Methods, systems and related equipment for detecting underground cavities in karst collapse hazard areas

Abstract

This application relates to the field of geological hazard detection technology, and discloses a method, system, and related equipment for detecting underground cavities in karst collapse hazard areas. The method includes: collecting geological data from the karst area and performing noise filtering and preprocessing to output denoised effective feature data; conducting analysis and modeling based on the effective feature data, using machine learning or rule-based algorithms to complete initial anomaly detection, and outputting cavity anomaly area marker information; constructing a three-dimensional spatial model based on the effective feature data and determining the target area to be detected by combining the marker information; performing underground cavity detection in the target area to obtain cavity existence results; extracting boundary and morphological information to form three-dimensional geometric data of the cavity and calculating the cavity volume when a cavity exists, generating a cavity volume result; updating the three-dimensional spatial model in real time using deployed sensors to continuously correct the cavity existence result and cavity volume result. This method achieves rapid identification, volume calculation, and real-time risk warning of underground cavities.

Description

Technical Field

[0001] This application relates to the field of geological disaster detection technology, and in particular to a method, system and related equipment for detecting underground cavities in karst collapse hazard areas. Background Technology

[0002] Karst collapse is a typical engineering geological hazard. Its precipitating mechanism is usually related to the formation and expansion of underground karst cavities and the instability of the overlying medium, manifesting as sudden surface subsidence or collapse, posing a significant threat to the safety of infrastructure such as buildings, roads, and pipelines. Karst collapse hazard areas are mostly distributed in karst development zones, where the underground medium has a complex structure and strong spatial heterogeneity. Cavity development is characterized by concealment and phased evolution, leading to high uncertainty in hazard identification and risk assessment. Current technologies for identifying and assessing karst collapse hazards typically employ a technical route of "regional investigation—geophysical exploration—drilling verification—long-term observation": on the one hand, risk zoning is conducted through engineering geological surveys, geomorphological and geological structure analysis, and verification of historical collapse data; on the other hand, geophysical exploration (such as electrical resistivity tomography, seismic surveys, and ground-penetrating radar) is used to obtain underground anomaly responses, which are then verified by borehole exposure, sampling, and in-situ testing; in some engineering scenarios, groundwater level and surface subsidence monitoring methods are also implemented, supplemented by manual inspections or periodic re-measurements for hazard tracking. The above methods are widely used in engineering practice and can, to a certain extent, achieve the preliminary delineation and verification of potential hazard areas. However, in engineering applications, the above technical routes still generally have the following technical problems: First, it is difficult to form a stable and verifiable quantitative characterization of the spatial boundaries, morphology, and scale parameters of underground cavities, resulting in insufficient accuracy in hazard location and risk assessment; second, monitoring is mostly based on periodic retesting or manual inspection, making it difficult to continuously track changes in cavities and their related responses, thus failing to meet the need for timely early warning. Summary of the Invention

[0003] The main technical problem addressed by the implementation method of this application is the difficulty of quickly identifying and continuously monitoring and warning of underground cavities in karst collapse hazard areas using traditional technologies.

[0004] To solve the above-mentioned technical problems, the first technical solution adopted in this application is: providing a method for detecting underground cavities in karst collapse hazard areas, comprising: collecting geological data of the karst area, and performing noise filtering and data preprocessing on the geological data to obtain denoised effective feature data; analyzing and modeling the preprocessed effective feature data, performing initial anomaly detection through machine learning algorithms or rule algorithms to obtain marking information of cavity anomaly areas; constructing a three-dimensional spatial model corresponding to the karst area based on the effective feature data, and determining the target area to be detected in the three-dimensional spatial model according to the marking information of the cavity anomaly areas; based on the effective feature data... The system uses the feature data and the three-dimensional spatial model to detect underground cavities in the target area, obtaining a cavity existence detection result. When the cavity existence detection result indicates the existence of an underground cavity, the boundary and morphological information of the underground cavity are extracted from the three-dimensional spatial model as three-dimensional geometric data for volume calculation. The volume of the existing underground cavity is calculated based on the cavity three-dimensional geometric data, obtaining a cavity volume calculation result. The three-dimensional spatial model is updated using real-time monitoring data continuously collected by sensors deployed in the karst area. The cavity existence detection result and the cavity volume calculation result are then updated based on the updated three-dimensional spatial model.

[0005] Optionally, the steps between collecting geological data of the karst region and obtaining the labeling information of the cavity anomaly region include: collecting initial detection data of underground rock strata and cavities through ground-penetrating radar, seismic wave detection, and / or geological drilling, and collecting multi-source monitoring data of the surface and / or underground through a deployed sensor array; performing noise suppression and filtering on the initial detection data and / or the multi-source monitoring data to obtain the denoised effective feature data; performing feature extraction and standardization on the effective feature data to obtain a feature set for anomaly detection; performing initial anomaly detection based on the feature set, wherein, when using a machine learning algorithm, the feature set is classified, and / or clustered, and / or anomaly scoring is output; when using a rule-based algorithm, the feature set is threshold-based and / or multi-index consistency is checked to obtain anomaly detection output; and regionalizing the anomaly data according to the anomaly detection output to generate the labeling information of the cavity anomaly region.

[0006] Optionally, the step of constructing a three-dimensional spatial model corresponding to the karst region based on the effective feature data, and determining the target region to be detected in the three-dimensional spatial model according to the marking information of the cavity anomaly region, includes: fusing and unifying data from different detection devices and / or sensors, and generating a unified data input using a Kalman filter multi-source data fusion method; constructing a three-dimensional spatial structure model corresponding to the karst region based on the unified data input using a finite element analysis three-dimensional modeling method to obtain the three-dimensional spatial model; associating the marking information of the cavity anomaly region in the three-dimensional spatial model, and extracting the corresponding spatial distribution parameters to determine the target region to be detected.

[0007] Optionally, the step of detecting underground cavities in the target area based on the effective feature data and the three-dimensional spatial model to obtain a cavity existence detection result includes: associating the effective feature data within the target area with the spatial structure data in the three-dimensional spatial model to obtain a target area dataset for cavity detection; performing feature extraction and discrimination processing on the target area dataset to construct an input feature set for existence discrimination, and executing a machine learning algorithm or rule-based algorithm to obtain a cavity existence discrimination value; if a machine learning algorithm is executed, the input feature set is used as the model input, and Output the cavity existence discrimination value; if the rule algorithm is executed, perform threshold discrimination and / or combination discrimination on the input feature set, and output the cavity existence discrimination value, which is used to characterize the cavity existence detection result; when the cavity existence detection result indicates the existence of an underground cavity, based on the spatial distribution parameters in the three-dimensional spatial model, predict the morphology, depth, and size parameters of the underground cavity through machine learning algorithms and / or deep learning algorithms, and output the prediction result; perform risk assessment based on the three-dimensional spatial model and the prediction result, and generate a collapse risk assessment result through Monte Carlo simulation.

[0008] Optionally, the step of extracting the boundary and morphological information of the underground cavity from the three-dimensional spatial model as the cavity's three-dimensional geometric data for volume calculation when the cavity existence detection result indicates the existence of an underground cavity includes: extracting the boundary information of the underground cavity from the effective feature data and / or the three-dimensional spatial model using an image processing algorithm based on Hough transform to generate the geometric contour of the underground cavity; converting the geometric contour into a three-dimensional geometric model using finite element analysis based on the boundary information to generate the cavity's three-dimensional geometric data; performing parametric analysis on the three-dimensional geometric model and extracting the cavity's depth, width, and length parameters from the analysis results; and performing accuracy correction on the three-dimensional geometric model based on the collected multi-source verification data to ensure that the cavity's three-dimensional geometric data is consistent with the actual cavity morphology.

[0009] Optionally, the step of calculating the volume of the existing underground cavity based on the three-dimensional geometric data of the cavity to obtain the cavity volume calculation result includes: selecting a volume calculation method based on the geometric features represented by the three-dimensional geometric data of the cavity; when the volume calculation method is Monte Carlo calculation, random sampling is performed within the known boundary area corresponding to the cavity, then the cavity volume is calculated as follows: ; in, Indicates the volume of the cavity. This represents the number of random points that fall within the cavity. This represents the total number of random points. This represents the total simulated region volume. When the volume calculation method is based on a profile function, the cavity profile is represented as a function, and the start and end positions of the profile are determined. The cavity volume is then calculated as follows: ; in, Indicates the volume of the cavity. The functional form representing the cavity profile. These represent the start and end positions of the profile, respectively; the uncertainty analysis of the cavity volume calculation results is performed, and error propagation is conducted to obtain the volume uncertainty in the following way: ; in, Indicates volume uncertainty. Represents the volume of the i-th variable The partial derivatives, This represents the uncertainty of the i-th input quantity or parameter used for volume calculation; when the volume uncertainty exceeds a preset threshold, the cavity three-dimensional geometric data is updated based on the newly acquired data, and the cavity volume calculation result is recalculated based on the updated cavity three-dimensional geometric data.

[0010] Optionally, after the step of calculating the volume of the existing underground cavity based on the cavity's three-dimensional geometric data to obtain the cavity volume calculation result, the method further includes: continuously collecting underground monitoring data of the karst area through sensors; performing real-time analysis on the monitoring data to detect changes in cavity volume and provide early warning of collapse risk; updating the cavity's three-dimensional geometric data and the cavity volume calculation result based on the monitored changes, and reassessing the collapse risk based on the updated data; displaying the cavity data and risk information in a three-dimensional graphical and / or map visualization format, and generating a structured risk analysis report based on the cavity volume calculation result, volume change information, and collapse risk assessment result.

[0011] To solve the above-mentioned technical problems, the second technical solution adopted in this application is: providing a karst collapse hazard area underground cavity detection system, including: a data acquisition and preprocessing module, used to acquire geological data of the karst area, and perform noise filtering and data preprocessing on the geological data to obtain denoised effective feature data; an anomaly detection and area marking module, used to analyze and model the preprocessed effective feature data, and perform initial anomaly detection through machine learning algorithms or rule algorithms to obtain marking information of cavity anomaly areas; a three-dimensional modeling and positioning module, used to construct a three-dimensional spatial model corresponding to the karst area based on the effective feature data, and determine the target area to be detected in the three-dimensional spatial model according to the marking information of the cavity anomaly area; and a cavity existence detection module, used to... The system uses the effective feature data and the three-dimensional spatial model to detect underground cavities in the target area, obtaining a cavity existence detection result. A cavity geometry data extraction module extracts the boundary and morphological information of the underground cavity from the three-dimensional spatial model when the cavity existence detection result indicates the existence of an underground cavity, using this information as the cavity's three-dimensional geometric data for volume calculation. A cavity volume calculation module calculates the volume of the existing underground cavity based on the cavity's three-dimensional geometric data, obtaining a cavity volume calculation result. A real-time monitoring and model update module updates the three-dimensional spatial model using real-time monitoring data continuously collected by sensors deployed in the karst area, and updates the cavity existence detection result and the cavity volume calculation result based on the updated three-dimensional spatial model.

[0012] To solve the above-mentioned technical problems, the third technical solution adopted in the embodiments of this application is: to provide an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the underground cavity detection method in the karst collapse hazard area as described above.

[0013] To solve the above-mentioned technical problems, the fourth technical solution adopted in the embodiments of this application is: to provide a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by an electronic device, the electronic device performs the underground cavity detection method in the karst collapse hazard area as described above.

[0014] Unlike related technologies, this application preprocesses multi-source geological data from karst areas to form stable and effective feature inputs, making subsequent judgments less susceptible to noise and sampling differences, thus improving the consistency and verifiability of cavity identification conclusions from a data perspective. Based on anomaly area marker information and combined with a 3D spatial model, spatial constraints on the target area to be detected are delineated, shifting cavity detection from empirical point selection to quantitative localization under structural constraints. This reduces computational redundancy caused by irrelevant areas and lowers the risk of missing key anomalies. Within the target area, the existence detection result of the cavity is output, and when existence is confirmed, boundary and morphological information is extracted to form 3D geometric data of the cavity for engineering calculations, elevating the detection conclusion from mere existence to a geometric expression that supports volume calculation. Based on the geometric data, the cavity volume calculation result is obtained, providing a quantitative basis for risk assessment and disposal decisions. Simultaneously, real-time sensor monitoring data is introduced to update the 3D spatial model and drive the existence detection result and volume calculation result to be updated synchronously, enabling the detection conclusion to continuously correct for changes in the cavity and its surrounding response, improving the timeliness and stability in continuous monitoring scenarios. Attached Figure Description

[0015] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements having the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0016] Figure 1 This is a schematic diagram of the operating environment for the underground cavity detection method in karst collapse hazard areas provided in this application embodiment.

[0017] Figure 2 This is a schematic diagram of the execution flow of the underground cavity detection method in karst collapse hazard areas provided in the embodiments of this application.

[0018] Figure 3 This is a schematic diagram of the execution flow for determining the target area to be detected in the underground cavity detection method for karst collapse hazard areas provided in the embodiments of this application.

[0019] Figure 4 This is a schematic diagram of the execution flow of extracting data from a three-dimensional spatial model in the method for detecting underground cavities in karst collapse hazard areas provided in the embodiments of this application.

[0020] Figure 5 This is a schematic diagram of the system structure of the underground cavity detection system for karst collapse hazard areas provided in this application embodiment.

[0021] Figure 6 This is a schematic diagram of the hardware structure of the electronic device used in the embodiment of this application for performing the method of detecting underground cavities in karst collapse hazard areas. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. Software tools, components, or servers not belonging to this company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.

[0023] It should be noted that, unless there is a conflict, the various features in the embodiments of this application can be combined with each other, all of which are within the protection scope of this application. Furthermore, although functional modules are divided in the system diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than the module division in the system diagram or the order in the flowchart.

[0024] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0025] To facilitate understanding of this embodiment, a detailed description of the method for detecting underground cavities in karst collapse hazard areas disclosed in this application will be provided first. Please refer to [link to relevant documentation]. Figure 1 , Figure 1 This is a schematic diagram of the operating environment for the method for detecting underground cavities in karst collapse hazard areas provided in this application embodiment, as shown below. Figure 1 As shown, the execution subject of the underground cavity detection method in karst collapse hazard areas provided in this application embodiment is generally an electronic device with a certain computing power, such as a computer. In some possible implementations, this underground cavity detection method in karst collapse hazard areas can be implemented by a processor calling computer-readable instructions stored in the memory. Figure 1 The computer equipment mentioned can be a server. A server can be a standalone server or a cloud server that provides 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 (CDNs), and big data and artificial intelligence platforms. This can be understood as... Figure 1 The number of computer devices shown is merely illustrative and can be expanded in any number according to actual needs.

[0026] Please continue reading. Figure 2 , Figure 2 This is a schematic diagram of the execution flow of the underground cavity detection method in karst collapse hazard areas provided in the embodiments of this application, as shown below. Figure 2 As shown, it includes the following steps: S1. Collect geological data of the karst area, and perform noise filtering and data preprocessing on the geological data to obtain effective feature data after noise reduction.

[0027] The identification of karst collapse hazards relies on the consistency and comparability of multi-source geophysical and monitoring data. Front-end data often contains noise components such as equipment system errors, environmental electromagnetic interference, coupling errors, discrete point spikes, and time-frequency drift. Engineering implementation typically requires calibrating the acquisition equipment, synchronizing time, and registering coordinates. It also involves performing detrending, bandpass filtering / wavelet denoising, spike suppression, amplitude normalization, missing segment processing, and resampling on the signal sequence to align data from different sources in terms of sampling rate, dimensions, and spatial grid. This results in feature representations that can be used for modeling, such as reflection intensity and continuity, arrival time and wave velocity anomalies, spectral energy distribution, and temporal change rate, to improve the signal-to-noise ratio and stability of subsequent identification.

[0028] S2. Analyze and model the preprocessed effective feature data, and perform initial anomaly detection using machine learning algorithms or rule-based algorithms to obtain the marking information of the cavity anomaly region.

[0029] In the anomaly detection stage, a discriminant model is constructed based on the separability of "cavity response" and "normal formation response." The input typically consists of a spatial location index and a multi-dimensional feature set. Machine learning paths usually output anomaly scores or category labels, with typical forms including classification models, clustering models, outlier detection models, and density models. Rule-based paths typically establish discriminant logic around multi-index threshold discrimination and consistency constraints, such as abrupt changes in reflection intensity and continuity disruption, the co-occurrence of arrival anomalies and wave velocity anomalies, and exceeding time-series change rate limits. To avoid false alarms triggered by noise points, engineering implementations often incorporate spatial aggregation and connected component filtering to integrate discrete anomaly points into anomaly region markers with spatial boundaries. These markers can include parameters such as confidence level, area / voxel size, center location, and boundary coordinates, facilitating subsequent 3D modeling, localization, and target region delineation.

[0030] As an optional implementation, the process between collecting geological data of the karst region in step S1 and obtaining the marking information of the cavity anomaly region in step S2 may also include the following steps S11 to S15.

[0031] S11. Initial detection data of underground rock strata and cavities are collected through ground-penetrating radar, seismic wave detection and / or geological drilling, and multi-source monitoring data of the surface and / or underground are collected through deployed sensor arrays.

[0032] Among them, ground-penetrating radar is used to acquire information on the electromagnetic response and reflection interface of the medium, and is suitable for imaging shallow cavities, filling differences, and interface abrupt changes; seismic wave detection is used to acquire information on wave velocity, amplitude attenuation, and arrival anomalies, and is suitable for identifying deep structural discontinuities, low-velocity anomalies, and loose and fractured zones; geological drilling is used to provide the basis for point exposure and sampling, and plays a role in calibration and verification; sensor arrays are used to form continuous, time-series monitoring inputs, and common measurements include underground pressure, pore water pressure, temperature and humidity, displacement / settlement, and vibration. The combined use of multi-source data can cover three levels: "structural anomalies, physical property anomalies, and dynamic responses," enabling anomaly localization to move from single responses to cross-constraints, and improving the reliability and verifiability of anomaly delineation.

[0033] S12. Perform noise suppression and filtering on the initial detection data and / or multi-source monitoring data to obtain denoised effective feature data.

[0034] Initial detection data and monitoring data exhibit different noise mechanisms and sampling structures, requiring separate suppression and filtering based on signal type. Common processing methods for ground-penetrating radar data include DC drift removal, background subtraction, time-zero correction, gain compensation, and bandpass filtering; common processing methods for seismic data include detrending, surface wave / multiple wave removal, power frequency interference removal, and time window filtering; common processing methods for drilling and sensor time-series data include outlier removal, sliding filtering, Kalman filtering, missing data interpolation, and time synchronization. The processing objective is to obtain stable feature inputs that can be used for unified modeling and reduce the disturbance of sporadic noise on anomaly detection thresholds and model discrimination.

[0035] S13. Perform feature extraction and standardization on the effective feature data to obtain a feature set for anomaly detection.

[0036] The feature extraction and standardization stages transform the original waveforms, profiles, or time series into a set of numerical features suitable for algorithm processing, ensuring dimensional consistency and scale alignment. Ground-penetrating radar can extract reflection amplitude, phase axis continuity, spectral energy ratio, attenuation coefficient, and attribute volumes (e.g., instantaneous amplitude / phase); seismic waves can extract travel time residuals, apparent wave velocity, dispersion indices, amplitude attenuation rate, and coherence coefficients; sensors can extract mean drift, rate of change, abrupt change points, periodicity indices, and multi-channel correlation coefficients. Standardization can employ normalization, zero-mean unit variance, quantile scaling, or robust scaling to avoid the dominant bias of "strong signal channels" in model training and discrimination, and to provide a comparable scale for multi-source joint discrimination.

[0037] S14. Perform initial anomaly detection based on the feature set. When using a machine learning algorithm, classify and / or cluster the feature set, and / or output anomaly scores. When using a rule-based algorithm, perform threshold discrimination and / or multi-indicator consistency verification on the feature set to obtain anomaly detection output.

[0038] The initial anomaly detection phase establishes a discrimination logic based on "abnormal responses caused by cavities / loose fracture zones." Machine learning paths can employ supervised classification or unsupervised anomaly detection: supervised classification labels validated point samples as "cavity-related / non-cavity," training a classifier to output anomaly categories and confidence levels; unsupervised paths can use clustering to divide data into several response clusters, or use anomaly scoring models to output outlier levels, identifying samples with significantly different responses from the background. Rule-based paths establish engineering-interpretable discrimination conditions, such as sudden increases in radar reflection intensity and interruptions in the phase axis, the co-occurrence of seismic travel time residuals and low-velocity anomalies, and simultaneous exceedances of monitoring displacement change rates and pressure anomalies. Multi-indicator consistency verification is used to eliminate false anomalies appearing only in a single channel, such as situations where only radar anomalies occur but there is no seismic or monitoring response. Anomaly detection outputs can include anomaly scores, anomaly categories, threshold trigger markers, and multi-source consistency markers, providing a foundation for subsequent regional labeling.

[0039] As an example, after the ground-penetrating radar profile and seismic attribute volume are preprocessed, the exploration section is divided into several small units according to spatial grids or sliding windows. For each unit, features such as reflection amplitude statistics, phase axis continuity index, spectral energy ratio, travel time residual, apparent wave velocity anomaly, and coherence coefficient are extracted. The "cavitary / non-cavitary" samples obtained from drilling or field verification are used as supervision labels. A binary classification model is trained using random forest or gradient boosting tree. During the model inference stage, each unit in the area to be detected outputs a "cavitary existence probability" or "anomaly score". Then, spatial connectivity aggregation and morphological filtering are performed on high-probability units to obtain continuous cavity anomaly area boundaries and labeling information. At the same time, the regional confidence level, area / voxel scale, and center coordinates are output for subsequent 3D modeling, positioning, and target area delineation.

[0040] As another example, the rule-based algorithm decomposes the cavity response into interpretable multi-index discrimination conditions and applies consistency constraints: it detects phase axis interruptions, reflection energy abrupt changes, and attenuation anomalies in ground-penetrating radar data; it calculates travel time residual exceeding limits and low-velocity anomaly characteristics in seismic wave data; and it calculates the duration and amplitude of displacement change rate, pore pressure abrupt changes, or humidity anomalies in sensor time series. When the radar index meets the condition of "continuity below the threshold and reflection energy above the threshold", and the seismic index meets the condition of "travel time residual above the threshold or apparent wave velocity below the threshold", and the sensor index meets the condition of "displacement change rate exceeding the threshold or pressure abrupt change exceeding the threshold", the corresponding spatial unit is identified as a cavity anomaly point. Subsequently, the anomaly points are aggregated according to spatial adjacency, isolated noise spots with area / voxel scale below the threshold are removed, and the boundary, center location, and regional consistency score of the anomaly region are output to form the marking information of the cavity anomaly region.

[0041] S15. Based on the anomaly detection output, the abnormal data is regionalized and labeled to generate marking information for the cavity anomaly area.

[0042] Regionalization labeling transforms discrete anomalies into spatially continuous anomaly regions, facilitating target area delineation and subsequent 3D modeling and localization. Regionalization typically includes spatial aggregation, connected component analysis, morphological processing, and scale filtering: spatial aggregation merges anomalies from adjacent meshes or voxels to form candidate regions; connected component analysis identifies independent anomalies and outputs their boundaries; morphological processing smooths boundaries and fills holes; and scale filtering removes noise spots that are too small in area / volume and lack persistence. The final labeling information usually includes anomaly region boundaries, center location, spatial extent, anomaly score summary, and a summary of temporal changes, supporting target area determination, subsequent exploration strategy development, and dynamic monitoring and updates.

[0043] Through steps S11 to S15, multi-source detection and monitoring data are calibrated, denoised, and characterized under a unified spatiotemporal reference. This transforms the response differences of different detection methods into a unified feature expression that can be fused and discriminated, reducing the impact of equipment noise, environmental interference, and deployment differences on anomaly identification conclusions, thereby improving the stability and verifiability of anomaly detection. On this basis, the anomaly detection output based on feature sets forms quantifiable discrimination results with multi-index constraints, reducing reliance on manual experience interpretation and single profile response, improving the distinguishability between cavity anomaly response and background response, and reducing the probability of false alarms and missed alarms. Furthermore, regional labeling transforms discrete anomaly output into spatially continuous cavity anomaly region marking information, providing clear boundaries and scale basis for 3D modeling, positioning, and delineation of the target area to be detected. This avoids computational redundancy caused by an excessively large target area or omission of key anomalies due to an excessively small target area, thereby improving the efficiency and consistency of subsequent cavity existence detection, boundary morphology extraction, and volume calculation.

[0044] S3. Construct a three-dimensional spatial model corresponding to the karst region based on effective feature data, and determine the target region to be detected in the three-dimensional spatial model according to the marking information of the cavity anomaly region.

[0045] The three-dimensional spatial model construction stage uses effective feature data as a unified input to complete the fusion and expression of multi-source information under the same spatial reference system, elevating the cavity-related response from two-dimensional profiles or discrete points to a spatial structural description. Engineering implementation typically includes time synchronization and coordinate registration of multi-source data, mapping ground-penetrating radar profiles, seismic wave attribute volumes, drilling points, and sensor monitoring points to a three-dimensional mesh or voxel frame; interpolation, resampling, and consistency constraints are applied to data with different spatiotemporal resolutions to form a three-dimensional attribute field or three-dimensional structural model that can be used for structural analysis. The determination of the target area to be detected is constrained by cavity anomaly region marking information. Anomalies or anomaly regions are projected and associated in the three-dimensional spatial model. The detection range is delineated through spatial aggregation, connected component extraction, and boundary expansion buffering operations. Priority detection areas can be selected by combining spatial distribution parameters (e.g., the spatial scale, depth range, and spatial relationship with the surface subsidence zone of the anomaly), thus providing a defined spatial domain and a stable spatial benchmark for subsequent existence detection, boundary extraction, and volume calculation.

[0046] As an alternative implementation method, please continue reading. Figure 3 , Figure 3 This is a flowchart illustrating the process of determining the target region to be detected, as provided in an embodiment of this application. Figure 3 As shown, it may include the following steps S31 to S33.

[0047] S31. Data from different detection devices and / or sensors are fused and processed for consistency, and a unified data input is generated by using Kalman filtering multi-source data fusion.

[0048] Among these, multi-source data fusion and consistency processing aim to establish a unified input for "consistent response of the same geological body across different sensor channels," reducing biases caused by noise from individual devices, differences in survey line layout, and uneven sampling density. Kalman filter fusion typically uses spatial grids or voxels as state carriers, taking radar reflection attributes, seismic attributes, drilling exposure information, and sensor time-series responses as observations. Smooth estimation and noise suppression are achieved through state prediction and observation updates. Consistency processing can introduce dimensional unification, time-scale alignment, outlier removal, and observation weight allocation, allowing high-reliability sources to contribute more to the fusion results. The output is a unified data input for 3D modeling, serving as boundary conditions or parameter fields for subsequent spatial structure solutions.

[0049] S32. Based on unified data input, a three-dimensional spatial structure model corresponding to the karst area is constructed through the three-dimensional modeling method of finite element analysis, and a three-dimensional spatial model is obtained.

[0050] In the finite element 3D modeling stage, a unified data input is transformed into a computable spatial structural representation. A common practice is to establish a 3D mesh within the study area, mapping medium properties, anomaly responses, or equivalent parameters to elements or nodes, and then obtaining a spatially continuous structural model or property distribution through finite element solution. Modeling can use stratigraphic interfaces and cavity anomaly boundaries as geometric constraints, and wave velocity / attenuation anomalies and reflection property anomalies as material parameters or field variables as inputs. The output 3D spatial structural model is used to express the discontinuities and spatial distribution of cavity-related structures, providing a verifiable 3D structural data foundation for subsequent target area delineation and parameter extraction.

[0051] S33. Associate the marking information of the cavity anomaly region in the three-dimensional spatial model, and extract the corresponding spatial distribution parameters to determine the target region to be detected.

[0052] In the target region determination stage, the cavity anomaly region marking information is spatially correlated with the 3D spatial model, transforming the anomaly markings from "identification information" into "operable spatial ranges." Spatial correlation is typically achieved through coordinate mapping, voxel indexing, or spatial querying, locating anomaly points / regions to the set of mesh cells in the 3D model. Spatial distribution parameter extraction can include the depth range, horizontal projection range, voxel size, connectivity, center position, and boundary envelope of the anomaly, used to define the boundaries and priorities of the detection domain. Based on the spatial distribution parameters, target region clipping, boundary buffer expansion, and multi-anomaly partitioning can be performed to form the target region to be detected required for subsequent existence detection, boundary morphology extraction, and volume calculation.

[0053] Through steps S31 to S33, multi-source detection and monitoring data are fused and processed to form a unified data input. This enables data from different devices, with different dimensions and sampling structures to be comparable and jointly constrained within the same spatial reference framework. This reduces the interference of single-source noise or local anomalies on spatial modeling, thereby improving the stability and consistency of the 3D spatial modeling input. The 3D spatial structure model constructed on the basis of the unified input elevates the 2D profile information and discrete point information into a continuous spatial expression. This provides a verifiable 3D structural basis for the spatial distribution, depth range, and connectivity of anomalies, enhancing the spatial interpretation capability of cavity-related responses. Furthermore, the spatial correlation between cavity anomaly region marking information and the 3D spatial model, as well as the extraction of spatial distribution parameters, transforms the determination of the target area to be detected from empirical delineation to quantitative delineation based on spatial structural constraints. This reduces computational redundancy caused by irrelevant areas participating in subsequent detection and lowers the risk of missing key anomalies. This provides clear spatial constraints and a reliable spatial benchmark for subsequent cavity existence detection, boundary morphology extraction, and volume calculation.

[0054] S4. Based on effective feature data and a three-dimensional spatial model, underground cavities are detected in the target area to obtain the cavity existence detection results.

[0055] In the underground cavity existence detection stage, the target area is used as the limited spatial domain. Effective feature data and spatial structural constraints provided by the 3D spatial model are used together to determine whether a cavity exists or not. Engineering implementation typically begins by aligning multi-source features within the target area with 3D mesh cells, forming a local dataset for discrimination. Subsequently, under the constraints of the 3D spatial model, discriminative features reflecting the cavity response are extracted, such as radar reflection interface interruptions and energy mutations, seismic wave velocity anomalies and travel time residual anomalies, attribute field gradient mutations and connected domain structural features, and the anomaly persistence and rate of change of the monitoring time series. The existence discrimination can output a binary conclusion or a threshold-based discrimination value, which characterizes the credibility of the cavity's existence. The role of the 3D spatial model in this stage is to provide constraints such as depth horizon, neighborhood connectivity, and anomaly spatial scale, reducing the impact of false anomalies from a single channel on the discrimination results. This yields verifiable cavity existence detection results and provides triggering conditions for subsequent boundary morphology extraction and volume calculation.

[0056] As an optional implementation, the process of step S4 above may also include the following steps S41 to S46.

[0057] S41. Associate the effective feature data within the target area to be detected with the spatial structure data in the three-dimensional spatial model to obtain the target area dataset for cavity detection.

[0058] In the target region dataset construction phase, the alignment of "data, model, and spatial location" is completed, enabling joint computation of effective feature data and the 3D spatial model under the same spatial indexing system. Engineering implementation typically maps the target region to be detected as a set of 3D mesh cells or voxels, extracting radar attributes, seismic attributes, drilling constraints, and sensor monitoring features within the target region. These features are then correlated with the structural information of the 3D spatial model (e.g., layered interfaces, attribute fields, anomaly candidate regions) to form a target region dataset containing spatial location, structural constraints, and multi-source features, providing a unified input carrier for subsequent existence determination.

[0059] S42. Perform feature extraction and discrimination processing on the target region dataset, construct an input feature set for existence discrimination, and execute machine learning algorithms or rule algorithms to obtain cavity existence discrimination values.

[0060] The existence discrimination input feature set uses "whether a cavity exists" as the discrimination target. Multi-source features are typically further processed into discriminative feature representations, such as spatial gradient, local contrast, connectivity indices, anomaly persistence, temporal change rate, and cross-source consistency indices. The discrimination process outputs a cavity existence discrimination value, which can be represented as a cavity existence probability, anomaly score, or discrete category identifier. The machine learning path focuses on learning the discrimination boundary from the sample distribution, while the rule-based path focuses on achieving verifiable discrimination through engineering-interpretable combinations of criteria. Both are based on the same input feature set, facilitating the selection of implementation routes according to data conditions in different engineering sites.

[0061] S43. If a machine learning algorithm is executed, the input feature set is used as the model input, and the cavity existence discrimination value is output.

[0062] In the machine learning discrimination stage, the input feature set is used as the model input. The model can output an existence discrimination value and confidence level to describe the credibility of the cavity's existence. Supervised learning can form a classifier based on validated samples, outputting "existence / non-existence" and probability; unsupervised anomaly detection can output anomaly scores and convert them into existence discrimination values ​​through thresholding. In engineering implementation, class imbalance handling, threshold calibration, and region-level aggregation are often combined to integrate point-level discrimination results into spatially continuous existence discrimination conclusions, reducing false alarms caused by noise points.

[0063] As an example, after constructing the target region dataset in the target region to be detected, the target region is divided into voxel units or sliding window units, and an input feature set is generated for each unit. The input feature set can consist of ground-penetrating radar reflection attributes (such as local reflection intensity statistics, phase axis continuity index, attenuation coefficient), seismic attributes (such as travel time residual, apparent wave velocity anomaly, coherence coefficient), and sensor time series features (such as displacement change rate, pressure mutation amplitude, anomaly duration), while adding spatial structural constraint features provided by the three-dimensional spatial model (such as depth horizon, neighborhood connectivity, anomaly voxel size). During the machine learning training phase, units corresponding to drilling findings, known cavity locations, or historical collapse points are selected as positive samples, while units with intact structures and no abnormal responses are selected as negative samples. The samples are standardized and missing values ​​are handled. A binary classification model is trained using random forest, gradient boosting tree, or support vector machine. After training, during the inference phase, the input feature set of each unit is input into the model, and the model outputs a cavity existence discrimination value, which can be represented as the cavity existence probability or anomaly score. Subsequently, the discrimination value is thresholded and combined with spatial adjacency relationships to perform region aggregation, resulting in continuous cavity existence determination regions. At the same time, a summary of region-level discrimination values ​​(e.g., maximum value or mean value) can be output to characterize the credibility of the cavity existence detection results within the target region.

[0064] S44. If the rule algorithm is executed, threshold discrimination and / or combination discrimination are performed on the input feature set, and the cavity existence discrimination value is output. The cavity existence discrimination value is used to characterize the cavity existence detection result.

[0065] In this process, the rule-based discrimination stage maps the input feature set into interpretable discrimination conditions. Threshold discrimination is used to identify exceedances of a single indicator, while combined discrimination is used to trigger existence determination when multiple indicators synergistically meet the conditions. Combined discrimination can be embodied in logics such as "simultaneous occurrence of radar structural damage characteristics and low-velocity seismic anomalies" or "spatial connectivity meeting the minimum scale and observing persistent temporal anomalies," suppressing single-source false anomalies through multi-indicator consistency constraints. The existence discrimination value output by the rules can be represented as a binary determination or a hierarchical score, used to characterize the cavity existence detection results and provide triggering conditions for subsequent morphological parameter prediction.

[0066] As another example, after constructing the input feature set for the target area to be detected, the features corresponding to each voxel unit or window unit are divided into three categories: ground-penetrating radar features, seismic wave features, and sensor time-series features. Engineering thresholds and consistency constraints are set for each category of features. In the threshold discrimination stage, single indicators are identified as exceeding limits. For example, a radar anomaly is generated when the continuity of the ground-penetrating radar phase axis is below the threshold and the local reflection intensity is above the threshold; a seismic anomaly is generated when the seismic travel time residual is above the threshold or the apparent wave velocity is below the threshold; and a monitoring anomaly is generated when the displacement rate of change exceeds the threshold or the pressure mutation amplitude exceeds the threshold and continues to exceed the preset time window. In the combined discrimination stage, multiple... The identifiers are logically combined to suppress single-source false anomalies. For example, a cavity is determined to exist when "both radar anomaly identifiers and seismic anomaly identifiers are true" or "the radar anomaly identifier is true and the monitoring anomaly identifier is true and the anomaly voxel size is greater than the minimum size threshold". A cavity is determined not to exist when "only a single identifier is true and the connected volume size is less than the threshold". In the output stage, the combined discrimination results are mapped to a cavity existence discrimination value. The cavity existence discrimination value can be set as a binary identifier or a hierarchical score, and the score can be weighted based on the degree of multi-source consistency, so that the cavity existence discrimination value can characterize the cavity existence detection result and be used for subsequent parameter prediction and risk assessment trigger condition determination.

[0067] S45. When the cavity existence detection result indicates the existence of an underground cavity, based on the spatial distribution parameters in the three-dimensional spatial model, the morphology, depth, and size parameters of the underground cavity are predicted by machine learning algorithms and / or deep learning algorithms, and the prediction results are output.

[0068] The morphological parameter prediction stage, based on the premise of "the existence of a cavity," uses the spatial distribution parameters in the 3D model as input constraints to parametrically estimate the cavity's shape, depth, and size. Machine learning or deep learning can output continuous parameters through regression prediction or scale levels through piecewise prediction. Inputs may include the voxel size, depth distribution, attribute field gradient, boundary coarse profile, and temporal variation characteristics of the anomaly. The output of this stage elevates the existence conclusion into a geometric parameter description usable for engineering evaluation, providing crucial inputs for volumetric calculation, construction avoidance, and risk assessment.

[0069] As another example, after the existence of the cavity is determined, the target area to be detected is clipped into three-dimensional data blocks of fixed spatial scale in the three-dimensional spatial model, and the multi-source data is organized into multi-channel three-dimensional tensor input: the ground-penetrating radar attribute volume is used as the first channel (reflection amplitude, phase axis continuity or attenuation attributes are written into the voxel grid after resampling), the seismic attribute volume is used as the second channel (travel time residual, apparent wave velocity or coherence coefficient is written into the same voxel grid), and the sensor temporal features are used as the third channel (displacement change rate, pressure change amplitude, etc. are mapped to the sensor spatial position and voxel field is generated by interpolation); during the input construction process, coordinate registration, voxel resolution unification, missing value filling and intensity normalization are performed, and candidate masks are generated using cavity anomaly area marker information, so that the network focuses on learning only within the voxel range covered by the candidate mask. Deep learning processing can employ a 3D convolutional network for multi-task prediction: the network backbone performs convolution on the multi-channel voxel field to extract spatial features, the segmentation branch outputs the "cavity voxel probability field", and the regression branch outputs continuous parameters such as "burial depth range, length, width, and height"; during the training phase, drilling and verified 3D geometric data of cavities are used to generate supervision labels, the segmentation branch uses voxel-level intersection-union ratio correlation loss to constrain the cavity boundary, and the regression branch uses absolute error or squared error to constrain the size and burial depth parameters, and introduces layer constraints in the depth direction to avoid unreasonable layer skipping. In the inference phase, the probability field of the cavity voxels is thresholded, and connected component extraction and small voxel cluster removal are performed to obtain the three-dimensional morphological contour of the cavity. The principal axis direction and bounding box size are calculated on the obtained connected components. The projection length and width of the bounding box in the horizontal direction are used as cavity size parameters. The correspondence between the vertical range of the connected components and the depth coordinates of the three-dimensional spatial model is used to obtain the burial depth parameter. Indicators such as the surface area / volume ratio, compactness, and principal axis length-to-length ratio of the connected components are used to characterize the cavity morphological features. Finally, the predicted results of the morphology, depth, and size parameters are output, and the prediction results are written back into the three-dimensional spatial model as input constraints for subsequent risk assessment and volume calculation.

[0070] S46. Risk assessment is conducted based on the three-dimensional spatial model and prediction results, and collapse risk assessment results are generated through Monte Carlo simulation.

[0071] In the risk assessment phase, the three-dimensional spatial model and prediction parameters are used together as conditions for risk calculation, and a quantitative assessment is conducted on the uncertainty of the collapse-causing mechanism. Monte Carlo simulation typically involves randomly sampling key uncertain parameters, repeatedly calculating collapse risk indicators and statistically analyzing their distribution characteristics, and outputting assessment results such as risk level, risk probability, or risk score. Input parameters can come from cavity scale, burial depth, spatial distribution, monitoring change characteristics, etc., and the output results can be used for setting early warning thresholds, prioritizing key areas, and allocating emergency resources, improving the interpretability and verifiability of risk assessment.

[0072] Through steps S41 to S46, the multi-source features and three-dimensional spatial structural constraints within the target area to be detected are organized into a target area dataset for discrimination. This ensures that the cavity existence judgment is based on the consistency of spatial location, structural background, and multi-source response, thereby reducing the impact of single-channel noise or local pseudo-anomalies on the discrimination conclusion and improving the stability and verifiability of the existence detection results. The existence discrimination value obtained by performing machine learning or rule-based discrimination based on the input feature set transforms the cavity existence from an empirical description into a quantifiable discrimination output, facilitating the comparison and thresholding of detection results in different regions and time periods. The system manages and provides clear triggering conditions for subsequent processing. After the existence conclusion is established, it predicts the cavity shape, depth, and size by combining the spatial distribution parameters of the three-dimensional spatial model, so that the detection result is upgraded from "whether it exists" to an engineering output of "geometric parameters are available", which in turn provides key input for volume calculation and risk assessment. By introducing Monte Carlo simulation for risk assessment, the uncertainty of key parameters enters the risk calculation process in a probabilistic statistical manner, so that the collapse risk assessment result can reflect the risk range characteristics brought about by parameter fluctuations, improve the robustness and interpretability of risk assessment, and support the setting of early warning thresholds and the prioritization of emergency response.

[0073] S5. When the cavity existence detection result indicates the existence of an underground cavity, extract the boundary and morphological information of the underground cavity from the three-dimensional spatial model as the three-dimensional geometric data of the cavity for volume calculation.

[0074] As an alternative implementation method, please continue reading. Figure 4 , Figure 4 This is a schematic diagram of the process for extracting data from a three-dimensional spatial model provided in an embodiment of this application, such as... Figure 4 As shown, it may include the following steps S51 to S34.

[0075] S51. The image processing algorithm of Hough transform is used to extract the boundary information of the underground cavity from the effective feature data and / or three-dimensional spatial model to generate the geometric contour of the underground cavity.

[0076] Among them, the Hough transform is suitable for extracting geometrically consistent boundary features from a noisy background. In engineering implementation, effective feature data or slices of 3D spatial models are first converted into edge enhancement representations, such as gradient calculation, threshold segmentation and edge detection of the attribute field. Then, candidate edge points are mapped to the Hough parameter space to accumulate votes and identify line segments, arcs or curve segments with stable geometric features. The geometric features corresponding to the voting peak are back-projected back into the original space to form a boundary point set. Closed or semi-closed cavity geometric contours are generated through connectivity filtering, boundary smoothing and discontinuity completion, providing boundary constraints for subsequent 3D reconstruction.

[0077] S52. Based on boundary information, the geometric contour is transformed into a three-dimensional geometric model using finite element analysis to generate three-dimensional geometric data of the cavity.

[0078] In the finite element 3D modeling stage, the boundary contour is transformed into a computable 3D geometric representation. In engineering implementation, a 3D mesh is usually generated using the boundary point set or contour line as geometric constraints. The contour is then surface-fitted or voxelized to form a 3D solid boundary. The solid is then meshed to obtain a set of elements and nodes. The finite element framework is used here to ensure the continuity and computability of the geometric reconstruction. The output cavity 3D geometric data can be represented as triangular mesh patches, tetrahedral meshes, or voxel solid sets, which facilitates subsequent parameter extraction and volume calculation.

[0079] S53. Perform parametric analysis on the three-dimensional geometric model and extract the depth, width, and length parameters of the cavity from the analysis results.

[0080] In the parametric analysis stage, the 3D geometric model is transformed from a "shape description" into a "set of parameters that can be used for engineering evaluation." In engineering implementation, principal axis analysis and bounding box fitting are performed on the 3D model to obtain dimensional parameters such as length, width, and height. Simultaneously, the vertical range of the model is mapped to the cavity burial depth range using the depth coordinate system of the 3D spatial model. For irregular shapes, the maximum cross-sectional size, average cross-sectional size, and scale distribution can be extracted through cross-sectional scanning to improve the adaptability of the parameters to subsequent risk assessment and design decisions.

[0081] S54. Based on the collected multi-source verification data, the accuracy of the three-dimensional geometric model is corrected so that the three-dimensional geometric data of the cavity is consistent with the actual cavity shape.

[0082] In the accuracy correction stage, multi-source verification data is used to constrain the geometric consistency of the 3D geometric model and correct errors. The goal is to reduce morphological deviations caused by geophysical resolution, boundary extraction errors, and modeling approximations. Multi-source verification data can come from drilling exposure points, sampling information, in-situ tests, field re-measurements, or stable response ranges of monitoring data. The correction process can be achieved through point constraints (convergence of boundary points to verification points), scale constraints (consistency of depth / width / length with the verification range), and local morphological correction (refitting or re-meshing contour segments with large deviations), so that the 3D geometric data of the cavity achieves a verifiable consistency with the actual cavity morphology.

[0083] Through steps S51 to S54, the boundaries and morphology of the underground cavity in the three-dimensional spatial model are transformed from "abnormal response" to "geometrically constrained three-dimensional entity expression," providing a calculable and verifiable geometric description basis for the cavity's outer surface contour and internal morphology. This avoids boundary ambiguity and morphological distortion caused by relying solely on cross-sectional interpretation or discrete points. The establishment and parameter extraction of the three-dimensional geometric model enable key engineering parameters such as cavity depth, width, and length to form a unified output, providing structured input for subsequent volume calculation, risk assessment, and disposal decisions. After multi-source verification data is introduced into the model accuracy correction, the impact of boundary extraction errors, meshing approximation errors, and data resolution differences on the geometric results is constrained. The consistency between the cavity's three-dimensional geometric data and the actual on-site morphology is improved, thereby enhancing the reliability and stability of the cavity volume calculation results and strengthening the traceability and verifiability of the results in engineering applications.

[0084] S6. Calculate the volume of the existing underground cavity based on the three-dimensional geometric data of the cavity, and obtain the cavity volume calculation result.

[0085] As an optional implementation, the process of obtaining the cavity volume calculation result in step S6 above may also include the following steps S61 to S65.

[0086] S61. Select the volume calculation method based on the geometric characteristics represented by the three-dimensional geometric data of the cavity.

[0087] The selection of the volume calculation method falls under the category of modeling strategy matching. The calculation path is determined based on the morphological complexity, boundary analyzability, and data resolution of the cavity's 3D geometric data. Generally, when the boundary shape is irregular, contains gaps, or has significant noise, random sampling calculations are more suitable to mitigate the impact of local errors. When the boundary is continuous and the profile function can be stably fitted, integral calculations are more suitable to achieve higher computational efficiency and repeatability. In engineering practice, indicators such as mesh density, voxel size, and survey line spacing are also considered to avoid selecting calculation paths sensitive to data quality that could lead to unstable volume results.

[0088] S62. When the volume is calculated using Monte Carlo simulation, random sampling is performed within the known boundary region corresponding to the cavity. The cavity volume is then calculated as follows: ; in, Indicates the volume of the cavity. This represents the number of random points that fall within the cavity. This represents the total number of random points. This represents the total simulated region volume. Monte Carlo calculations estimate the cavity proportion by randomly sampling within a known boundary region, transforming the volume estimation of geometrically complex bodies into a probabilistic statistical problem. The proportion of random points falling inside the cavity is proportional to the cavity volume; the more sampling points, the smaller the estimation variance; the closer the sampling region fits the cavity envelope, the higher the estimation efficiency. In engineering implementation, the cavity's circumscribed envelope or a known simulated region is typically constructed first, followed by random point generation and "point inside cavity" determination, and the volume estimate is obtained by statistically analyzing the proportion of points falling. For example, when the cavity boundary is irregular or has a multi-connected structure, random sampling can avoid being overly sensitive to local fitting errors at the boundary.

[0089] S63. When the volume calculation method is based on the profile function, the cavity profile is represented as a function, and the start and end positions of the profile are determined. Then, the cavity volume is calculated as follows: ; in, Indicates the volume of the cavity. The functional form representing the cavity profile. These represent the start and end positions of the profile, respectively. Profile function integral calculation is suitable for scenarios where the cavity has a relatively stable profile representation in a certain direction. The core is to decompose the cavity into continuous profiles along the main direction and integrate and sum the profile area or equivalent function. In engineering implementation, the cutting direction passing through the main axis of the cavity is usually selected first. The profile function form is fitted based on boundary points or voxel sections, and the integration domain is limited by the start and end positions to obtain a volume estimate. For example, when the cavity shape is approximately "elongated" or "lens-shaped," and the profile contour is continuously fittable along the cutting direction, integral calculation is more likely to obtain consistent volume results and facilitates result verification.

[0090] S64. The uncertainty analysis of the cavity volume calculation results is performed, and the error propagation is carried out to obtain the volume uncertainty in the following way: ; in, Indicates volume uncertainty. Represents the volume of the i-th variable The partial derivatives, This represents the uncertainty of the i-th input or parameter used in volume calculation. Uncertainty analysis quantifies the sensitivity of the volume result to input errors, propagating measurement errors, modeling errors, and parameter estimation errors to the final volume result, forming a reliability characterization of the result that can be used for engineering decision-making. Error propagation is typically based on a combination of the volume's sensitivity coefficient to each input and the uncertainty of the input. Inputs can come from boundary point coordinate errors, voxel size errors, profile fitting parameter errors, sampling statistical errors, etc. In engineering applications, uncertainty results are often used to identify dominant error sources and guide the priority of subsequent data acquisition. For example, when boundary point positioning errors contribute the most to the volume, priority should be given to improving boundary extraction accuracy rather than increasing the number of sampling points.

[0091] S65. When the volume uncertainty exceeds the preset threshold, the cavity three-dimensional geometric data is updated based on the newly collected data, and the cavity volume calculation result is recalculated based on the updated cavity three-dimensional geometric data.

[0092] The threshold-triggered update in step S65 establishes a closed-loop mechanism for "quality gating." When the volume uncertainty reaches an unacceptable level, new data is added to improve the stability of geometric constraints and volume calculation. The newly acquired data can come from supplementary survey lines, denser survey points, additional monitoring windows, or higher-resolution detection results. The update process typically involves expanding the boundary point set, refitting the profile, re-subdividing the mesh, or re-estimating the attribute field, followed by recalculating the volume and assessing the uncertainty. In engineering practice, monitoring data from the rainy season or after drainage disturbances can be included in the update trigger conditions. Phased updates maintain consistency between the volume results and the on-site evolution, ensuring the validity of the risk assessment input parameters.

[0093] Through steps S61 to S65, the cavity volume calculation forms a repeatable volume solution process based on the selection of calculation paths through geometric feature matching. This allows complex irregular cavities and cavities with analytical cross-sections to obtain suitable calculation strategies, thereby reducing the sensitivity of a single calculation path to data quality and morphological complexity and improving the stability of the volume results. Simultaneously, the volume uncertainty obtained through error propagation explicitly quantifies the impact of measurement errors, modeling errors, and parameter estimation errors on the volume results, transforming the volume results from a single numerical output to an engineering output with quality characterization, facilitating reliability assessment and decision grading. Furthermore, a threshold-triggered update mechanism based on volume uncertainty forms a quality gating closed loop. When the volume uncertainty exceeds the limit, newly acquired data is introduced to update the three-dimensional geometric data of the cavity and recalculate the volume results, thereby suppressing the risk of misjudgment under high uncertainty conditions and improving the applicability and verifiability of the volume calculation results in dynamic monitoring and risk assessment scenarios.

[0094] As an optional implementation, after step S6 above, which calculates the volume of the existing underground cavity based on the three-dimensional geometric data of the cavity and obtains the cavity volume calculation result, the following steps S66 to S69 may also be included.

[0095] S66. Continuously collect underground monitoring data in karst areas through sensors.

[0096] In the underground monitoring data acquisition phase, continuous, time-series data streams support cavity evolution tracking. The sensor array can cover response quantities such as surface subsidence, underground displacement, pore water pressure, underground pressure, temperature and humidity, and vibration. The monitoring data is mapped to a three-dimensional spatial model through a unified timestamp and spatial location index. For example, displacement gauges and pore water pressure gauges are deployed in karst development zones to form minute-level sampling sequences, which are used to capture abrupt changes in underground response caused by factors such as rainy season replenishment and pumping disturbances, providing continuous input for subsequent volume change identification.

[0097] S67. Perform real-time analysis of monitoring data to detect changes in cavity volume and provide early warning of collapse risk.

[0098] The real-time analysis phase transforms monitoring data from raw time series into state variables that can be used for early warning and judgment. Common processing methods include sliding window statistics, rate of change calculation, abrupt change detection, trend decomposition, and multi-channel consistency verification. It also establishes a correlation between monitored anomalies and changes in cavity volume. For example, when the displacement rate of change exceeds limits in multiple consecutive time windows and pore pressure undergoes synchronous abrupt changes, the anomaly is marked as a "rapid evolution event." Combined with the existing cavity volume baseline, it is determined that the cavity may expand or the overlying medium may become unstable, triggering a collapse risk early warning output.

[0099] S68. Update the three-dimensional geometric data of the cavity and the cavity volume calculation results based on the monitored changes, and reassess the collapse risk based on the updated data.

[0100] The update and reassessment phase involves feeding back monitored change data to the geometric and volumetric layers to ensure the cavity description remains consistent with the on-site response. Updates can be reflected in local corrections of cavity boundaries, re-estimation of cavity height / width parameters, and expansion or contraction of voxel connectivity domains, thereby recalculating the cavity volume. Risk reassessment uses updated parameters such as volume, depth, and rate of change as inputs for risk calculation, outputting a new risk level or risk score. For example, if monitoring shows accelerated local settlement and underground displacement concentrating towards the cavity, the 3D geometric data correspondingly expands the boundary in that direction, leading to an upward adjustment of the volume calculation result and an update of the risk assessment result from "medium risk" to "high risk."

[0101] S69. Display cavity data and risk information in the form of 3D graphics and / or map visualization, and generate a structured risk analysis report based on the cavity volume calculation results, volume change information and collapse risk assessment results.

[0102] The visualization and reporting phase transforms the calculation and evaluation results into a readable and deliverable representation. 3D graphics can be used to display the spatial location, boundary morphology, and volume changes of cavities within the strata, while maps can be overlaid with surface buildings, roads, pipelines, engineering zoning, and risk zoning. Structured risk analysis reports typically include fields such as cavity location and extent, volume and its changing trends, risk level and early warning records, key supporting data summaries, and recommended handling points, facilitating archiving and decision-making processes. For example, the report outputs the cavity center coordinates, burial depth, current volume, 24-hour volume change rate, risk level, and recommended inspection / handling level in tabular fields, along with accompanying 3D view screenshots and risk distribution maps as attachments for engineering review meetings.

[0103] Through steps S66 to S69, real-time analysis driven by continuous sensor monitoring data transforms the cavity status from static judgment to dynamic tracking. Changes in cavity volume are promptly identified and incorporated into the risk warning process, making collapse risk identification timely and reducing the lag risk caused by relying solely on periodic retests. After the change data feedback is used to update the cavity's three-dimensional geometric data and volume calculation results, the cavity description and quantification results can be synchronously corrected with on-site response. Risk assessment is based on the latest geometric and evolutionary information, thereby improving the robustness and consistency of risk assessment results and supporting dynamic adjustment of disposal priorities. Visualization and structured report output form an engineering-readable and deliverable expression of the cavity's spatial location, volume, and change trends, as well as the risk assessment results. This reduces information fragmentation caused by the dispersion of multi-source data, improves cross-disciplinary communication efficiency, and provides traceable evidence for review, inspection, and emergency response.

[0104] S7. Update the three-dimensional spatial model using real-time monitoring data continuously collected by sensors deployed in the karst area, and update the cavity existence detection results and cavity volume calculation results based on the updated three-dimensional spatial model.

[0105] In the real-time monitoring-driven model update stage, the time-series responses continuously acquired by sensors are used as dynamic constraints. An iterative update process for the 3D spatial model is introduced, enabling the model to reflect the time-varying characteristics of the karst medium and cavity-related responses. In engineering implementation, sensor monitoring data is typically synchronized in time, outlier removal, and drift correction are performed before mapping it to the spatial location index of the 3D spatial model. Response quantities such as displacement, pore water pressure, and underground pressure are transformed into spatial attribute fields or boundary condition updates. The 3D spatial model update can manifest as local attribute field reestimation, adjustment of anomalous voxel ranges, fine-tuning of boundary morphology, or reestimation of key parameters, ensuring consistency between the 3D spatial model and the real-time monitoring response. Based on the updated 3D spatial model, the cavity existence detection results are corrected by recalculating the discriminant features and discriminant values ​​of the target area. The cavity volume calculation results are updated by recalculating the updated cavity boundary / morphology data, improving the consistency and stability of identification under continuous monitoring conditions.

[0106] The underground cavity detection method for karst collapse hazard areas provided in this application reduces the interference of noise and sampling differences on the identification conclusions through preprocessing and characterization of multi-source detection and sensor monitoring data, thus providing a stable input basis for anomaly identification. Based on anomaly area marking information and combined with three-dimensional spatial modeling, the method achieves quantitative delineation of the target area under spatial structural constraints, reducing computational redundancy caused by irrelevant areas and lowering the risk of missing key anomalies. It forms a cavity existence discrimination within the target area and extracts cavity boundary and morphological information, outputting three-dimensional geometric data of the cavity that can be used for engineering calculations, thereby supporting the quantitative calculation of cavity volume. Volume uncertainty analysis and threshold-triggered update mechanism make the reliability of the results explicit and form a quality closed loop, improving the verifiability of volume results under data fluctuation conditions. Continuous monitoring-driven model and result updates improve the timeliness and consistency of detection conclusions, and with the output of visualization and structured reports, it provides a deliverable basis for risk assessment and engineering treatment.

[0107] Please continue reading. Figure 5 , Figure 5 This is a schematic diagram of the system structure of the underground cavity detection system for karst collapse hazard areas provided in this application embodiment, as shown below. Figure 5 As shown, the underground cavity detection system 50 in the karst collapse hazard area includes: a data acquisition and preprocessing module 51, an anomaly detection and area marking module 52, a three-dimensional modeling and positioning module 53, a cavity existence detection module 54, a cavity geometric data extraction module 55, a cavity volume calculation module 56, and a real-time monitoring and model update module 57.

[0108] The data acquisition and preprocessing module 51 is specifically used to acquire geological data of karst areas, and to perform noise filtering and data preprocessing on the geological data to obtain effective feature data after noise reduction.

[0109] The anomaly detection and region marking module 52 is specifically used to analyze and model the preprocessed effective feature data, perform initial anomaly detection through machine learning algorithms or rule algorithms, and obtain marking information of cavity anomaly regions.

[0110] The three-dimensional modeling and positioning module 53 is specifically used to construct a three-dimensional spatial model corresponding to the karst region based on the effective feature data, and to determine the target region to be detected in the three-dimensional spatial model according to the marking information of the cavity anomaly region.

[0111] The cavity existence detection module 54 is specifically used to detect underground cavities in the target area to be detected based on the effective feature data and the three-dimensional spatial model, and obtain the cavity existence detection result.

[0112] The cavity geometry data extraction module 55 is specifically used to extract the boundary and morphological information of the underground cavity from the three-dimensional spatial model when the cavity existence detection result indicates the existence of an underground cavity, and use it as the cavity three-dimensional geometry data for volume calculation.

[0113] The cavity volume calculation module 56 is specifically used to calculate the volume of the existing underground cavity based on the cavity's three-dimensional geometric data, and obtain the cavity volume calculation result.

[0114] The real-time monitoring and model update module 57 is specifically used to update the three-dimensional spatial model by continuously collecting real-time monitoring data from sensors already deployed in the karst area, and to update the cavity existence detection result and the cavity volume calculation result based on the updated three-dimensional spatial model.

[0115] As an optional implementation, the data acquisition and preprocessing module 51 is further specifically used to acquire initial detection data of underground rock strata and cavities through ground-penetrating radar, seismic wave detection, and / or geological drilling, and to acquire multi-source monitoring data of the surface and / or underground through a deployed sensor array; to perform noise suppression and filtering on the initial detection data and / or the multi-source monitoring data to obtain the denoised effective feature data; to perform feature extraction and standardization on the effective feature data to obtain a feature set for anomaly detection; to perform initial anomaly detection based on the feature set, wherein, when using a machine learning algorithm, the feature set is classified, and / or clustered, and / or anomaly score output is performed; when using a rule-based algorithm, the feature set is threshold-based and / or multi-index consistency verification is performed to obtain anomaly detection output; and to perform regional labeling on the anomaly data according to the anomaly detection output to generate labeling information of the cavity anomaly area.

[0116] As an optional implementation, the 3D modeling and positioning module 53 is further used to fuse and unify data from different detection devices and / or sensors, generate unified data input by using Kalman filtering multi-source data fusion; based on the unified data input, construct a 3D spatial structure model corresponding to the karst region through finite element analysis 3D modeling, and obtain the 3D spatial model; associate the marking information of the cavity anomaly region in the 3D spatial model, and extract the corresponding spatial distribution parameters to determine the target region to be detected.

[0117] As an optional implementation, the cavity existence detection module 54 is further configured to associate the effective feature data within the target area to be detected with the spatial structure data in the three-dimensional spatial model to obtain a target area dataset for cavity detection; perform feature extraction and discrimination processing on the target area dataset to construct an input feature set for existence discrimination, and execute a machine learning algorithm or rule algorithm to obtain a cavity existence discrimination value; if a machine learning algorithm is executed, the input feature set is used as the model input, and the cavity existence discrimination value is output; if a rule algorithm is executed, threshold discrimination and / or combination discrimination are performed on the input feature set, and the cavity existence discrimination value is output, which is used to characterize the cavity existence detection result; when the cavity existence detection result indicates the existence of an underground cavity, based on the spatial distribution parameters in the three-dimensional spatial model, the morphology, depth, and size parameters of the underground cavity are predicted by machine learning algorithms and / or deep learning algorithms, and the prediction result is output; a risk assessment is performed based on the three-dimensional spatial model and the prediction result, and a collapse risk assessment result is generated by Monte Carlo simulation.

[0118] As an optional implementation, the cavity geometry data extraction module 55 is further configured to extract the boundary information of the underground cavity from the effective feature data and / or the three-dimensional spatial model using an image processing algorithm based on Hough transform, so as to generate the geometric contour of the underground cavity; convert the geometric contour into a three-dimensional geometric model using finite element analysis based on the boundary information, so as to generate the three-dimensional geometric data of the cavity; perform parametric analysis on the three-dimensional geometric model, and extract the depth, width and length parameters of the cavity from the analysis results; and perform accuracy correction on the three-dimensional geometric model based on the collected multi-source verification data, so as to make the three-dimensional geometric data of the cavity consistent with the actual cavity shape.

[0119] As an optional implementation, the cavity volume calculation module 56 is further configured to select a volume calculation method based on the geometric features represented by the three-dimensional geometric data of the cavity; when the volume calculation method is Monte Carlo calculation, random sampling is performed within the known boundary region corresponding to the cavity, and the cavity volume is calculated as follows: ; in, Indicates the volume of the cavity. This represents the number of random points that fall within the cavity. This represents the total number of random points. This represents the total simulated region volume. When the volume calculation method is based on a profile function, the cavity profile is represented as a function, and the start and end positions of the profile are determined. The cavity volume is then calculated as follows: ; in, Indicates the volume of the cavity. The functional form representing the cavity profile. These represent the start and end positions of the profile, respectively; the uncertainty analysis of the cavity volume calculation results is performed, and error propagation is conducted to obtain the volume uncertainty in the following way: ; in, Indicates volume uncertainty. Represents the volume of the i-th variable The partial derivatives, This represents the uncertainty of the i-th input quantity or parameter used for volume calculation; when the volume uncertainty exceeds a preset threshold, the cavity three-dimensional geometric data is updated based on the newly acquired data, and the cavity volume calculation result is recalculated based on the updated cavity three-dimensional geometric data.

[0120] As an optional implementation, the underground cavity detection system 50 for karst collapse hazard areas further includes a cavity data monitoring and analysis module. This module is specifically used to continuously collect underground monitoring data from the karst area via sensors; perform real-time analysis of the monitoring data to detect changes in cavity volume and provide early warnings of collapse risks; update the three-dimensional geometric data of the cavity and the cavity volume calculation results based on the monitored changes, and reassess the collapse risk based on the updated data; display the cavity data and risk information in a three-dimensional graphical and / or map visualization format, and generate a structured risk analysis report based on the cavity volume calculation results, volume change information, and collapse risk assessment results.

[0121] It should be noted that the above-mentioned underground cavity detection system for karst collapse hazard areas can execute the underground cavity detection method for karst collapse hazard areas provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects of the method. Technical details not described in detail in the embodiments of the underground cavity detection system for karst collapse hazard areas can be found in the underground cavity detection method for karst collapse hazard areas provided in the embodiments of this application.

[0122] Figure 6 This is a schematic diagram of the hardware structure of the electronic device for performing the method for detecting underground cavities in karst collapse hazard areas, as provided in the embodiments of this application. Figure 6 As shown, the electronic device 600 includes: One or more processors 610 and memory 620, Figure 6 Take the 610 processor as an example.

[0123] The processor 610 and the memory 620 can be connected via a bus or other means. Figure 6 Taking the example of a connection between China and Israel via a bus.

[0124] The memory 620, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the method for detecting underground cavities in karst collapse hazard areas in this embodiment. The processor 610 executes various server functions and data processing by running the non-volatile software programs, instructions, and modules stored in the memory 620, thereby implementing the method for detecting underground cavities in karst collapse hazard areas described in the above embodiment.

[0125] The memory 620 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the underground cavity detection system for karst collapse hazard areas. Furthermore, the memory 620 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 620 may optionally include memory remotely located relative to the processor 610, and these remote memories can be connected to the underground cavity detection system for karst collapse hazard areas via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0126] The one or more modules are stored in the memory 620. When executed by the one or more processors 610, they perform the underground cavity detection method for karst collapse hazard areas in any of the above method embodiments, for example, performing the above-described... Figure 2 Method steps S1 to S7, Figure 3 Method steps S31 to S33, Figure 4 Steps S51 to S54 in the method are implemented. Figure 5 The functions of modules 51-57 in the document.

[0127] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0128] This application provides a non-volatile computer-readable storage medium storing computer-executable instructions that are executed by one or more processors, for example... Figure 6 One of the processors 610 enables the above-described one or more processors to execute the underground cavity detection method for karst collapse hazard areas in any of the above method embodiments, for example, to execute the above-described... Figure 2 Method steps S1 to S7, Figure 3 Method steps S31 to S33, Figure 4 Steps S51 to S54 in the method are implemented. Figure 5 The functions of modules 51-57 in the document.

[0129] This application provides a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium. The computer program includes program instructions, which, when executed by an electronic device, enable the electronic device to perform the underground cavity detection method for karst collapse hazard areas described in any of the above method embodiments. For example, it can execute the above-described method. Figure 2 Method steps S1 to S7, Figure 3 Method steps S31 to S33, Figure 4 Steps S51 to S54 in the method are implemented. Figure 5 The functions of modules 51-57 in the document.

[0130] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units 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.

[0131] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software and a general-purpose hardware platform, or of course, using hardware. Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of this application as described above, which are not provided in detail for the sake of brevity; although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for detecting underground cavities in a karst collapse hazard area, characterized in that, include: Geological data of karst areas are collected, and noise filtering and data preprocessing are performed on the geological data to obtain effective feature data after noise reduction. The preprocessed effective feature data is analyzed and modeled, and initial anomaly detection is performed using machine learning algorithms or rule-based algorithms to obtain the marking information of the cavity anomaly region; Based on the effective feature data, a three-dimensional spatial model corresponding to the karst region is constructed, and the target region to be detected is determined in the three-dimensional spatial model according to the marking information of the cavity anomaly region; Based on the effective feature data and the three-dimensional spatial model, underground cavity detection is performed on the target area to be detected to obtain the cavity existence detection result. When the cavity existence detection result indicates the existence of an underground cavity, the boundary and morphological information of the underground cavity are extracted from the three-dimensional spatial model and used as the three-dimensional geometric data of the cavity for volume calculation. The volume of the existing underground cavity is calculated based on the three-dimensional geometric data of the cavity, and the cavity volume calculation result is obtained. The three-dimensional spatial model is updated using real-time monitoring data continuously collected by sensors deployed in the karst area. Based on the updated three-dimensional spatial model, the cavity existence detection results and the cavity volume calculation results are updated.

2. The method according to claim 1, wherein, The steps between collecting geological data from the karst region and obtaining the marker information for the cavity anomaly region include: Initial detection data of underground rock strata and cavities are collected through ground-penetrating radar, seismic wave detection and / or geological drilling, and multi-source monitoring data of the surface and / or underground are collected through deployed sensor arrays. The initial detection data and / or the multi-source monitoring data are subjected to noise suppression and filtering to obtain the denoised effective feature data. The effective feature data is subjected to feature extraction and standardization to obtain a feature set for anomaly detection; Initial anomaly detection is performed based on the feature set. When using a machine learning algorithm, the feature set is classified, and / or clustered, and / or anomaly score is output. When using a rule-based algorithm, the feature set is thresholded and / or multi-indicator consistency is checked to obtain anomaly detection output. The abnormal data is regionalized and labeled according to the anomaly detection output to generate the labeling information of the cavity anomaly region.

3. The method according to claim 1, wherein, The step of constructing a three-dimensional spatial model corresponding to the karst region based on the effective feature data, and determining the target region to be detected in the three-dimensional spatial model according to the marking information of the cavity anomaly region, includes: Data from different detection devices and / or sensors are fused and processed for consistency, and a unified data input is generated by using Kalman filtering multi-source data fusion. Based on the unified data input, a three-dimensional spatial structure model corresponding to the karst region is constructed using the three-dimensional modeling method of finite element analysis, thus obtaining the three-dimensional spatial model; The marking information of the cavity anomaly region is associated in the three-dimensional spatial model, and the corresponding spatial distribution parameters are extracted to determine the target region to be detected.

4. The method according to claim 1, wherein, The step of detecting underground cavities in the target area based on the effective feature data and the three-dimensional spatial model to obtain the cavity existence detection result includes: The effective feature data within the target area to be detected are associated with the spatial structure data in the three-dimensional spatial model to obtain a target area dataset for cavity detection. Feature extraction and discrimination processing are performed on the target region dataset to construct an input feature set for existence discrimination, and machine learning algorithms or rule algorithms are executed to obtain cavity existence discrimination values; If a machine learning algorithm is executed, the input feature set is used as the model input, and the cavity existence discrimination value is output. If the rule algorithm is executed, threshold discrimination and / or combination discrimination are performed on the input feature set, and the cavity existence discrimination value is output. The cavity existence discrimination value is used to characterize the cavity existence detection result. When the cavity existence detection result indicates the existence of an underground cavity, based on the spatial distribution parameters in the three-dimensional spatial model, the shape, depth, and size parameters of the underground cavity are predicted by machine learning algorithms and / or deep learning algorithms, and the prediction results are output. Risk assessment is conducted based on the three-dimensional spatial model and the prediction results, and collapse risk assessment results are generated through Monte Carlo simulation.

5. The method for detecting underground cavities in karst collapse hazard areas according to claim 1, characterized in that, The step of extracting the boundary and morphological information of the underground cavity from the three-dimensional spatial model as the three-dimensional geometric data of the cavity for volume calculation when the cavity existence detection result indicates the existence of an underground cavity includes: The Hough transform image processing algorithm is used to extract the boundary information of the underground cavity from the effective feature data and / or the three-dimensional spatial model to generate the geometric contour of the underground cavity; Based on the boundary information, the geometric contour is transformed into a three-dimensional geometric model using finite element analysis to generate the three-dimensional geometric data of the cavity; The three-dimensional geometric model is subjected to parametric analysis, and the depth, width, and length parameters of the cavity are extracted from the analysis results. The accuracy of the three-dimensional geometric model is corrected based on the collected multi-source verification data so that the three-dimensional geometric data of the cavity is consistent with the actual cavity shape.

6. The method for detecting underground cavities in karst collapse hazard areas according to claim 1, characterized in that, The step of calculating the volume of the existing underground cavity based on the three-dimensional geometric data of the cavity, and obtaining the cavity volume calculation result, includes: The volume calculation method is selected based on the geometric features represented by the three-dimensional geometric data of the cavity; When the volume calculation method is Monte Carlo calculation, random sampling is performed within the known boundary region corresponding to the cavity. The cavity volume is then calculated as follows: ； in, Indicates the volume of the cavity. This represents the number of random points that fall within the cavity. This represents the total number of random points. This represents the total volume of the simulation region; When the volume calculation method is based on a profile function, the cavity profile is represented as a function, and the start and end positions of the profile are determined. The cavity volume is then calculated as follows: ； in, Indicates the volume of the cavity. The functional form representing the cavity profile. These indicate the start and end positions of the cross-section, respectively. Uncertainty analysis is performed on the calculated cavity volume, and error propagation is used to obtain the volume uncertainty as follows: ； in, Indicates volume uncertainty. Represents the volume of the i-th variable The partial derivatives, This represents the uncertainty of the i-th input or parameter used for volume calculation; When the volume uncertainty exceeds a preset threshold, the cavity three-dimensional geometric data is updated based on the newly acquired data, and the cavity volume calculation result is recalculated based on the updated cavity three-dimensional geometric data.

7. The method for detecting underground cavities in karst collapse hazard areas according to claim 1, characterized in that, After the step of calculating the volume of the existing underground cavity based on the three-dimensional geometric data of the cavity to obtain the cavity volume calculation result, the method further includes: The underground monitoring data of the karst area is continuously collected through sensors; The monitoring data is analyzed in real time to detect changes in cavity volume and provide early warning of collapse risk. The cavity's three-dimensional geometric data and volume calculation results are updated based on the monitored changes, and the collapse risk is reassessed based on the updated data. The cavity data and risk information are displayed in the form of 3D graphics and / or maps, and a structured risk analysis report is generated based on the cavity volume calculation results, volume change information and collapse risk assessment results.

8. A system for detecting underground cavities in karst collapse hazard areas, characterized in that, include: The data acquisition and preprocessing module is used to acquire geological data of karst areas and perform noise filtering and data preprocessing on the geological data to obtain effective feature data after noise reduction. The anomaly detection and region labeling module is used to analyze and model the preprocessed effective feature data, perform initial anomaly detection through machine learning algorithms or rule algorithms, and obtain labeling information of cavity anomaly regions. The three-dimensional modeling and positioning module is used to construct a three-dimensional spatial model corresponding to the karst region based on the effective feature data, and to determine the target region to be detected in the three-dimensional spatial model according to the marking information of the cavity anomaly region; The cavity existence detection module is used to detect underground cavities in the target area to be detected based on the effective feature data and the three-dimensional spatial model, and obtain the cavity existence detection result. The cavity geometry data extraction module is used to extract the boundary and morphological information of the underground cavity from the three-dimensional spatial model when the cavity existence detection result indicates the existence of an underground cavity, and use it as the cavity three-dimensional geometry data for volume calculation. The cavity volume calculation module is used to calculate the volume of the existing underground cavity based on the cavity's three-dimensional geometric data, and obtain the cavity volume calculation result. The real-time monitoring and model update module is used to update the three-dimensional spatial model by continuously collecting real-time monitoring data from sensors deployed in the karst area, and to update the cavity existence detection results and cavity volume calculation results based on the updated three-dimensional spatial model.

9. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method for detecting underground cavities in karst collapse hazard areas as described in any one of claims 1-7.

10. A non-volatile computer-readable storage medium, characterized in that, The non-volatile computer-readable storage medium stores computer-executable instructions, which, when executed by an electronic device, cause the electronic device to perform the method for detecting underground cavities in karst collapse hazard areas as described in any one of claims 1-7.

Images