temporal optical satellite images of the study area at least two time phases;

[0007] Spatial registration and data-level fusion are performed on the UAV aerial survey images and the airborne lidar point cloud data of the same time phase to generate a first fused terrain model of the corresponding time phase.

[0008] The first fused terrain model from different time phases is time-series aligned with the multi-temporal optical satellite imagery to generate a multi-temporal three-dimensional terrain model sequence for the study area.

[0009] Based on the multi-temporal three-dimensional terrain model sequence, the surface deformation region is extracted by elevation difference calculation, and the deformation evolution boundary of the surface deformation region is identified.

[0010] Based on the deformation evolution boundary, the cumulative deformation area and the total landslide area of ​​each time phase are calculated, and then the landslide area ratio sequence reflecting the disaster evolution process is obtained.

[0011] Based on the landslide area ratio sequence and the preset creep stage threshold, when the landslide area ratio exceeds the corresponding creep stage threshold, the disaster is determined to have entered the corresponding evolution stage, and an early warning message containing the stage determination result is generated.

[0012] To address the aforementioned problems, this application also provides a geological hazard intelligent interpretation system based on multi-source remote sensing data fusion, the system comprising:

[0013] The data acquisition and preprocessing module is used to acquire UAV aerial survey images, airborne lidar point cloud data, and multi-temporal optical satellite images of the study area at least two time phases.

[0014] The multi-source data fusion modeling module is used to spatially register and data-level fuse the UAV aerial survey images and the airborne lidar point cloud data of the same time phase to generate a first fused terrain model of the corresponding time phase.

[0015] The temporal 3D reconstruction and enhancement module is used to temporally align the first fused terrain model with the multi-temporal optical satellite imagery at different time phases to generate a multi-temporal 3D terrain model sequence for the study area.

[0016] The automatic surface deformation extraction module is used to extract surface deformation areas based on the multi-temporal three-dimensional terrain model sequence through elevation difference calculation, and to identify the deformation evolution boundary of the surface deformation area.

[0017] The evolution process quantification analysis module is used to calculate the cumulative deformation area and the total area of ​​the landslide area at each time phase based on the deformation evolution boundary, and then calculate the landslide area ratio sequence that reflects the evolution process of the disaster.

[0018] The intelligent early warning decision output module is used to make a judgment based on the landslide area ratio sequence and the preset creep stage threshold. When the landslide area ratio exceeds the corresponding creep stage threshold, the disaster is determined to have entered the corresponding evolution stage, and an early warning information containing the stage judgment result is generated.

[0019] Compared with the prior art, this application has the following beneficial effects:

[0020] This application effectively overcomes the limitations of single data by systematically coordinating UAV imagery, airborne LiDAR point clouds, and multi-temporal satellite imagery, and employing an innovative data-level fusion strategy. Specifically, it utilizes the vegetation-penetrating properties of LiDAR to obtain a true digital elevation model, which is then deeply fused with a high-resolution digital surface model generated by the UAV and orthophotos based on an inverse distance weighting algorithm to generate a first fused terrain model. This model retains the fine texture details of the imagery in exposed areas, while accurately restoring the true terrain that is obscured in vegetation-covered areas. This fundamentally solves the bottleneck of difficulty in obtaining true deformation information under vegetation interference, providing a reliable three-dimensional data foundation for subsequent high-precision analysis.

[0021] Building upon the high-quality single-temporal fusion model, this application further utilizes temporal alignment technology to combine the limited fusion model with long-term satellite imagery, constructing a multi-temporal 3D terrain model sequence with unified spatiotemporal references. This enables continuous 3D digital representation of the monitoring area over a long period, supporting dynamic retrospective analysis of disaster evolution. Based on this sequence, deformation boundaries are extracted through automated elevation difference and threshold segmentation. Furthermore, the application creatively proposes the normalized quantitative indicator "landslide area ratio" and its temporal sequence, transforming the complex spatial deformation expansion process into a concise... With a clear time-series evolution curve, and by automatically comparing this curve with preset creep stage thresholds, the application achieves fully automated, standardized intelligent identification and early warning information output for the three stages of disasters: primary creep, steady-state creep, and accelerated creep (prone to slippage). This application has established a complete technical chain from multi-source data acquisition, deep fusion, time-series modeling, quantitative deformation extraction to intelligent early warning decision-making, realizing a fundamental transformation of geological disaster interpretation from the traditional manual experience model to a fully automated, quantitative, and intelligent model, significantly improving the timeliness, objectivity, and reliability of monitoring and early warning. Attached Figure Description

[0022] Figure 1 A flowchart illustrating a method for intelligent interpretation of geological hazards based on multi-source remote sensing data fusion, provided in an embodiment of this application;

[0023] Figure 2 A functional block diagram of a geological disaster intelligent interpretation system based on multi-source remote sensing data fusion provided in an embodiment of this application;

[0024] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0025] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0026] This application provides a method for intelligent interpretation of geological hazards based on multi-source remote sensing data fusion. The executing entity of this method includes, but is not limited to, at least one electronic device that can be configured to execute the method provided in this application, such as a server or a terminal. In other words, the method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, and big data and artificial intelligence platforms.

[0027] Reference Figure 1 The diagram shown is a flowchart illustrating a method for intelligent interpretation of geological hazards based on multi-source remote sensing data fusion, according to an embodiment of this application. In this embodiment, the method includes:

[0028] S1. Acquire UAV aerial survey images, airborne lidar point cloud data, and multi-temporal optical satellite images of the study area at least two time phases.

[0029] In some embodiments, acquiring UAV aerial survey images, airborne lidar point cloud data, and multi-temporal optical satellite images of the study area at least two temporal phases specifically includes:

[0030] Ground control points were deployed within the study area;

[0031] Control the drone equipped with an oblique photography camera to conduct aerial photography of the study area according to the preset flight altitude, heading overlap rate and lateral overlap rate, and acquire the drone aerial survey images;

[0032] The drone equipped with lidar is controlled to perform aerial scanning of the study area to acquire raw point cloud data;

[0033] The original point cloud data is filtered, registered, and ground point extracted to generate the airborne lidar point cloud data with non-ground points removed.

[0034] Obtain multi-temporal optical satellite images from a preset satellite image database that cover the study area and span at least two time phases.

[0035] In this embodiment of the application, step S1 aims to systematically acquire complementary multi-source remote sensing data for subsequent intelligent interpretation. It directly addresses the shortcomings of data singularity and the severe vegetation interference in some technologies. This step is the foundation for all subsequent data fusion and intelligent analysis, and the quality and characteristics of the acquired data directly determine the reliability and accuracy of the final interpretation.

[0036] In this embodiment, the "temporal phase" refers to a specific point in time when the data is collected, and data from at least two temporal phases form the basis for time-series analysis for dynamic change detection and process backtracking. The "UAV aerial survey imagery" is a series of high spatial resolution digital images obtained by taking pictures of the ground surface from the air using an oblique photography camera mounted on a UAV platform. These images record the rich texture and spectral information of the ground surface. The "airborne lidar point cloud data" is a dense dataset composed of massive three-dimensional spatial coordinate points directly obtained by a lidar scanning system mounted on a UAV, which emits laser pulses to the ground surface and receives their echoes. The active detection characteristics of lidar enable it to partially penetrate the vegetation canopy and obtain the three-dimensional information of the real ground surface or objects under the vegetation that are covered by vegetation. The "multi-temporal optical satellite imagery" refers to multiple image data covering the same geographical area acquired by optical remote sensing satellites at different time points. These data provide a long-term historical observation record of the study area.

[0037] In this embodiment, the specific implementation of deploying ground control points within the study area is as follows: Personnel must select stable, easily identifiable locations with good visibility within the study area. Using high-precision global navigation satellite system receivers, such as real-time dynamic measurement equipment, they must accurately determine the three-dimensional geographic coordinates of these locations. These ground control points provide an absolute benchmark for the geometric accuracy correction and spatial coordinate unification of all subsequent aerial remote sensing data. For example, in a monitoring mission targeting a mining slope, no fewer than ten such control points will be deployed at the top, middle, and surrounding stable areas of the slope.

[0038] In this embodiment, the specific technical means for controlling a drone equipped with an oblique photography camera to perform aerial photography according to preset parameters are as follows. First, a key parameter is determined based on the requirements of the final result for ground detail resolution, namely the ground sampling distance. The ground sampling distance represents the actual ground size corresponding to a pixel in the image. The smaller the value, the higher the resolution. To achieve centimeter-level high-precision modeling, the ground sampling distance can be set to 0.05 meters. Then, the relative flight altitude of the drone is calculated according to the formula "relative flight altitude equals camera focal length multiplied by ground sampling distance divided by camera pixel size". Camera focal length and pixel size are inherent physical parameters of the camera. By calculating these, a specific flight altitude value can be obtained to ensure imaging accuracy. Then, the flight path is planned to ensure the overlap between adjacent photos. The forward overlap rate refers to the overlap ratio of preceding and subsequent images along the flight direction, and the lateral overlap rate refers to the overlap ratio of images between adjacent flight paths. To ensure the quality of subsequent 3D model reconstruction, both the forward and lateral overlap rates are usually preset to a high overlap rate of no less than 70%. Finally, the UAV is controlled to automatically execute the flight mission according to the set altitude, flight path, and overlap rate to collect high-overlap, high-resolution image sequences covering the entire study area, i.e., the UAV aerial survey images. This series of operations ensures that the acquired image data itself has the potential to generate a high-precision 3D model.

[0039] In this embodiment, the specific technical means for controlling a drone equipped with a lidar to perform aerial scanning to obtain raw point cloud data is as follows. During operation, another drone, or the same drone that has been modified and adapted, is used, equipped with a lidar sensor. The lidar system emits laser beams at extremely high frequencies towards the ground during flight and accurately measures the round-trip time of each laser pulse, thereby calculating the distance between the sensor and the ground point. Combined with the high-precision position and attitude information recorded by the drone in real time, the system can directly calculate the three-dimensional spatial coordinates of each laser footpoint, forming raw point cloud data containing hundreds of millions or even billions of spatial points. This process can be carried out in conjunction with oblique photogrammetry aerial scanning or independently, but it is necessary to ensure that the flight range covers the target area.

[0040] In this embodiment, the specific technical means for filtering, registering, and extracting ground points from the original point cloud data to generate airborne lidar point cloud data are as follows: Since the original point cloud contains a large amount of noise, flyby points, and points from non-surface targets such as vegetation and buildings, it must be processed with precision. First, statistical filtering or radius filtering algorithms are used to remove obvious outlier noise points. Next, if the data is collected in multiple flight strips, point cloud registration algorithms such as the iterative nearest point algorithm are used to accurately align the point clouds of different flight strips to the same coordinate system. Finally, and most importantly, is ground point extraction. This embodiment employs a progressive triangulation filtering algorithm. This algorithm starts from initial sparse ground seed points, iteratively constructs a triangulation network, and sets angle and distance thresholds to determine whether surrounding new points belong to the ground, thereby effectively separating points representing the real ground from non-ground points such as vegetation and buildings. After this series of processing steps, the final output airborne lidar point cloud data is a pure set of three-dimensional points on the ground surface, free from interference such as vegetation. This lays a solid foundation for generating a true digital elevation model. This step directly solves the key bottleneck problem that optical imagery and traditional photogrammetry cannot obtain true terrain data in densely vegetated areas.

[0041] In this embodiment, the specific technical means for acquiring multi-temporal optical satellite images from a preset satellite image database are as follows. The preset satellite image database can be a commercial satellite data distribution platform or a public remote sensing data acquisition platform. During operation, the database is searched and filtered based on the geographical boundary coordinates of the study area and the required time range (this range must cover the temporal phases of the acquired UAV and lidar data, and trace back as far as possible). Filtering conditions include spatial resolution, cloud cover, and shooting date. For example, to complement the analysis of UAV data with a ground sampling distance of 0.05 meters, sub-meter level commercial satellite images (such as WorldView and GeoEye series) with a spatial resolution better than 1 meter, or publicly available medium-to-high resolution images such as Sentinel-2, are selected. Finally, a series of temporally ordered satellite image products that have undergone radiometric and system-level geometric corrections are downloaded, constituting the multi-temporal optical satellite image dataset. These data provide long-term continuous observation capabilities.

[0042] In this embodiment, the technical effect of step S1 is that it systematically collects three complementary raw data sources, creating the necessary conditions for subsequent fusion and interpretation. Specifically, UAV aerial survey imagery provides high-resolution surface texture; airborne lidar point cloud data provides real three-dimensional terrain information penetrating vegetation; and multi-temporal optical satellite imagery provides long-term historical observation records covering the disaster development process. This initially solves the problems of one-sided information from a single data source, lack of terrain information in vegetation-covered areas, and scarcity of long-term dynamic data in the background technology.

[0043] S2. Spatial registration and data-level fusion of the UAV aerial survey imagery and the airborne lidar point cloud data of the same time phase are performed to generate a first fused terrain model corresponding to the time phase.

[0044] In some embodiments, the step of spatially registering and data-level fusing the UAV aerial survey imagery and the airborne lidar point cloud data of the same temporal phase to generate a first fused terrain model corresponding to that temporal phase specifically includes:

[0045] Based on the UAV aerial survey images of the same time phase, a digital orthophoto model and a digital ground model of the study area are generated;

[0046] A digital elevation model of the study area is generated based on the airborne lidar point cloud data of the same time phase.

[0047] Register the digital orthophoto model, the digital terrain model, and the digital elevation model to the same geographic coordinate system;

[0048] The registered digital terrain model and the digital elevation model are fused at the data level to generate the first fused terrain model corresponding to the time phase.

[0049] In some embodiments, the step of performing data-level fusion of the registered digital ground model and the digital elevation model to generate the first fused terrain model of the corresponding time phase specifically includes: using an inverse distance weighted interpolation algorithm to perform fusion calculation on the registered digital ground model and the digital elevation model to generate the first fused terrain model.

[0050] In this embodiment, step S2 is the core step in solving the problems of multi-source data fusion and vegetation interference. Its purpose is to deeply integrate UAV image data and LiDAR point cloud data with different information characteristics acquired at the same time point to generate a single, high-quality three-dimensional model with both high-resolution texture and real terrain elevation, namely the first fused terrain model. This step directly addresses the two defects of data uniformity and severe interference from vegetation cover in the background technology. Through creative data-level fusion technology, it provides a unique and reliable high-precision three-dimensional benchmark for subsequent accurate deformation analysis and dynamic evolution interpretation.

[0051] In this embodiment, the digital orthophoto model is an image map that has undergone geometric correction to eliminate projection differences caused by terrain undulations and sensor tilt. It has accurate geographic coordinates and is represented as a two-dimensional planar map with true surface color and texture information. The digital ground model is a three-dimensional model that describes the undulations of the top surfaces of all objects on the ground. It includes the height information of all above-ground elements such as vegetation and buildings. The digital elevation model is a three-dimensional model that only reflects the topographic undulations of bare ground. It removes non-surface elements such as vegetation and buildings through processing, retaining only the elevation information of the true ground. The data-level fusion refers to the process of integrating data from different sources at the level of raw observation data or basic geographic information products that have undergone preliminary processing, in order to generate a new dataset with more complete information and higher quality.

[0052] In this embodiment of the application, the first fused terrain model is a novel three-dimensional terrain representation generated by data-level fusion. It integrates the real texture from the digital orthophoto model, the rich details from the digital ground model, and the real terrain from the digital elevation model to form a unified, high-fidelity three-dimensional scene.

[0053] In this embodiment, the specific technical means for generating digital orthophoto models and digital ground models based on UAV aerial survey images of the same time phase are as follows. This step is crucial for achieving high-precision modeling. The UAV aerial survey images (a set of highly overlapping photos) obtained in step S1 are imported into professional 3D reconstruction software, such as ContextCapture. The software first uses the feature points contained in the images for automatic matching. Through aerial triangulation calculations, it calculates the precise position and attitude parameters of each image in space and constructs a sparse point cloud. This process relies on the high heading and lateral overlap rates preset in step S1. The high overlap rate ensures... The system ensures the sufficiency and reliability of feature matching. Next, based on the calculated precise parameters, the software performs dense matching on the image to generate a dense point cloud. Then, it automatically constructs a triangular mesh surface model based on the dense point cloud. This surface model reflects the top morphology of all ground features, including vegetation and buildings, and is the geometric prototype of the digital terrain model. Finally, the system uses the original UAV image as a texture and accurately maps it onto the surface of this triangular mesh model, while simultaneously generating a two-dimensional orthophoto image with the correct spatial position, i.e., a digital orthophoto model. The textured three-dimensional triangular mesh model itself, after output processing, becomes a usable digital terrain model.

[0054] For example, aerial survey data of a mining slope can be processed in this way to obtain a realistic 3D model (digital terrain model) that shows the texture of every rock and every clump of vegetation on the slope, as well as a slope plan map (digital orthophoto model) that can accurately measure distances.

[0055] In this embodiment, the specific technical means for generating a digital elevation model based on airborne lidar point cloud data from the same time phase is as follows. This step is crucial for achieving the acquisition of realistic terrain through vegetation. The airborne lidar point cloud data obtained in step S1, after noise removal and ground point extraction, is used as input. These ground point clouds consist of a large number of discrete points with three-dimensional coordinates. To generate a continuous terrain surface, a spatial interpolation algorithm is required. A common and effective method is the triangulation network construction algorithm. This algorithm uses these discrete ground points as vertices and connects them into a non-overlapping triangular patch network according to the empty circumcircle criterion (i.e., the Delaunay criterion). This triangulation network directly expresses the irregular shape of the ground surface and constitutes the geometric basis of the digital elevation model. Since the input point cloud data has undergone rigorous filtering and ground point extraction, the terrain expressed by the constructed triangulation network is a "bare surface" terrain free from the interference of trees and houses.

[0056] For example, for the same mining slope covered by trees, the digital elevation model generated by this method will clearly show the shape of the real slope, gullies and even cracks hidden under the tree canopy, rather than a green shell model of a vegetation surface.

[0057] In this embodiment, the specific technical means for registering the digital orthophoto model, digital terrain model, and digital elevation model to the same geographic coordinate system is as follows: This step is the foundation for achieving spatial consistency of multi-source data. During the generation process, the spatial coordinate systems of the digital orthophoto model and the digital terrain model have been absolutely oriented using aerial triangulation and the ground control points established in step S1, and unified to, for example, the WGS84 coordinate system. The digital elevation model is generated from lidar point clouds, and the lidar point clouds have also been unified to the same WGS84 coordinate system using high-precision integrated navigation system data and ground control points during data preprocessing. To ensure absolute accuracy, a precise spatial registration check is still required before fusion. In practice, the digital orthophoto model and the digital elevation model can be loaded simultaneously in the geographic information system software. The process involves creating a digital elevation model (usually in raster form). This is achieved by manually or automatically matching clearly identifiable and stable corresponding feature points on both images, such as road intersections or the corners of large buildings. Using these matched point pairs, a small translation, rotation, or affine transformation parameter is calculated to precisely calibrate the digital elevation model (DEM) and the digital orthophoto model. Since the DEM and DORphoto model originate from the same processing flow and are naturally spatially aligned, it is only necessary to ensure that the DEM is aligned with them. Through this step, the three models achieve pixel-level or sub-pixel-level spatial alignment within the same geographic coordinate framework, enabling precise correlation between texture information (from the DORphoto model), top elevation (from the DEM), and true ground elevation (from the DEM) at the same geographic location.

[0058] In this embodiment, the specific technical means of performing data-level fusion of the registered digital ground model and digital elevation model to generate the first fused terrain model is to use an inverse distance weighted interpolation algorithm. This step is a creative core data processing step, which determines the quality of the fused model. The digital ground model has high spatial resolution and can finely describe the top morphology of surface objects (including vegetation), but its elevation value is significantly higher than that of the real ground in vegetation-covered areas. The digital elevation model reflects the real ground elevation, but its data source (laser point cloud) may have relatively low point density in bare areas without vegetation, resulting in terrain details that are sometimes not as rich as those of the digital ground model generated by dense image matching.

[0059] In this embodiment, the goal of the inverse distance weighted interpolation algorithm is to combine the advantages of both. The specific implementation process is as follows: First, the registered digital terrain model and digital elevation model are resampled to the same regular grid size. Each grid point has an elevation value. For each target grid point P in the final fusion model, the algorithm will simultaneously search for source elevation data points from the digital terrain model and digital elevation model within a certain neighborhood. The elevation contribution weight of each source data point to the target point P is inversely proportional to the p-th power of the distance from the source point to point P (p is usually taken as 2). This means that the closer the source data point is to point P, the greater its elevation value has an impact on the final fused elevation value of point P. The key application logic is as follows: In areas with buildings or exposed rock, the data points of the digital terrain model (derived from dense image matching) are typically denser and closer to the target point P, thus carrying higher weight in the calculation. This allows the fusion result to inherit the rich detail of the digital terrain model. In densely vegetated areas, the data points of the digital terrain model reflect the elevation of the treetops, while the data points of the digital elevation model (derived from LiDAR ground points), although sparser, reflect the obscured true ground. Because the algorithm treats all data points equally by assigning weights based on distance, those sparse but closer to the true ground will have a dominant influence on the fusion result, thereby pulling the elevation values ​​back to the true ground position. By traversing and calculating each target grid point, a new, unified elevation matrix is ​​finally generated. The 3D terrain model represented by this matrix is ​​the first fused terrain model, which maintains extremely high detail resolution in exposed areas and reveals the obscured true terrain in vegetated areas.

[0060] For example, in the fusion model of the mining area slope, the exposed rock blocks on the slope have clear textures and sharp shapes, while the forested areas on the slope, the real slope undulations below, and even potential cracks and gullies can be accurately presented.

[0061] In this embodiment, the technical effect of step S2 is to fundamentally improve the quality and reliability of the three-dimensional terrain data used for disaster interpretation. The first fused terrain model it creates successfully integrates the rich texture details of optical images with the ability of lidar to penetrate vegetation and obtain real terrain, overcoming the limitations of a single data source. In particular, it effectively eliminates the serious interference of vegetation cover on terrain information acquisition, providing an irreplaceable high-precision data foundation for subsequent accurate identification of deformation boundaries and calculation of displacement.

[0062] S3. Align the first fused terrain model from different time phases with the multi-temporal optical satellite imagery to generate a multi-temporal three-dimensional terrain model sequence for the study area.

[0063] In some embodiments, the step of temporally aligning the first fused terrain model from different time phases with the multi-temporal optical satellite imagery to generate a multi-temporal three-dimensional terrain model sequence for the study area specifically includes:

[0064] Acquire the first fused terrain model generated at different time phases, and the multi-temporal optical satellite imagery;

[0065] Based on the acquisition time of the first fused terrain model, the corresponding images are selected from the multi-temporal optical satellite images as matching images;

[0066] Using the first fused terrain model as a spatial reference, the matching image is spatially registered with it;

[0067] Based on the registered optical satellite imagery, the surface texture of the first fused terrain model is enhanced to generate three-dimensional terrain models of each time phase with temporal texture information.

[0068] The three-dimensional terrain models generated at each time phase are arranged according to their corresponding time sequence to form a multi-temporal three-dimensional terrain model sequence for the study area.

[0069] In this embodiment, step S3 aims to address the core problems in the background technology, namely, the difficulty in real-time dynamic monitoring and the inadequacy in observing long-term dynamic change characteristics and tracing historical deformation trends. This step creatively combines a high-precision but costly and temporally sparse first fused terrain model with easily acquired multi-temporal optical satellite imagery with long-term historical archives. Through temporal alignment and texture enhancement techniques, a unified three-dimensional observation sequence with both high geometric accuracy and long time span is constructed.

[0070] In this embodiment, the temporal alignment refers to arranging data from different sensors and acquired at different times in the order of their acquisition time and performing precise spatial matching so that they can represent the continuous state of the same geographical area on the time axis; the multi-temporal three-dimensional terrain model sequence is an ordered collection of multiple three-dimensional terrain models organized in chronological order, and each model in the sequence represents the three-dimensional state of the study area at a specific time point.

[0071] In this embodiment, the specific technical means for acquiring the first fused terrain model and multi-temporal optical satellite images generated at different times are as follows. This step is the data preparation stage. The first fused terrain model comes directly from the output of step S2. For example, for the study area, step S2 has generated a first fused terrain model Fusion_T1 representing the "August 2019" time phase and a first fused terrain model Fusion_T2 representing the "May 2022" time phase. The multi-temporal optical satellite images come from the dataset acquired in step S1. This dataset not only includes images close to Fusion_T1 and Fusion_T2 in time, but also images from earlier historical periods, such as sub-meter resolution satellite images from 2015 and 2017. The system or operator retrieves these pre-processed model files and image files from storage as input for this step.

[0072] In this embodiment, the specific technical means for selecting the corresponding image from multi-temporal optical satellite images as the matching image based on the acquisition time of the first fused terrain model is as follows. This step is key to achieving temporal logical correspondence. Each first fused terrain model has a precise metadata timestamp. The system traverses the multi-temporal optical satellite image library to find the satellite image that is closest in time for each fused model. The matching principle is temporal proximity, aiming to use the surface spectral and texture features recorded by the satellite image of that time phase to reflect the surface conditions at that point in time represented by the fused model and its adjacent periods.

[0073] For example, for Fusion_T1 (August 2019), the satellite imagery Sat_2019Summer acquired in the summer of 2019 is selected; for Fusion_T2 (May 2022), the satellite imagery Sat_2022Spring acquired in the spring of 2022 is selected. For earlier historical periods, although there is no corresponding high-precision fusion model, the satellite imagery itself (such as Sat_2015, Sat_2017) is also included in this process. They will be registered with a selected high-precision fusion model benchmark to construct a 3D model of the earlier time period.

[0074] In this embodiment of the application, the specific technical means of spatially registering the matching image with the first fused terrain model as a spatial reference is as follows: This step ensures that all data are strictly aligned in space. The first fused terrain model itself contains a high-precision digital orthophoto model as its texture reference. During operation, in remote sensing or GIS software, the digital orthophoto model of Fusion_T1 and the selected matching satellite image Sat_2019Summer are loaded simultaneously. Due to the possible slight systematic geometric errors or projection differences between the two, precise image-to-image registration is required. Specifically, multiple clear, stable, and unchanging corresponding feature points are manually or automatically selected on the two images, such as road intersections, corners of large buildings, and stable rock outcrops. Using these point pairs, a polynomial transformation model (such as a quadratic polynomial) is calculated to spatially and geometrically correct the satellite image Sat_2019Summer to a position completely consistent with the digital orthophoto model of Fusion_T1. This process ensures that every pixel on the satellite image can be accurately mapped to the corresponding 3D position of the first fused terrain model. For historical satellite images (such as Sat_2015), the one with better quality between Fusion_T1 and Fusion_T2 is selected as the unified spatial reference for registration, thereby ensuring that the data of all time phases are under the same high-precision spatial reference frame.

[0075] In this embodiment, the specific technical means for enhancing the surface texture of the first fused terrain model based on registered optical satellite imagery to generate 3D terrain models with temporal texture information for each phase are as follows: The first fused terrain model itself already possesses high-resolution textures from UAV imagery. The registered satellite imagery provides texture information for the same area at similar time phases, which may have different spectral characteristics (such as seasonal changes in vegetation) or record newly occurring surface changes (such as the appearance of cracks or vegetation destruction). Texture enhancement is not a simple replacement, but a fusion or supplement; one implementation method is texture mosaicking and harmonization, using the registered satellite imagery as a new texture layer and fusing it with the original texture of the model. In areas where no changes have occurred and the UAV texture is clearer, the original texture is retained; in areas where the satellite imagery shows significant signs of change (such as the dark linear features of new cracks) but the UAV texture is not clearly captured due to resolution or lighting reasons, the texture features of the satellite imagery are enhanced or integrated; another method is the reconstruction of historical periods; for early phases such as 2015, although there is no corresponding first fused terrain model, we can utilize the Sat_2015 imagery that has been registered with the baseline model. The specific approach involves using the geometric framework (i.e., its 3D mesh) of a baseline fusion model (such as Fusion_T1) as a basis, and mapping the Sat_2015 imagery as texture onto this unchanging geometric framework. This generates a 3D terrain model whose geometric framework is based on high-precision data from the T1 time phase, but whose surface texture reflects the surface conditions in 2015. In this way, a 3D terrain model is generated for each key time phase (whether it is a time phase with a high-precision fusion model or a time phase with only satellite imagery). These models share a consistent high-precision geometric framework but possess surface texture information specific to their respective time phases, thus forming a geometrically stable and texture-evolving model sequence. For example, the generated 2015 model can show the early vegetation distribution and scattered cracks on the slope; the 2019 model can simultaneously display high-precision terrain details and the vegetation and crack conditions at that time; and the 2022 model clearly presents the broken surface texture after the landslide.

[0076] In this embodiment of the application, the specific technical means of arranging the three-dimensional terrain models generated in each time phase according to their corresponding time sequence to form a multi-temporal three-dimensional terrain model sequence is as follows: This step is the organization and output of the results. All generated three-dimensional terrain model files with clear time phase labels are stored in an ordered list or database structure according to their representation time from early to late. This sequence is the multi-temporal three-dimensional terrain model sequence. In the software platform, this sequence can be played sequentially, thereby realizing the dynamic three-dimensional visualization and retrospective of the changes in the surface morphology and texture of the study area over time, and intuitively displaying the development process of disasters.

[0077] In this embodiment, the technical effect of step S3 is to extend the time dimension of high-precision three-dimensional terrain monitoring. It cleverly uses a high-precision fusion model as a geometric skeleton and multi-temporal satellite imagery as a skin for recording historical changes, constructing a three-dimensional observation sequence with consistent spatial accuracy spanning several years. This directly overcomes the inherent limitations of UAVs and airborne radars due to high monitoring costs and sparse temporal phases, realizing continuous three-dimensional digital archiving of the long-term process of geological disaster formation and development, and providing the only possible data carrier for quantitatively tracing back the evolution history.

[0078] S4. Based on the multi-temporal three-dimensional terrain model sequence, extract the surface deformation region through elevation difference calculation and identify the deformation evolution boundary of the surface deformation region.

[0079] In some embodiments, the step of extracting surface deformation regions based on the multi-temporal three-dimensional terrain model sequence through elevation difference calculation and identifying the deformation evolution boundaries of the surface deformation regions specifically includes:

[0080] From the multi-temporal three-dimensional terrain model sequence, select at least two temporal three-dimensional terrain models in chronological order, one as the baseline model and the other as the comparison model;

[0081] Elevation difference calculations are performed on the benchmark model and the comparison model to obtain the surface elevation change displacement field of the study area;

[0082] Based on a preset elevation change threshold, regions whose elevation change exceeds the preset displacement threshold are extracted from the elevation change displacement field and used as the surface deformation regions.

[0083] Based on the extracted boundaries of the surface deformation regions, the deformation evolution boundaries of the surface deformation regions are identified and determined.

[0084] In this embodiment, step S4 is a crucial transformation step from static 3D model to dynamic deformation information extraction. Its core objective is to automatically and quantitatively detect the spatial location and extent of significant surface deformation from the temporal 3D observation sequence constructed in step S3, and to accurately characterize its boundaries. This step directly addresses the problems of low interpretation efficiency, reliance on manual visual interpretation, and difficulty in quantitatively characterizing deformation in the background technology. Through fully automated elevation difference calculation and thresholding analysis, it achieves objective, accurate, and batch identification of deformation areas, laying the foundation for subsequent quantification of the evolution process.

[0085] In this embodiment, the elevation difference calculation is a mathematical operation process that subtracts the elevation values ​​of two three-dimensional terrain models from different time phases at the same spatial location to obtain the elevation change at that location. The surface elevation change displacement field is a two-dimensional data layer, where the value of each pixel or grid point represents the vertical elevation change of the corresponding surface location between two specific time phases; positive values ​​usually indicate uplift, and negative values ​​indicate subsidence. The surface deformation region refers to the continuous or discontinuous set of all spatial pixels in the surface elevation change displacement field whose absolute value of elevation change exceeds a certain preset significance threshold. The deformation evolution boundary refers to the closed geometric outline surrounding a surface deformation region, which defines the spatial distribution shape of the deformation influence range during that period.

[0086] In this embodiment, the specific technical means for selecting at least two temporal 3D terrain models from a multi-temporal 3D terrain model sequence in chronological order, and using one as the baseline model and the other as the comparison model, is as follows: This step is the decision-making stage for determining the time span of deformation analysis. The input is the ordered model sequence generated in step S3. Model selection must be based on the monitoring target. For example, to analyze the changes before and after a landslide event, the model before the event is selected as the baseline model, and the model after the event is selected as the comparison model. If analyzing long-term slow deformation, two models with a long time interval may be selected. The system automatically completes the selection based on the timestamp in the model metadata, according to user instructions or preset rules. For example, to analyze the deformation of the Faer landslide, the system automatically selects the model from May 2022 (after the landslide) as the comparison model and the model from August 2019 (before the landslide) as the baseline model. The baseline model represents the initial reference state for deformation analysis.

[0087] In this embodiment of the application, the specific technical means for performing elevation difference calculation on the benchmark model and the comparison model to obtain the surface elevation change displacement field of the study area is as follows. This step is the core calculation for realizing the quantification of deformation. First, ensure that the two models are in the same spatial coordinate system and the same grid resolution. The calculation process is a pixel-by-pixel algorithm operation. For each target pixel position in the displacement field, the algorithm reads the elevation value of the benchmark model at that position and reads the elevation value of the comparison model at the same position. Then, the latter is subtracted from the former to obtain the elevation change value of that pixel. This calculation traverses every pixel in the study area and finally generates a new raster dataset with the same spatial range as the original model, but each pixel value represents the elevation change amount, i.e., the surface elevation change displacement field. For example, at a pixel point at the rear edge of a landslide, the elevation of the benchmark model is value A, and the elevation of the comparison model is value B, and B is much smaller than A. Then, the calculated elevation change value of that point is a significant negative value, which intuitively reflects that a large subsidence has occurred at that point.

[0088] In this embodiment, the specific technical means for extracting areas with elevation changes exceeding a preset displacement threshold from the elevation change displacement field as surface deformation areas based on a preset elevation change threshold is as follows. This step is a key discrimination step in separating significant deformation areas from a continuous change field. Due to measurement errors, vegetation phenology changes, etc., small, non-structural elevation fluctuations can occur, so a threshold must be set to distinguish noise from actual deformation. The preset displacement threshold is a parameter combining empirical values ​​and accuracy analysis. Its setting is mainly based on two aspects: first, the elevation accuracy estimate of the terrain model after multi-source data fusion, such as the mean square error obtained based on ground control point checks; second, the significance of deformation of the monitored object. At a minimum, in practice, the threshold can be set to a multiple of the mean error value. For example, if the elevation accuracy of the fused terrain model is estimated to be about 0.1 meters, and significant deformation is defined as a change of at least 3 times the noise level, then the preset displacement threshold can be set to 0.3 meters. The extraction process is a raster condition filtering algorithm. The system traverses each pixel in the displacement field and determines whether the absolute value of its elevation change is greater than the preset displacement threshold. If it is greater, the pixel is marked as a deformed pixel; otherwise, it is marked as an undeformed pixel. The set of all pixels marked as deformed pixels constitutes the preliminary surface deformation area. This process automatically eliminates areas with minor changes and focuses on the range where substantial displacement has occurred.

[0089] In this embodiment, the specific technical means for identifying and determining the deformation evolution boundary based on the extracted surface deformation area is as follows: This step vectorizes and normalizes the spatial morphology of the deformation area. The deformation area obtained in the previous step is in the form of a raster, and its edges may be jagged. The boundary identification usually adopts edge detection or region growing contour extraction algorithms in image processing. One specific implementation is to first perform morphological filtering on the binarized deformation area raster, such as erosion followed by dilation, to smooth the boundary and remove small isolated noise patches. Then, the edge detection operator is used to traverse the raster image and automatically track the outer contour line of each continuous deformation patch. Another more direct way is to call the raster to vectorization algorithm. This algorithm can automatically aggregate continuous deformation pixels into polygonal surface features, and the boundary line of the polygon is the required deformation evolution boundary. The final output is one or more closed polygon vector files. Each polygon accurately delineates the deformation influence range identified in this analysis. For example, for a landslide, this step will automatically output a smooth closed boundary line depicting the planar projection range of the landslide.

[0090] In this embodiment of the application, the technical effect of step S4 is to realize the automation and quantification of surface deformation information extraction. It uses a precise three-dimensional model sequence, through rigorous mathematical calculation and objective threshold judgment, to automatically output the location, range and boundary of the deformation area. This completely changes the problem that traditional methods rely on manual visual delineation, which is inefficient and highly subjective. This provides a direct technical means for quickly and accurately assessing the scope and scale of disaster impact.

[0091] S5. Based on the deformation evolution boundary, calculate the cumulative deformation area and the total area of ​​the landslide area at each time phase, and then calculate the landslide area ratio sequence that reflects the disaster evolution process.

[0092] In some embodiments, the step of calculating the cumulative deformation area and the total landslide area at each time phase based on the deformation evolution boundary, and then calculating the landslide area ratio sequence reflecting the disaster evolution process, specifically includes:

[0093] Based on the multi-temporal three-dimensional terrain model sequence, the deformation evolution boundary corresponding to each time is obtained;

[0094] For each time phase, the cumulative deformation region area of ​​that time phase is calculated based on its corresponding deformation evolution boundary.

[0095] The landslide area with the largest coverage in the multi-temporal phase is identified, and the total area of ​​the landslide area is calculated using its deformation evolution boundary.

[0096] For each time phase, the ratio of the cumulative deformation area to the total area of ​​the collapse and slippage area in that time phase is calculated to obtain the collapse and slippage area ratio for that time phase;

[0097] Arrange the landslide area ratios of all time phases according to their corresponding time sequence to generate a landslide area ratio sequence that reflects the evolution of the disaster.

[0098] In some embodiments, determining the landslide area with the largest coverage in multiple time phases and calculating the total area of ​​the landslide area using its deformation evolution boundary specifically includes:

[0099] The boundary of the unstable slope region, which covers all cumulative deformation areas and is determined by the deformation evolution boundary based on all time phases throughout the entire monitoring period, is defined as the landslide area.

[0100] Calculate the area of ​​the unstable slope region as the total area of ​​the landslide area.

[0101] In the embodiments of this application, step S5 is the core creative step for realizing the intelligent quantitative interpretation of the disaster evolution process. Its purpose is to transform the boundary data extracted in step S4, which are scattered across various time phases and describe the spatial range of deformation, into a unified quantitative index sequence that characterizes the evolution of the disaster development process over time. This step directly addresses the fundamental problems in the background technology, such as the difficulty in observing long-term dynamic change characteristics and retrospectively tracking historical deformation trends, as well as the low degree of automation and quantification. By creatively defining and calculating the landslide area ratio, the complex spatial deformation process is compressed into a clear, measurable, and comparable time-series signal, providing a unique and crucial quantitative basis for the final intelligent stage identification.

[0102] In this embodiment, the cumulative deformation area refers to the total area enclosed by the deformation evolution boundaries of all surface deformation areas identified in a specific time phase; the total landslide area refers to the area corresponding to the maximum extent that the entire potentially unstable slope may eventually fail or has already failed during the entire monitoring period; the landslide area ratio sequence is a sequence arranged in chronological order, where each value is the landslide area ratio calculated for the corresponding time phase, and this sequence intuitively reflects the relative expansion of the disaster impact range over time.

[0103] In this embodiment of the application, the specific technical means for obtaining the deformation evolution boundary corresponding to each time based on the multi-temporal three-dimensional terrain model sequence is as follows: This step is data connection and input. The system directly calls the processing result of step S4. For each model in the multi-temporal three-dimensional terrain model sequence, step S4 has already output its corresponding deformation evolution boundary vector file. These boundary files correspond one-to-one with the model time phase and have timestamp attributes. By reading the timestamp attributes, the system automatically associates and matches the boundary files with the three-dimensional model sequence, thereby preparing the corresponding spatial range data for each analysis time phase. For example, the system loads the deformation evolution boundary polygon files for four time phases: 2015, 2017, 2019, and 2022. Each file accurately records the deformation area range monitored in that year.

[0104] In this embodiment, the specific technical means for calculating the cumulative deformation area of ​​each time phase based on its corresponding deformation evolution boundary is as follows: This step quantifies the deformation scale of a single time phase. The calculation is automatically completed in the GIS software environment. For a single time phase, its deformation evolution boundary may contain one or more polygons. The core of calculating the cumulative deformation area is to calculate the total area of ​​these polygons. Specifically, the GIS software reads all the boundary polygons of the time phase and directly calculates the planar projection area of ​​each polygon through its built-in geometry engine. If there are multiple non-adjacent polygons, the software will automatically sum the areas of all polygons. This calculation process is strictly based on a unified map projection coordinate system to ensure the accuracy of the area measurement results. The calculated area value is stored as an attribute field in the boundary file of the time phase. For example, for the boundary data of 2017, the system calculates the sum of the areas of all the polygons it contains to obtain the value A_2017, which is the cumulative deformation area in 2017.

[0105] In this embodiment, the specific technical means for determining the largest collapse-slip region in multiple temporal phases and calculating the total area of ​​the collapse-slip region using its deformation evolution boundary is as follows: This step is a key step in determining the quantification benchmark, and its purpose is to define an invariant denominator. The technical implementation involves spatial overlay analysis and geometric union operation. The system loads the deformation evolution boundary polygons of all temporal phases into the same layer, and then performs a geometric union operation. This operation merges all these polygons into one (or more) new polygons. The range of this new polygon covers the deformation regions that have appeared in all historical temporal phases, thus representing... The maximum potential or final failure range of the unstable slope during the entire monitoring period is defined as the boundary of the new polygon generated by the union operation. The system then calculates the area of ​​this new polygon, and the resulting value is the total area of ​​the landslide area. For example, the deformation boundaries of 2015, 2017, 2019, and 2022 are combined to obtain a complete large boundary covering the entire area from early sporadic deformation to the final landslide. The area of ​​this large boundary is calculated to obtain the value A_total. This A_total will serve as the unified benchmark for all subsequent ratio calculations.

[0106] In this embodiment of the application, the specific technical means for calculating the ratio of the cumulative deformation area to the total collapse area for each time phase to obtain the collapse area ratio for that time phase is as follows: This step is the calculation of generating core quantitative indicators. For each time phase, the system performs a division operation, dividing the cumulative deformation area (A_cumulative_i) of that time phase by the total collapse area (A_total). The calculation formula can be expressed as the collapse area ratio equals the cumulative deformation area divided by the total collapse area. This calculation is automatically completed by the program in a loop, traversing each time phase. The calculation result is a dimensionless value between 0 and 1. This value has a clear physical meaning. It indicates that in the current time phase, the deformation area has occupied the proportion of the final failure range. For example, to calculate the collapse area ratio R_2017 in 2017, that is, to divide A_2017 by A_total, we get a decimal less than 1, indicating that in 2017, the slope instability range had only developed to a few tens of percent of the final scale.

[0107] In this embodiment of the application, the specific technical means for arranging the landslide area ratios of all time phases according to their corresponding time sequence to generate a landslide area ratio sequence reflecting the evolution process of the disaster is as follows: This step is to organize the quantification results to form time series data. The system sorts the calculated landslide area ratio values ​​from early to late according to the timestamp of each time phase, forming an ordered array or list. This sequence is the landslide area ratio sequence. It can be stored as a table or visualized as a curve that changes over time. For example, the final generated sequence may be represented as [0.15, 0.32, 0.55, 0.60, 1.00], corresponding to the observation results in 2015, 2017, 2019, 2020, and 2022, respectively. This sequence curve clearly shows the process of the deformation range ratio accelerating over time until it reaches 1 (complete destruction).

[0108] In the embodiments of this application, the technical effect of step S5 is to achieve a concise, powerful and objective quantitative description of the long-term evolution process of geological disasters. Through the innovative index of landslide area ratio, it normalizes the spatial information of deformation at different times and scales to a unified scale, so that the entire process of disaster from gestation and development to outbreak can be characterized by a monotonically increasing curve. This completely solves the problem that traditional methods rely on subjective experience and lack quantitative scales for dividing evolution stages, and provides a solid mathematical foundation for automated and intelligent disaster early warning.

[0109] S6. Based on the landslide area ratio sequence and the preset creep stage threshold, a judgment is made. When the landslide area ratio exceeds the corresponding creep stage threshold, the disaster is determined to have entered the corresponding evolution stage, and an early warning message containing the stage judgment result is generated.

[0110] In some embodiments, the judgment is based on the landslide area ratio sequence and a preset creep stage threshold. When the landslide area ratio exceeds the corresponding creep stage threshold, the disaster is determined to have entered the corresponding evolution stage, and early warning information containing the stage determination result is generated. Specifically, this includes:

[0111] Obtain the sequence of collapse area ratios;

[0112] Read the preset creep stage thresholds, which include at least a primary creep threshold, a secondary creep threshold, and a tertiary creep threshold;

[0113] The latest phase of the collapse area ratio sequence is compared with the creep stage threshold in sequence.

[0114] When the collapse-slip area ratio of the latest time phase exceeds the primary creep threshold but does not exceed the secondary creep threshold, the disaster evolution stage is determined to be the primary creep stage;

[0115] When the collapse area ratio of the latest time phase exceeds the secondary creep threshold but does not exceed the tertiary creep threshold, the disaster evolution stage is determined to be the secondary creep stage;

[0116] When the collapse area ratio of the latest time phase exceeds the third-level creep threshold, the disaster evolution stage is determined to be the third-level creep stage;

[0117] Based on the determined disaster evolution stage, early warning information is generated, which includes information on the current stage and the corresponding early warning level.

[0118] In this embodiment, step S6 is the final decision-making and output step for realizing intelligent interpretation and early warning of geological disasters. Its purpose is to automatically diagnose the current evolution stage of the disaster based on the disaster evolution process signal—the landslide area ratio sequence—obtained by quantification in step S5, and generate early warning information with clear guiding significance through a set of preset and scientific judgment rules. This step directly addresses the core defects of the background technology, such as low automation and quantification and reliance on human experience judgment. By combining objective quantitative data with clear stage thresholds, it realizes fully automatic, standardized, and reproducible intelligent identification of the disaster evolution stage, completing the closed loop from data perception to decision support.

[0119] In this embodiment of the application, the creep stage threshold is a set of numerical criteria pre-set based on statistical patterns, theoretical model analysis, and expert experience from a large number of geological disaster cases. It is used to map the continuous landslide area ratio to discrete disaster evolution stages. The early warning information is a structured data output that includes at least the name of the currently determined disaster evolution stage, the corresponding suggested early warning level, and a timestamp.

[0120] In this embodiment of the application, the specific technical means for obtaining the landslide area ratio sequence is as follows: This step involves reading input data. The system directly reads the generated ordered sequence file or database record from the output result storage location of step S5. This sequence is a list sorted by timestamp, such as [(t1,R1),(t2,R2),...,(t_n,R_n)], where t_n is the latest time phase and R_n is the corresponding latest landslide area ratio. The system ensures that the latest and complete sequence data is read.

[0121] In this embodiment, the specific technical means for reading the preset creep stage threshold is as follows: This step involves loading the identification rules. The threshold, as a core judgment parameter, is pre-stored in the system's configuration file or parameter database. These thresholds include at least three key values: primary creep threshold, secondary creep threshold, and tertiary creep threshold. Their setting is based on the typical three-stage division in geological disaster creep theory, and the specific dividing point is determined by statistical analysis of the landslide area ratio evolution curve in historical cases. For example, by analyzing multiple landslide cases, it was found that when the deformation range ratio (landslide area ratio) is... When the product ratio is less than about one-third of the total potential range, deformation is usually slow and dispersed, corresponding to primary creep; when the ratio is between one-third and two-thirds, the deformation region expands rapidly and interconnects, corresponding to secondary (steady-state) creep; when the ratio exceeds two-thirds, it often indicates that deformation has entered an accelerated stage and is close to overall instability, corresponding to tertiary creep. Based on this, the primary creep threshold can be preset to a small value, the secondary creep threshold to a medium value, and the tertiary creep threshold to a large value close to 1. The system loads these specific values ​​from a specified path during runtime.

[0122] In this embodiment of the application, the specific technical means of comparing the latest phase's collapse area ratio in the collapse area ratio sequence with the creep stage threshold in sequence is as follows: This step is to perform the core logical judgment. The system first extracts the last data pair from the read sequence, namely the latest phase t_n and its collapse area ratio R_n. This is a simple program value retrieval operation. Subsequently, the system performs a series of condition judgments. It compares the value of R_n with the primary creep threshold read from the configuration file. This is a greater than or less than or equal to logical operation. According to the comparison result, the program flow will enter different branches. This comparison process is carried out step by step. First, it is judged whether the minimum threshold is exceeded, and then it is judged whether the higher threshold is exceeded in sequence, until the value range of R_n is determined.

[0123] In this embodiment of the application, the specific technical means for determining the disaster evolution stage as the primary creep stage when the latest phase of the landslide area ratio exceeds the primary creep threshold but does not exceed the secondary creep threshold is as follows. This is the specific implementation of the first judgment branch. In the program, this is represented by an if conditional statement, where the condition is (R_n > primary creep threshold) and (R_n ≤ secondary creep threshold). If this condition is true, the system sets the value of an internal state variable or output variable to an identifier representing the primary creep stage, such as the string PrimaryCreep or code 1. This means that the system determines that the current disaster is in the early stage of deformation, and the range is expanding slowly. For example, if the preset primary creep threshold is 0.3, the secondary creep threshold is 0.6, and the latest landslide area ratio R_n is 0.45, then this condition is met, and the system determines that it is in the primary creep stage.

[0124] In this embodiment of the application, the specific technical means for determining the disaster evolution stage as the secondary creep stage when the collapse area ratio of the latest time phase exceeds the secondary creep threshold but does not exceed the tertiary creep threshold is as follows. This is the specific implementation of the second judgment branch. Its program condition is (R_n > secondary creep threshold) and (R_n ≤ tertiary creep threshold). If this condition is met, the system sets the state to the identifier of the secondary creep stage (or steady-state creep stage), such as SecondaryCreep or code 2. This indicates that the system determines that the disaster has entered a dangerous stage where the deformation range is rapidly expanding but has not yet reached the critical point. Using the threshold in the previous example, if R_n is 0.65, then this condition is met, and the system determines it to be the secondary creep stage.

[0125] In this embodiment of the application, the specific technical means for determining the disaster evolution stage as the third-level creep stage when the collapse-slip area ratio of the latest time phase exceeds the third-level creep threshold is as follows: This is the highest-level judgment branch, and its condition is (R_n > third-level creep threshold). When the condition is met, the system sets the state to the identifier of the third-level creep stage (or accelerated creep stage, imminent slip stage), such as TertiaryCreep or code 3. This indicates that the system determines that the disaster has entered a critical state in which large-scale, rapid instability is highly likely to occur. If the third-level creep threshold is set to 0.8, and R_n reaches 0.9 or even 1.0, this judgment is triggered.

[0126] In this embodiment, the specific technical means for generating early warning information containing current stage information and corresponding early warning levels based on the determined disaster evolution stage is as follows: This step is the final information encapsulation and output. The system maps the stage identifier obtained from the above determination to a preset early warning level rule. For example, the mapping rule can be: primary creep stage corresponds to attention level (blue), secondary creep stage corresponds to warning level (yellow), and tertiary creep stage corresponds to alarm level (orange) or red alarm. Then, the system automatically generates a structured early warning information object or message. This information includes at least the following fields: disaster body identifier, current time phase, determined creep stage, corresponding early warning level, determination time, and optional deformation rate or ratio and other auxiliary information. This information can be pushed to the monitoring platform interface through the application programming interface to trigger an audible and visual alarm, or sent to relevant personnel via SMS, email, etc. For example, the system finally generates a message: {Early warning area: XX slope; Time phase: October 2023; Current stage: Level 2 creep (steady-state creep); Early warning level: Yellow; Note: The deformation range has exceeded 60% of the potential area, and patrols need to be strengthened}.

[0127] In this embodiment, the technical effect of step S6 is to realize the terminal intelligence of the interpretation process and the automation of early warning. It condenses all the complex multi-source data processing and deformation analysis results into a concise and clear stage judgment and early warning instruction, so that the monitoring data can be directly understood by non-professional managers and used for emergency decision-making, which greatly improves the timeliness, objectivity and practicality of geological disaster monitoring and early warning.

[0128] In this embodiment, step S6 is the final value realization link in relation to other steps in the whole method. It relies entirely on the collapse area ratio sequence provided by step S5, which accurately reflects the evolution process, as the only judgment input. Without the quantification of S5, the judgment of S6 will have no data to rely on. At the same time, S6 is the end point and output of the entire method technology chain. The early warning information it generates is the final product of the whole system serving disaster prevention and mitigation decision-making. All the technical efforts from S1 to S5 are ultimately for the purpose of enabling S6 to make this timely and accurate automated intelligent judgment.

[0129] like Figure 2 The diagram shown is a functional block diagram of a geological disaster intelligent interpretation system based on multi-source remote sensing data fusion, provided in an embodiment of this application.

[0130] The intelligent geological disaster interpretation system 100 based on multi-source remote sensing data fusion described in this application can be installed in an electronic device. Depending on the functions implemented, the intelligent geological disaster interpretation system 100 may include a data acquisition and preprocessing module 101, a multi-source data fusion modeling module 102, a temporal 3D reconstruction and enhancement module 103, an automatic surface deformation extraction module 104, an evolution process quantitative analysis module 105, and an intelligent early warning decision output module 106. The modules described in this application can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and can perform fixed functions, stored in the memory of the electronic device.

[0131] In this embodiment, the functions of each module / unit are as follows:

[0132] The data acquisition and preprocessing module 101 is used to acquire UAV aerial survey images, airborne lidar point cloud data and multi-temporal optical satellite images of the study area at least two time phases.

[0133] The multi-source data fusion modeling module 102 is used to spatially register and data-level fuse the UAV aerial survey images and the airborne lidar point cloud data of the same time phase to generate a first fused terrain model of the corresponding time phase.

[0134] The temporal 3D reconstruction and enhancement module 103 is used to temporally align the first fused terrain model of different time phases with the multi-temporal optical satellite imagery to generate a multi-temporal 3D terrain model sequence for the study area.

[0135] The automatic surface deformation extraction module 104 is used to extract the surface deformation area based on the multi-temporal three-dimensional terrain model sequence by calculating the elevation difference, and to identify the deformation evolution boundary of the surface deformation area.

[0136] The evolution process quantification analysis module 105 is used to calculate the cumulative deformation area and the total area of ​​the landslide area at each time phase based on the deformation evolution boundary, and then calculate the landslide area ratio sequence that reflects the evolution process of the disaster.

[0137] The intelligent early warning decision output module 106 is used to make a judgment based on the landslide area ratio sequence and the preset creep stage threshold. When the landslide area ratio exceeds the corresponding creep stage threshold, the disaster is determined to have entered the corresponding evolution stage, and an early warning information containing the stage judgment result is generated.

[0138] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0139] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0140] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0141] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application.

[0142] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0143] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of this application without departing from the spirit and scope of the technical solutions of this application.

Claims

1. A method for intelligent interpretation of geological hazards through multi-source remote sensing data fusion, characterized in that, The method includes: Acquire UAV aerial survey images, airborne lidar point cloud data, and multi-temporal optical satellite images of the study area at least two time phases; Spatial registration and data-level fusion are performed on the UAV aerial survey images and the airborne lidar point cloud data of the same time phase to generate a first fused terrain model of the corresponding time phase. The first fused terrain model from different time phases is time-series aligned with the multi-temporal optical satellite imagery to generate a multi-temporal three-dimensional terrain model sequence for the study area. Based on the multi-temporal three-dimensional terrain model sequence, the surface deformation region is extracted by elevation difference calculation, and the deformation evolution boundary of the surface deformation region is identified. Based on the deformation evolution boundary, the cumulative deformation area and the total landslide area of ​​each time phase are calculated, and then a landslide area ratio sequence reflecting the disaster evolution process is obtained. Specifically, this includes: obtaining the deformation evolution boundary corresponding to each time phase based on the multi-time phase three-dimensional terrain model sequence; for each time phase, calculating the cumulative deformation area of ​​that time phase according to its corresponding deformation evolution boundary; defining the boundary of the unstable slope area covering all cumulative deformation areas, determined based on the deformation evolution boundaries of all time phases throughout the entire monitoring period, as the landslide area; calculating the area of ​​the unstable slope area as the total landslide area; for each time phase, calculating the ratio of the cumulative deformation area of ​​that time phase to the total landslide area to obtain the landslide area ratio of that time phase; arranging the landslide area ratios of all time phases according to their corresponding time sequence to generate a landslide area ratio sequence reflecting the disaster evolution process. Based on the landslide area ratio sequence and the preset creep stage threshold, when the landslide area ratio exceeds the corresponding creep stage threshold, the disaster is determined to have entered the corresponding evolution stage, and an early warning message containing the stage determination result is generated.

2. The intelligent interpretation method for geological hazards based on multi-source remote sensing data fusion as described in claim 1, characterized in that, The acquisition of at least two temporal phases of UAV aerial survey imagery, airborne lidar point cloud data, and multi-temporal optical satellite imagery of the study area specifically includes: Ground control points were deployed within the study area; Control the drone equipped with an oblique photography camera to conduct aerial photography of the study area according to the preset flight altitude, heading overlap rate and lateral overlap rate, and acquire the drone aerial survey images; The drone equipped with lidar is controlled to perform aerial scanning of the study area to acquire raw point cloud data; The original point cloud data is filtered, registered, and ground point extracted to generate the airborne lidar point cloud data with non-ground points removed. Obtain multi-temporal optical satellite images from a preset satellite image database that cover the study area and span at least two time phases.

3. The intelligent interpretation method for geological hazards based on multi-source remote sensing data fusion as described in claim 1, characterized in that, The step of spatially registering and data-level fusing the UAV aerial survey imagery and the airborne lidar point cloud data of the same time phase to generate a first fused terrain model corresponding to the same time phase specifically includes: Based on the UAV aerial survey images of the same time phase, a digital orthophoto model and a digital ground model of the study area are generated; A digital elevation model of the study area is generated based on the airborne lidar point cloud data of the same time phase. Register the digital orthophoto model, the digital terrain model, and the digital elevation model to the same geographic coordinate system; The registered digital terrain model and the digital elevation model are fused at the data level to generate the first fused terrain model corresponding to the time phase.

4. The intelligent interpretation method for geological hazards based on multi-source remote sensing data fusion as described in claim 3, characterized in that, The step of performing data-level fusion of the registered digital ground model and the digital elevation model to generate the first fused terrain model of the corresponding time phase specifically includes: using an inverse distance weighted interpolation algorithm to perform fusion calculation on the registered digital ground model and the digital elevation model to generate the first fused terrain model.

5. The intelligent interpretation method for geological hazards based on multi-source remote sensing data fusion as described in claim 1, characterized in that, The step of temporally aligning the first fused terrain model from different time phases with the multi-temporal optical satellite imagery to generate a multi-temporal three-dimensional terrain model sequence for the study area specifically includes: Acquire the first fused terrain model generated at different time phases, and the multi-temporal optical satellite imagery; Based on the acquisition time of the first fused terrain model, the corresponding images are selected from the multi-temporal optical satellite images as matching images; Using the first fused terrain model as a spatial reference, the matching image is spatially registered with it; Based on the registered optical satellite imagery, the surface texture of the first fused terrain model is enhanced to generate three-dimensional terrain models of each time phase with temporal texture information. The three-dimensional terrain models generated at each time phase are arranged according to their corresponding time sequence to form a multi-temporal three-dimensional terrain model sequence for the study area.

6. The intelligent interpretation method for geological hazards based on multi-source remote sensing data fusion as described in claim 1, characterized in that, The step of extracting surface deformation regions based on the multi-temporal three-dimensional terrain model sequence through elevation difference calculation and identifying the deformation evolution boundaries of the surface deformation regions specifically includes: From the multi-temporal three-dimensional terrain model sequence, select at least two temporal three-dimensional terrain models in chronological order, one as the baseline model and the other as the comparison model; Elevation difference calculations are performed on the benchmark model and the comparison model to obtain the surface elevation change displacement field of the study area; Based on a preset elevation change threshold, regions whose elevation change exceeds the preset displacement threshold are extracted from the elevation change displacement field and used as the surface deformation regions. Based on the extracted boundaries of the surface deformation regions, the deformation evolution boundaries of the surface deformation regions are identified and determined.

7. The intelligent interpretation method for geological hazards based on multi-source remote sensing data fusion as described in claim 1, characterized in that, The judgment is based on the landslide area ratio sequence and a preset creep stage threshold. When the landslide area ratio exceeds the corresponding creep stage threshold, the disaster is determined to have entered the corresponding evolution stage, and an early warning message containing the stage determination result is generated, specifically including: Obtain the sequence of collapse area ratios; Read the preset creep stage thresholds, which include at least a primary creep threshold, a secondary creep threshold, and a tertiary creep threshold; The latest phase of the collapse area ratio sequence is compared with the creep stage threshold in sequence. When the collapse-slip area ratio of the latest time phase exceeds the primary creep threshold but does not exceed the secondary creep threshold, the disaster evolution stage is determined to be the primary creep stage; When the collapse area ratio of the latest time phase exceeds the secondary creep threshold but does not exceed the tertiary creep threshold, the disaster evolution stage is determined to be the secondary creep stage; When the collapse area ratio of the latest time phase exceeds the third-level creep threshold, the disaster evolution stage is determined to be the third-level creep stage; Based on the determined disaster evolution stage, early warning information is generated, which includes information on the current stage and the corresponding early warning level.

8. A multi-source remote sensing data fusion-based intelligent interpretation system for geological hazards, used to implement the multi-source remote sensing data fusion-based intelligent interpretation method for geological hazards as described in any one of claims 1-7, characterized in that, The system includes: The data acquisition and preprocessing module is used to acquire UAV aerial survey images, airborne lidar point cloud data, and multi-temporal optical satellite images of the study area at least two time phases. The multi-source data fusion modeling module is used to spatially register and data-level fuse the UAV aerial survey images and the airborne lidar point cloud data of the same time phase to generate a first fused terrain model of the corresponding time phase. The temporal 3D reconstruction and enhancement module is used to temporally align the first fused terrain model with the multi-temporal optical satellite imagery at different time phases to generate a multi-temporal 3D terrain model sequence for the study area. The automatic surface deformation extraction module is used to extract surface deformation areas based on the multi-temporal three-dimensional terrain model sequence through elevation difference calculation, and to identify the deformation evolution boundary of the surface deformation area. The evolution process quantification analysis module is used to calculate the cumulative deformation area and the total landslide area of ​​each time phase based on the deformation evolution boundary, and then calculate the landslide area ratio sequence reflecting the disaster evolution process. Specifically, it includes: obtaining the deformation evolution boundary corresponding to each time phase based on the multi-time phase three-dimensional terrain model sequence; calculating the cumulative deformation area of ​​each time phase according to its corresponding deformation evolution boundary; defining the boundary of the unstable slope area covering all cumulative deformation areas as the landslide area within the entire monitoring period based on the deformation evolution boundaries of all time phases; calculating the area of ​​the unstable slope area as the total landslide area; calculating the ratio of the cumulative deformation area to the total landslide area of ​​each time phase to obtain the landslide area ratio of that time phase; arranging the landslide area ratios of all time phases according to their corresponding time sequence to generate the landslide area ratio sequence reflecting the disaster evolution process. The intelligent early warning decision output module is used to make a judgment based on the landslide area ratio sequence and the preset creep stage threshold. When the landslide area ratio exceeds the corresponding creep stage threshold, the disaster is determined to have entered the corresponding evolution stage, and an early warning information containing the stage judgment result is generated.

Images