Ground subsidence danger early warning method based on multi-data fusion

Abstract

The invention discloses a ground subsidence risk early warning method based on multi-data fusion, and relates to the technical field of geological disaster monitoring and early warning, and the method comprises the following steps: S1, constructing an albedo distribution field, a humidity distribution field and a ground feature brightness distribution field, fusing the three types of data, and generating a multi-dimensional fingerprint spectrum in combination with time sequence changes, outputting an abnormal fingerprint band; and S2, performing continuous connection line comparison corresponding to the abnormal fingerprint band on the multi-period remote sensing image, identifying a phase delay change path, extracting a sudden change section in phase delay change as a phase delay inflection point region, and calibrating a delay time window for atmospheric disturbance stripping based on the phase delay inflection point region. According to the method, effective separation of phase errors in a high-humidity and high-albedo environment is realized through multi-dimensional fingerprint spectrum construction and disturbance stripping, and stability and continuity of earth surface deformation data are improved by combining shadow path injection and time window rearrangement, so that accuracy and reliability of settlement early warning are enhanced.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring and early warning technology, specifically to a method for early warning of ground subsidence risk based on multi-data fusion. Background Technology

[0002] Land subsidence hazard early warning based on multi-data fusion refers to the comprehensive utilization of data from different sources and types, such as satellite remote sensing monitoring data (InSAR), Global Navigation Satellite System (GNSS) displacement observation data, groundwater level monitoring data, geological drilling data, surface building subsidence records, and meteorological and hydrological information, to establish a unified space-time analysis model for dynamic assessment of the occurrence, development, and affected areas of land subsidence. Simultaneously, an intelligent sensing system is introduced, deploying multiple types of surface and underground sensing units in key areas to continuously sense and collect data in real time on surface deformation, groundwater dynamics, soil strain, and environmental parameters, providing a high-timeliness and high-precision foundation for multi-source data. This method achieves the complementarity and correction of multi-dimensional information through data fusion algorithms. Combined with the continuous monitoring results output by the intelligent sensing system, it dynamically corrects remote sensing and discrete observation data, identifies subtle trend changes in ground deformation, and then constructs a hazard index model to provide graded early warnings for potential subsidence areas. This allows for the early identification of geological hazards, assisting government and engineering management departments in formulating scientific prevention and control measures, and achieving a shift from passive response to proactive prevention.

[0003] The existing technology has the following shortcomings: In environments with extreme high humidity or high albedo (such as snow-covered ground, metal roofs, saline-alkali tidal flats, and wetland evaporation zones), the strong reflection of radar waves by the surface and the propagation refraction effect caused by atmospheric humidity will superimpose to produce an abnormal phase delay, causing nonlinear distortion of the radar echo signal along its propagation path. This phase delay is mistakenly identified as a surface displacement signal during surface deformation inversion, thus introducing systematic errors. When multiple periods of remote sensing data are overlaid and analyzed, this small deviation will continue to accumulate, leading to a false subsidence trend in the deformation estimation results. Ultimately, the system will misjudge signal disturbances affected by meteorological changes as actual geological subsidence processes, resulting in overestimation of risk levels or false triggering of warnings, severely weakening the accuracy and reliability of land subsidence hazard warnings.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a ground subsidence hazard early warning method based on multi-data fusion to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a ground subsidence hazard early warning method based on multi-data fusion, comprising the following steps: S1. Construct albedo distribution field, humidity distribution field and ground object brightness distribution field, fuse the three types of data, combine time series changes to generate multidimensional fingerprint map, and output anomaly fingerprint band containing temporal and spatial anomaly features. S2 performs continuous line comparison with the anomalous fingerprint band on multiple remote sensing images, identifies the phase delay change path, extracts the abrupt change segment in the phase delay change as the phase delay inflection point region, and calibrates the lag time window for atmospheric disturbance stripping based on the phase delay inflection point region. S3, initiate the meteorological disturbance stripping process based on the lag time window, use the phase delay inflection point region as the change constraint condition, separate the specular reflection component and the atmospheric water vapor tail component in the remote sensing phase data, and generate the corresponding meteorological stripping data sequence. S4 projects the meteorological stripped data sequence onto a spatiotemporal grid composed of land cover classifications, extracts the locations of abrupt changes in albedo and humidity, and outputs a list of risk anchor points containing error-sensitive areas based on spatial distribution characteristics. S5 takes the risk anchor list as input, performs shadow path injection operation based on alternating shading mechanism on high albedo remote sensing pixels, and applies periodically changing time window rearrangement in the time axis direction to offset phase delay error and restore the stability of surface displacement change trend.

[0007] Preferably, step S1 includes: The remote sensing image sequence covering the target area is organized in chronological order to construct an albedo distribution field, and the albedo transition state and persistence characteristics of each pixel are marked. In the spatial grid corresponding to the albedo distribution field, a humidity distribution field is constructed, and information on humidity change gradient, humidity propagation direction, and abrupt change points is extracted. A luminance distribution field is constructed based on the albedo distribution field and the humidity distribution field. The luminance fluctuation region is extracted and the three types of distribution fields are made spatially and temporally consistent. The albedo distribution field, humidity distribution field and brightness distribution field are jointly modeled to generate multidimensional variation curves, and abnormal spatiotemporal fingerprint bands are extracted by sliding comparison and trend superposition.

[0008] Preferably, step S2 includes: Within the range of the anomalous spatiotemporal fingerprint, remote sensing image sequences within a continuous time period are selected, and phase change observations at the corresponding spatial locations are extracted to generate a continuous line sequence across the time dimension. Analyze the continuous line sequences, extract the phase change segments with abrupt jump characteristics as abrupt change candidate regions, and further aggregate them into phase change corridors; Based on the phase change corridor, the interference core region with hysteresis effect is calibrated, and a hysteresis time window group is established; By combining the lag time window group with the anomalous spatiotemporal fingerprint band, a time reference structure with a dual constraint mechanism is constructed.

[0009] Preferably, the starting point of the time window group in the time reference structure is used to trigger the meteorological disturbance stripping process, and the spatial connectivity of the anomalous spatiotemporal fingerprint band is used to limit the stripping range, so that the meteorological disturbance stripping process is only executed within the spatiotemporal unit containing the phase change corridor.

[0010] Preferably, step S3 includes: Based on the start and end boundaries of the lag time window, multiple remote sensing image sequences are located, and phase data consistent with the phase delay inflection point area are extracted to construct a set of phase change trajectories. Using the phase delay inflection point region as a variation constraint, the specular reflection component data set is extracted; Based on the identification of specular reflection components, the atmospheric water vapor tail component data set is extracted; The specular reflection component and the atmospheric water vapor tail component are combined to construct a meteorological stripping data sequence, and the start and end time range of the stripping and the source of interference are retained.

[0011] Preferably, the meteorological stripping data sequence maintains the same spatial resolution and temporal density as the original remote sensing phase data during the construction process, and distinguishes the source type of specular reflection component and atmospheric water vapor tail component by pixel value identification, so as to ensure the continuity and accuracy of subsequent surface displacement trend recovery.

[0012] Preferably, step S4 includes: Based on the land cover information of the remote sensing image sequence, a land cover classification layer is constructed, and the meteorological stripping data sequence is projected onto the spatiotemporal grid composed of land cover classification to form an interference response set; Extract the regions that overlap with albedo abrupt changes in the projected data, identify albedo abrupt change points, and generate albedo perturbation bands; Based on the albedo perturbation band, humidity abrupt change regions were extracted, and high-sensitivity regions of humidity abrupt change with spatiotemporal overlap between humidity transition and phase perturbation were screened out. By combining the response intensity and persistence of changes in meteorological stripping data, spatial units that have been in a state of high frequency of disturbances for a long time are selected, and a list of risk anchor points containing error-sensitive areas is output.

[0013] Preferably, the spatial units in the risk anchor point list are sorted according to meteorological stripping intensity and land cover classification characteristics. Anchor points at the intersection of albedo change zones and humidity change high-sensitivity areas are selected as error-sensitive core areas for subsequent image masking and trend restoration processing.

[0014] Preferably, step S5 includes: Extract the spatial location set corresponding to high albedo features from the risk anchor list, and construct a time-series disturbance response matrix to form a continuous phase change record; Differentiated masking operations are performed based on the disturbance intensity level of the risk anchor point, and rhythmically adjusted shadow paths are injected into the masking interval to supplement the observation data; After completing the shadow path injection, the timeline is periodically rearranged, and a breathing time window group is set to alternately form occlusion and opening cycles. After the time window is rearranged, the time series structure of high albedo pixels is reorganized, and the adjusted surface displacement trend record is output.

[0015] Preferably, during the shadow path injection process, rhythmic interpolation adjustment is performed on the phase response trajectory of neighboring pixels to ensure a continuous transition of the injection path in time and space. After the shadow path injection is completed, the phase delay error of high albedo pixels is dynamically offset and the stability of the surface displacement trend is restored through the alternating action of breathing time window groups.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention constructs a multidimensional fingerprint map integrating albedo, humidity, and ground feature brightness, and uses anomalous fingerprint bands as clues to perform cross-temporal identification and disturbance separation processing, effectively avoiding the accumulation of nonlinear errors in radar phase information in high-humidity and high-reflection environments. By introducing a lag time window and combining it with phase inflection point regions for constrained localization, anomalous signals and the actual deformation process are separated in both time and space, significantly improving the stability of deformation trend representation and enhancing the targeted identification of surface disturbances.

[0017] This invention introduces shadow path injection and time window rearrangement mechanisms based on the risk anchor point list. Through rhythmic masking and dynamic replacement, it weakens the interference of high-risk pixels on the overall sequence, ensuring consistency in data continuity and trend extension of surface deformation results. Particularly in areas with frequent albedo jumps or persistent meteorological disturbances, this processing path effectively suppresses error propagation, improving the reliability and practical value of subsidence early warning results. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a flowchart of the method for early warning of ground subsidence risk based on multi-data fusion according to the present invention. Detailed Implementation

[0020] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0021] This invention provides, for example Figure 1 The land subsidence hazard early warning method based on multi-data fusion shown includes the following steps: S1. Construct albedo distribution field, humidity distribution field and ground object brightness distribution field, fuse the three types of data, combine time series changes to generate multidimensional fingerprint map, and output anomaly fingerprint band containing temporal and spatial anomaly features. To effectively extract anomalous regions with temporal and spatial variation characteristics, it is necessary to first complete a comprehensive fusion process of albedo, humidity, and ground feature brightness information, and then combine this with a temporal dimension to establish a fingerprint map capable of characterizing the evolutionary trajectory. This process includes the following steps: A sequence of remote sensing images covering the target area was acquired, ensuring temporal continuity. For each image, surface albedo data was calculated based on optical and thermal infrared bands, as well as observation parameters suitable for brightness estimation, and an albedo distribution map was constructed using a rasterized approach. During the construction process, albedo threshold ranges were set for different types of surface areas to distinguish between various surface conditions such as bare soil, water bodies, snow and ice, vegetation, and buildings. By processing the albedo maps for each period in the time series pixel by pixel, an albedo distribution field was established according to the observation time sequence, identifying its transition states and persistence characteristics in the temporal dimension. Transition states refer to the abrupt changes in surface albedo at different times, while persistence characteristics describe its stability or gradual change within adjacent time intervals. This albedo distribution field serves as the spatial reference benchmark for subsequent fusion operations.

[0022] Surface humidity observation data within the same time period are incorporated into the spatial grid corresponding to the albedo distribution map. Humidity information can be derived from thermal infrared band inversion, ground observation data interpolation, or other effective spatial data sources. By spatially registering and temporally processing multiple periods of humidity data, a humidity distribution field is constructed according to the same spatial resolution and time nodes as the albedo distribution field. To enhance the expressive power of the humidity distribution field in terms of dynamic characteristics, the humidity change gradient of each pixel over continuous time slices needs to be extracted to identify the spatial propagation direction and concentration trend of humidity, and abrupt changes and increasing bands are classified and labeled in conjunction with the time series. The establishment of the humidity distribution field not only needs to maintain consistency with the albedo distribution field in spatial coding but also needs to construct an evolution path at the temporal level that reflects the accumulation of high-humidity disturbances to capture potential atmospheric humidity disturbance areas.

[0023] Building upon existing albedo and humidity distribution fields, this paper further incorporates ground feature brightness information to perform panoramic brightness modeling of remote sensing imagery. The brightness data is constructed based on the raw reflectance values ​​from the remote sensing images, combining various surface features to quantify the light scattering capacity, thereby generating a brightness distribution map describing the overall brightness intensity of the surface. Brightness distribution maps from different periods are uniformly resampled to form a brightness distribution field. Within this field, brightness fluctuation areas are extracted, identifying brightness change bands that overlap with albedo transition zones and humidity fluctuation zones. By establishing a continuous data sequence of the brightness distribution field along the time axis, spatial alignment and temporal label consistency with the previous two types of distribution fields are further ensured, guaranteeing structural uniformity in the multidimensional feature representation of the three data distribution fields. The addition of the brightness distribution field enhances the ability to identify highly reflective areas such as buildings, metal roofs, and snow-covered areas, thereby improving the clarity of identifying areas with abnormal changes.

[0024] The albedo, humidity, and brightness distribution fields are jointly modeled to generate a fused map containing temporal trends. In this process, the change trajectories of each pixel in the three dimensions are merged according to a unified spatial grid structure and observation time axis, forming a set of multi-dimensional change curves. These curves are categorized and extracted, and a change feature matrix is ​​generated based on their directionality, consistency, and variability. Furthermore, by sliding comparison in the temporal dimension and trend superposition in the spatial dimension, continuous regions exhibiting anomalous combinations of changes across multiple time periods are identified. These regions are extracted to form anomaly spatiotemporal fingerprints, which fully characterize the co-occurrence of three types of features—albedo mutation, anomalous humidity fluctuation, and anomalous brightness increase—under specific time series, and possess traceable spatial continuity. These anomalous spatiotemporal fingerprints serve as crucial inputs for subsequent analysis of phase delay sources and identification of surface deformation anomalies, providing stable and accurate prior information support for subsequent time window calibration and error stripping processes.

[0025] S2 performs continuous line comparison with the anomalous fingerprint band on multiple remote sensing images, identifies the phase delay change path, extracts the abrupt change segment in the phase delay change as the phase delay inflection point region, and calibrates the lag time window for atmospheric disturbance stripping based on the phase delay inflection point region. To fully utilize the constructed spatiotemporal fingerprint bands and further acquire dynamic evolution information capable of representing potential phase disturbance regions, continuous cross-temporal alignment operations are required along the spatial path of the anomaly fingerprint bands in multiple remote sensing images. This identifies potential phase delay change paths and further extracts abrupt change feature regions as the spatial representation basis for phase delay inflection point regions. Based on the spatial distribution and evolution trend of phase delay inflection point regions, a time window structure for meteorological disturbance stripping is established. This process includes the following steps: Within the established spatiotemporal fingerprint zone of anomalies, remote sensing image sequences covering the same area within consecutive time periods are selected and organized chronologically. For each time point, phase change observations at the spatial location corresponding to the anomaly fingerprint zone are extracted, and these observations are connected to form a continuous change curve. This change curve is unfolded along the main axis of the anomaly fingerprint zone, maintaining consistency between the pixel location and the multidimensional fingerprint structure from the previous step. By overlaying the phase change curves at different time points in a unified coordinate system, a set of continuous lines across the time dimension is formed. This sequence demonstrates the phase evolution trend along the same spatial path, laying the data foundation for the subsequent extraction of abrupt change segments.

[0026] For the phase change curves at each time node in the continuous linear sequence, the extension mode and change amplitude of each segment in the time dimension are analyzed segment by segment. Locations where the phase difference changes abruptly between adjacent time periods are identified, and these segments with abrupt jump characteristics are extracted as preliminary abrupt change candidate areas. This identification process uses the spatial extension trajectory of the anomalous fingerprint band as a clue, focusing on analyzing those phase discontinuities that persist in albedo transition regions and humidity abrupt change regions. Each abrupt change candidate area corresponds to a specific combination of land cover type and environmental conditions, thus forming a one-to-one correspondence with the albedo, humidity, and brightness feature sequences constructed in the previous stage. By further aggregating abrupt change candidate areas with temporal consistency, a spatially coherent phase abrupt change corridor is formed.

[0027] Based on the identified phase abrupt change corridors, local time series structure analysis is conducted to track the continuation characteristics of each abrupt change segment in subsequent time nodes. If a certain abrupt change segment maintains high amplitude changes across multiple adjacent time nodes without rapidly stabilizing, this segment is defined as a core disturbance region with hysteresis. Using these core disturbance regions as a basis, the position of their first appearance is traced back along the time axis, and the time range before the apparent abrupt change is traced forward, thus defining the temporal extension interval of the disturbance. This interval is the core component of the hysteresis time window. Furthermore, the hysteresis time intervals of different abrupt change segments are aligned and integrated to form a group of hysteresis time windows with spatial dependence. This window group not only covers the main path of phase disturbances but also captures the temporal shift structure caused by differences in surface albedo, humidity persistence, or brightness fluctuations during disturbance propagation.

[0028] By combining the spatial distribution relationship between lag time window groups and anomalous fingerprint bands, a time reference structure with a dual constraint mechanism is established. On the one hand, the starting point of the time window group serves as the initiation signal for meteorological disturbance stripping operations, initiating subsequent processing procedures for all phase data entering the window range. On the other hand, the spatial connectivity of the anomalous fingerprint bands serves as the boundary condition for the disturbance evolution path, limiting the stripping operation to only apply to spatiotemporal units containing identified phase inflection point regions. This dual constraint mechanism ensures the accuracy and spatial targeting of subsequent operations in target localization and provides sufficient lag expression capability in time control. The lag time window not only expresses the time interval of disturbance initiation and duration but also closely integrates the phase change trends in multi-period remote sensing images with the dynamic evolution of previous surface albedo and humidity characteristics, forming a time management strategy with proactive response capability to meteorological disturbance effects, providing an operable time structure basis for subsequent meteorological stripping stages.

[0029] S3, initiate the meteorological disturbance stripping process based on the lag time window, use the phase delay inflection point region as the change constraint condition, separate the specular reflection component and the atmospheric water vapor tail component in the remote sensing phase data, and generate the corresponding meteorological stripping data sequence. To effectively address the interference of high humidity and high albedo surfaces on remote sensing phase data, a meteorological disturbance stripping process guided by both spatial and temporal characteristics needs to be initiated based on the established lag time window. This process should use the phase delay inflection point region as the core constraint to separate the non-geological components affected by disturbance from a temporally continuous remote sensing image sequence. This process not only needs to identify the spatiotemporal distribution characteristics of different disturbance sources but also maintain consistency with the original phase structure during the stripping process. Specifically, it includes the following steps: Based on the clearly defined start and end boundaries of the lag time window, multiple remote sensing image sequences within this time interval are located, and a set of phase data consistent with the spatial location of the phase delay inflection point region is extracted from these images. To ensure spatial consistency in the stripping process, pixel-level coordinate anchoring is performed on the boundary of the phase delay inflection point region, establishing a spatial matching relationship between its corresponding positions in each image sequence. Based on this, a set of phase change trajectories under a time series structure is constructed to describe the phase evolution process of this region within the lag time. This process serves as the input basis for subsequent interference separation, possessing clear temporal clues and spatial positioning, ensuring that each set of data maintains consistent source and continuous path during processing.

[0030] Using the identified phase delay inflection point region as a constraint, the response differences of phase data at different time points within the lag time window are analyzed, with a focus on phase anomaly change segments that occur temporally close to the inflection point. These segments often exhibit abrupt changes in the time series, and the direction of change is often inconsistent with geological trends, reflecting the nonlinear disturbance effect introduced by external interference. Based on the spatial distribution characteristics of albedo transition zones and brightness change bands, these phase anomaly segments are initially attributed to specular reflection interference, as they often appear on building roofs, snow-covered surfaces, or highly reflective areas. Further comparative analysis of the brightness characteristics of these areas at different times is conducted to extract the temporal relationship between brightness abrupt change time and phase anomaly response, clarify the temporal influence segment of specular reflection components, and separately identify them from the original phase series to form a specular reflection component data set.

[0031] Building upon the identification of specular reflection components, the study further identifies the phase response change paths caused by atmospheric humidity variations. For phase change sequences located in high-humidity areas, evaporation zones, or wetland edges within the lag time window, the study analyzes their tailing characteristics on the time axis, i.e., the phase response continues to slowly extend after abrupt changes. This tailing behavior is common in areas with slow air mass changes and continuous humidity accumulation, exhibiting a temporal offset characteristic from albedo perturbations. By comparing gradient migration paths in the humidity distribution field, the consistency between the humidity propagation direction and the phase change direction is extracted, serving as the temporal localization basis for the humidity tailing effect. Phase response segments exhibiting temporal tailing characteristics and spatially supported by humidity fluctuations are labeled as atmospheric water vapor tailing components and extracted separately from the original phase sequences to form independent datasets.

[0032] After identifying and separating the specular reflection component and the atmospheric water vapor tail component, their spatial and temporal data sequences are unified and aggregated to construct a meteorological stripping data sequence covering all areas affected by meteorological disturbances within the lag time window. This data sequence maintains the same spatial resolution and temporal density as the original remote sensing phase data, ensuring that the original phase data can be corrected or canceled in a differential manner during subsequent processing. Each pixel value in the stripped data sequence is marked with a source identifier, distinguishing whether it belongs to the specular reflection effect or the atmospheric humidity tail effect, and indicating the corresponding stripping start and end time range. This structure not only provides targeted input for subsequent surface displacement trend recovery but also provides quantifiable disturbance intensity information for subsequent risk area extraction and early warning classification. Through the meteorological stripping data sequence output by this process, the land subsidence hazard early warning method can effectively remove non-deformation interference factors introduced by high humidity and high albedo environments without affecting the actual geological deformation trend, improving the temporal stability and spatial consistency in continuous remote sensing data analysis.

[0033] S4 projects the meteorological stripped data sequence onto a spatiotemporal grid composed of land cover classifications, extracts the locations of abrupt changes in albedo and humidity, and outputs a list of risk anchor points containing error-sensitive areas based on spatial distribution characteristics. To more accurately identify potential error interference areas under complex meteorological disturbances, it is necessary to further combine the output meteorological stripping data sequence with surface physical characteristics, and spatially label and classify disturbance hotspots through a joint temporal and spatial organization method. To this end, the meteorological stripping data sequence needs to be projected onto a spatiotemporal grid constructed based on land cover classification, and combined with the spatial evolution characteristics of albedo and humidity abrupt changes, key areas with disturbance sensitivity are extracted, ultimately outputting a list of risk anchor points with spatial positioning capabilities. This process includes the following steps: Based on the original ground cover information of the area covered by the remote sensing image sequence, a set of ground cover classification layers are constructed, and these layers are spatially divided into a unified grid structure. Each grid cell corresponds to a set of spatial coordinates, possesses a unique ground cover identifier, and is consistent with the remote sensing observation time node in the temporal dimension. The ground cover classification information comes from the interpretation results of the fusion of visible light band, near-infrared band, and thermal infrared band, and may include, but is not limited to, building roofs, bare land, vegetation-covered areas, water areas, snow-covered areas, and saline-alkali bare beaches. During the grid construction process, it is necessary to ensure that the ground cover classification layers and the meteorological stripping data sequence are completely matched in spatial resolution and time node, so as to achieve accurate projection of the stripped data in the spatiotemporal grid. During the projection process, each pixel value in the stripped data will be mapped to its corresponding ground cover cell, forming a set of interference responses with ground cover characteristic attributes.

[0034] Based on the projected data, and according to the spatial distribution of the specular reflection component and atmospheric water vapor trail component identified in the previous stage, the concentrated areas of these components in the land cover classification grid are analyzed, and areas overlapping with albedo abrupt changes are extracted. This process focuses on locating highly reflective land cover features, such as metal roofs, snow-covered areas, and saline-alkali tidal flats, and the corresponding pixel sets within these areas. By extracting the gradient changes in albedo over time in these areas, locations where albedo jumps occur in a short period of time are identified, i.e., albedo abrupt change areas. Each albedo abrupt change point is associated with a specific land cover type and assigned a time tag to record the location and time of its first albedo abrupt change. Using these locations as centers, and combining the stripping intensity values ​​within the surrounding grid points, the spatial diffusion trend and attenuation boundary of the disturbance are further derived, thus forming albedo disturbance bands with directionality and extent.

[0035] Based on the identification of albedo disturbance bands, further perturbation regions related to abrupt humidity changes are located. These abrupt humidity change regions are mainly concentrated in low-lying areas, wetland edges, water evaporation zones, and artificially irrigated areas—land cover types with high moisture accumulation capacity. Within these regions, by analyzing the temporal trajectory of humidity data, humidity response segments exhibiting rapid changes or sustained increases within the lag time window are identified. The locations of these abrupt humidity change segments are extracted and compared with the spatial trajectories of atmospheric water vapor tail components in meteorological stripping data to screen out intersection areas where humidity transitions and phase perturbations have a spatiotemporal overlap. These intersection areas are marked as high-sensitivity areas for humidity changes, and their corresponding land cover type, spatial location, and time period are recorded. Through this process, a stable humidity perturbation identification structure can be established within the spatiotemporal grid, forming a cross-over relationship with the aforementioned albedo disturbance bands, providing multidimensional constraint support for subsequent target extraction.

[0036] After identifying and locating albedo disturbance zones and highly sensitive areas of humidity abrupt changes, the response intensity and persistence of meteorological stripping data within these areas are comprehensively considered. Spatial units that have been in a state of high disturbance frequency for a long time and exhibit anomalous responses at multiple time points are selected as representative locations of disturbance-sensitive areas. These locations are further mapped onto a spatiotemporal grid to form a set of spatial anchor points with time-series tracking capabilities. Each anchor point carries parameters such as timestamp, land cover classification, disturbance type (albedo or humidity), and meteorological stripping intensity, used to express its role as the core of the disturbance response throughout the early warning analysis process. Finally, these anchor points are sorted by time sequence and spatial location, and a risk anchor point list with a complete structure is output. This list can serve as the target input for anomaly compensation and masking operations in subsequent steps, and can also be used to assist in risk level classification and response mechanism adjustment, improving the adaptability and spatial positioning capability of the entire early warning method under complex surface conditions.

[0037] S5, taking the risk anchor list as input, performs shadow path injection operation based on alternating shading mechanism on high albedo remote sensing pixels, and applies periodically changing time window rearrangement in the time axis direction to offset phase delay error and restore the stability of surface displacement change trend. To mitigate the persistent phase delay error caused by high-albedo areas in multi-period remote sensing imagery and to achieve rhythmic attenuation of anomalous interference signals over time, intervention processing is required on the corresponding high-albedo pixels in the remote sensing data based on an extracted list of risk anchor points. This processing reconstructs the true trend trajectory of surface deformation by implementing an alternating shading mechanism and shadow path injection, combined with rhythmic reconstruction over time. The process includes the following steps: Based on the generated list of risk anchor points, the spatial location set of high-albedo land cover categories is extracted, and the corresponding pixel index is located in multi-period remote sensing image sequences. Each high-albedo pixel has clear spatial coordinates, land cover type label, and disturbance start and end time range, and carries risk level description information. These pixels are uniformly arranged at each time node, and a cross-time disturbance response matrix is ​​constructed, using the spatial index as the primary key to organize a continuous phase change record. This record sequence usually exhibits frequent fluctuations and obvious jumps in the anomaly occurrence area, which can easily lead to false trend extensions. Based on this, according to the disturbance intensity level of the risk anchor points, a differentiated masking strategy is implemented for different pixels. That is, the data participation of these high-interference pixels is artificially excluded at certain observation time nodes, forming discontinuous observation windows, thereby breaking the accumulation path of anomalous phase values ​​on the time axis.

[0038] To ensure a smooth spatial transition during the shading operation, a local spatial neighborhood region needs to be constructed around each high-albedo pixel. Within this neighborhood region, normal pixels unaffected by meteorological disturbances during the same time period are identified as alternative path sources. The phase response trajectories of these neighboring pixels are used as candidate reference paths, and these paths are selected based on factors such as spatial proximity, relative differences in disturbance levels, and consistency of temporal evolution to determine the shadow path used to supplement the shading location. This shadow path does not simply copy the data of neighboring pixels, but rather rhythmically interpolates and adjusts it according to the vertical position of the shading node on the time axis, creating a transitional state consistent with the original path structure in both time and space. By injecting the shadow path into the shading interval, data breaks caused by missing continuous observations are avoided, and the cumulative extension of original outliers is eliminated, providing an intermediate support structure for the stable recovery of surface deformation trends.

[0039] After shadow path injection, to further reduce the rhythm inconsistencies that may be introduced into time series composed of discontinuous data, the time axis needs to be periodically rearranged. Specifically, this involves setting breathing-style time window groups at continuous observation time nodes. This involves introducing equidistant short-period occlusion periods and open periods into the long-period time axis, forming a time-segment structure consisting of alternating static suppression and dynamic response zones. During the occlusion period, phase observations of pixels corresponding to high-risk anchor points are temporarily suspended; during the open period, observations processed by shadow path injection are collected. In this way, the observation nodes across the entire time axis are rhythmically adjusted, establishing a regular data usage pattern. This time window rearrangement not only effectively reduces rhythm disturbances caused by high-frequency perturbations but also improves the consistency of the phase sequence in expressing the overall trend, facilitating the smooth recovery of deformation trends along the time axis.

[0040] After shadow path injection and time window rearrangement, the time series structure of all high-albedo remote sensing pixels is reorganized. The phase change trajectory after removing perturbation values ​​is compared sequentially with the original sequence, and an adjusted surface displacement trend record is output. This trend record exhibits enhanced continuity, abrupt change elimination, and uniform rhythm in the temporal dimension, while maintaining spatial structural integrity. For areas under long-term observation missions, this method is applicable to the dynamic access and real-time processing of subsequent images, possessing continuous application capabilities. Through risk anchor-driven masking and substitution mechanisms, combined with a breathing-style time control structure, not only is error propagation under high-albedo environments effectively suppressed, but a more accurate trend expression basis is also provided for surface deformation monitoring, contributing to the stable operation and accurate response of land subsidence hazard early warning methods under complex interference conditions.

[0041] This invention constructs a multidimensional fingerprint map integrating albedo, humidity, and ground feature brightness, and uses anomalous fingerprint bands as clues to perform cross-temporal identification and disturbance separation processing, effectively avoiding the accumulation of nonlinear errors in radar phase information in high-humidity and high-reflection environments. By introducing a lag time window and combining it with phase inflection point regions for constrained localization, anomalous signals and the actual deformation process are separated in both time and space, significantly improving the stability of deformation trend representation and enhancing the targeted identification of surface disturbances.

[0042] This invention introduces shadow path injection and time window rearrangement mechanisms based on the risk anchor point list. Through rhythmic masking and dynamic replacement, it weakens the interference of high-risk pixels on the overall sequence, ensuring consistency in data continuity and trend extension of surface deformation results. Particularly in areas with frequent albedo jumps or persistent meteorological disturbances, this processing path effectively suppresses error propagation, improving the reliability and practical value of subsidence early warning results.

[0043] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for early warning of land subsidence hazard based on multi-data fusion, characterized in that, Includes the following steps: S1, construct the albedo distribution field, humidity distribution field and ground feature brightness distribution field, fuse the three types of data, combine them with time series changes to generate a multidimensional fingerprint map, and output the abnormal fingerprint band; S2 performs continuous line comparison with the anomalous fingerprint band on multiple remote sensing images, identifies the phase delay change path, extracts the abrupt change segment in the phase delay change as the phase delay inflection point region, and calibrates the lag time window for atmospheric disturbance stripping based on the phase delay inflection point region. S3, initiate the meteorological disturbance stripping process based on the lag time window, use the phase delay inflection point region as the change constraint condition, separate the specular reflection component and the atmospheric water vapor tail component in the remote sensing phase data, and generate the corresponding meteorological stripping data sequence. S4 projects the meteorological stripped data sequence onto a spatiotemporal grid composed of land cover classifications, extracts the locations of abrupt changes in albedo and humidity, and outputs a list of risk anchor points based on spatial distribution characteristics. S5 takes the risk anchor list as input, performs shadow path injection operation based on alternating shading mechanism on high albedo remote sensing pixels, and applies periodically changing time window rearrangement in the time axis direction to offset phase delay error and restore the stability of surface displacement change trend.

2. The method for early warning of ground subsidence hazard based on multi-data fusion according to claim 1, characterized in that, Step S1 includes: The remote sensing image sequence covering the target area is organized in chronological order to construct an albedo distribution field, and the albedo transition state and persistence characteristics of each pixel are marked. In the spatial grid corresponding to the albedo distribution field, a humidity distribution field is constructed, and information on humidity change gradient, humidity propagation direction, and abrupt change points is extracted. A brightness distribution field is constructed based on the albedo distribution field and the humidity distribution field. Brightness fluctuation regions are extracted and the three types of distribution fields are made spatially and temporally consistent. The albedo distribution field, humidity distribution field and brightness distribution field are jointly modeled to generate multidimensional variation curves, and abnormal spatiotemporal fingerprint bands are extracted by sliding comparison and trend superposition.

3. The method for early warning of ground subsidence hazard based on multi-data fusion according to claim 2, characterized in that, Step S2 includes: Within the range of the anomalous spatiotemporal fingerprint, remote sensing image sequences within a continuous time period are selected, and phase change observations at the corresponding spatial locations are extracted to generate a continuous line sequence across the time dimension. Analyze the continuous line sequences, extract the phase change segments with abrupt jump characteristics as abrupt change candidate regions, and further aggregate them into phase change corridors; Based on the phase change corridor, the interference core region with hysteresis effect is calibrated, and a hysteresis time window group is established; By combining the lag time window group with the anomalous spatiotemporal fingerprint band, a time reference structure with a dual constraint mechanism is constructed.

4. The method for early warning of ground subsidence hazard based on multi-data fusion according to claim 3, characterized in that, The starting point of the time window group in the time reference structure is used to trigger the meteorological disturbance stripping process, and the spatial connectivity of the anomalous spatiotemporal fingerprint band is used to limit the stripping range, so that the meteorological disturbance stripping process is only executed within the spatiotemporal unit containing the phase change corridor.

5. The method for early warning of ground subsidence hazard based on multi-data fusion according to claim 3, characterized in that, Step S3 includes: Based on the start and end boundaries of the lag time window, multiple remote sensing image sequences are located, and phase data consistent with the phase delay inflection point area are extracted to construct a set of phase change trajectories. Using the phase delay inflection point region as a variation constraint, the specular reflection component data set is extracted; Based on the identification of specular reflection components, the atmospheric water vapor tail component data set is extracted; The specular reflection component and the atmospheric water vapor tail component are combined to construct a meteorological stripping data sequence, and the start and end time range of the stripping and the source of interference are retained.

6. The method for early warning of ground subsidence hazard based on multi-data fusion according to claim 5, characterized in that, During the construction process, the meteorological stripping data sequence maintains the same spatial resolution and temporal density as the original remote sensing phase data. The source types of specular reflection components and atmospheric water vapor tail components are distinguished by pixel value identification to ensure the continuity and accuracy of subsequent surface displacement trend recovery.

7. The method for early warning of ground subsidence hazard based on multi-data fusion according to claim 5, characterized in that, Step S4 includes: Based on the land cover information of the remote sensing image sequence, a land cover classification layer is constructed, and the meteorological stripping data sequence is projected onto the spatiotemporal grid composed of land cover classification to form an interference response set; Extract the regions that overlap with albedo abrupt changes in the projected data, identify albedo abrupt change points, and generate albedo perturbation bands; Based on the albedo perturbation band, humidity abrupt change regions were extracted, and high-sensitivity regions of humidity abrupt change with spatiotemporal overlap between humidity transition and phase perturbation were screened out. By combining the response intensity and persistence of changes in meteorological stripping data, spatial units that have been in a state of high frequency of disturbances for a long time are selected, and a list of risk anchor points is output.

8. The method for early warning of ground subsidence hazard based on multi-data fusion according to claim 7, characterized in that, The spatial units in the risk anchor point list are sorted according to meteorological stripping intensity and land cover classification characteristics. Anchor points at the intersection of albedo change zones and humidity change high-sensitivity areas are selected as error-sensitive core areas for subsequent image masking and trend restoration processing.

9. The method for early warning of ground subsidence hazard based on multi-data fusion according to claim 7, characterized in that, Step S5 includes: Extract the spatial location set corresponding to high albedo features from the risk anchor list, and construct a time-series disturbance response matrix to form a continuous phase change record; Differentiated masking operations are performed based on the disturbance intensity level of the risk anchor point, and rhythmically adjusted shadow paths are injected into the masking interval to supplement the observation data; After completing the shadow path injection, the timeline is periodically rearranged, and a breathing time window group is set to alternately form occlusion and opening cycles. After the time window is rearranged, the time series structure of high albedo pixels is reorganized, and the adjusted surface displacement trend record is output.

10. The method for early warning of ground subsidence hazard based on multi-data fusion according to claim 9, characterized in that, During the shadow path injection process, the phase response trajectory of neighboring pixels is rhythmically interpolated to ensure a continuous transition of the injection path in time and space. After the shadow path injection is completed, the phase delay error of high albedo pixels is dynamically offset and the stability of the surface displacement trend is restored through the alternating action of breathing time window groups.

Images