A method and system for fast modeling of karst cave based on multi-source data

Abstract

The application provides a kind of karst cave fast modeling method and system based on multi-source data, the method comprises: collecting multi-source data including geological exploration data, remote sensing data and unmanned aerial vehicle aerial data, and pre-processing the multi-source data;The data after pre-processing is fused using weighted fusion or principal component analysis method;Shape feature, size feature and position feature are extracted from the fused data;The extracted karst cave features are converted into point cloud data or grid data, and a karst cave model is constructed using a three-dimensional modeling software;The initial model is edited by vertex, adjusted by face sheet, and the model morphology is corrected in combination with expert experience;The constructed karst cave model is compared with the actual geological data, and if the error exceeds the preset threshold, the model parameters are iteratively adjusted until the accuracy requirement is met.The application can realize efficient extraction, accurate modeling and dynamic optimization of karst geological information.

Description

Technical Field

[0001] This invention relates to the field of geological modeling technology, specifically a method and system for rapid karst cave modeling based on multi-source data, aiming to achieve rapid and accurate construction of karst cave models by integrating multiple data sources. Background Technology

[0002] In the current field of cave geological modeling, with the continuous advancement of information technology, data collection methods have become increasingly diverse, including geological exploration data, remote sensing imagery, and drone aerial photography. However, these data sources are often fragmented, lacking unified data processing standards and fusion mechanisms, resulting in low data utilization efficiency and difficulty in comprehensively and accurately reflecting the true geological characteristics of caves. Furthermore, existing cave modeling methods mostly rely on a single data source, making the modeling process cumbersome and limiting its accuracy, failing to meet the demands for rapid and precise modeling. Therefore, how to efficiently integrate multi-source data to achieve accurate extraction of cave features and rapid construction of 3D models has become a critical issue that urgently needs to be addressed in the current field of cave modeling.

[0003] The main problems with current cave modeling technology lie in insufficient data fusion, inaccurate feature extraction, and a lack of model validation and optimization methods. First, differences in data formats and inconsistencies in coordinate systems among multiple data sources limit effective data fusion, resulting in incomplete and inaccurate foundational data for model construction. Second, cave feature extraction methods are relatively outdated, making it difficult to accurately identify key cave features from complex geological data, thus affecting the model's accuracy and practicality. Finally, model validation and optimization steps lack systematicity and scientific rigor, often relying on manual experience and judgment, making it difficult to ensure the model's accuracy and reliability. These problems collectively constrain the development and application of cave modeling technology. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for rapid modeling of karst caves based on multi-source data. Through multi-source data fusion, intelligent analysis and three-dimensional modeling technology, it can achieve efficient extraction, accurate modeling and dynamic optimization of karst cave geological information, thereby providing high-precision and high-reliability karst cave geological model support for geological research, resource exploration, disaster prevention and other fields.

[0005] A rapid modeling method for karst caves based on multi-source data includes the following steps:

[0006] Step S101: Collect multi-source data including geological exploration data, remote sensing data and UAV aerial photography data, and preprocess the multi-source data;

[0007] Step S102: The preprocessed data are fused using weighted fusion or principal component analysis (PCA) to form a comprehensive dataset of karst cave geological information.

[0008] Step S103: Extract shape features, size features, and position features from the fused data using machine learning algorithms or deep learning models;

[0009] Step S104: Convert the extracted cave features into point cloud data or mesh data, and construct a cave model using 3D modeling software; edit vertices and adjust faces of the initial model using human-computer interaction tools, and correct the model shape based on expert experience;

[0010] Step S105: Compare the constructed cave model with the actual geological data to calculate the spatial error, morphological error and volume error; if the error exceeds the preset threshold, iteratively adjust the model parameters until the accuracy requirements are met.

[0011] Furthermore, the geological exploration data includes borehole data, seismic data, and geological profiles; the remote sensing data includes satellite imagery and radar data; and the UAV aerial photography data includes LiDAR point clouds and high-resolution imagery.

[0012] Furthermore, the preprocessing of the multi-source data includes:

[0013] Data cleaning: removing noise through filtering, interpolating and filling in missing data, and correcting erroneous data;

[0014] Format unification: Format unification includes format conversion and coordinate system unification. Format conversion is the process of converting geological exploration data, remote sensing data and UAV aerial photography data into geospatial raster vector format.

[0015] Data calibration: Based on ground control point pairs, spatial alignment of multi-source data is performed to eliminate spatial biases caused by sensors or platforms; time series of multi-source data are aligned through spectrum analysis to ensure data consistency on the time axis.

[0016] Furthermore, the weighted fusion includes dynamically allocating weights based on the accuracy and resolution of the data source to generate a comprehensive dataset;

[0017] The PCA fusion includes extracting principal components through covariance matrix decomposition, eliminating redundant information, and constructing a dimensionality-reduced dataset.

[0018] Furthermore, step S104 specifically includes:

[0019] Step S501: By integrating, preprocessing, and extracting features from multi-source data, a high-precision cave model is constructed using 3D modeling technology. The extracted cave feature dataset includes shape features, size features, and location features. Shape features include the major axis, minor axis, and orientation angle of the cave. Size features include the diameter or volume of the cave. Location features include the latitude and longitude coordinates or relative position of the cave. The feature data is then converted into point cloud data or mesh data supported by 3D modeling software.

[0020] Point cloud data is generated based on the location and shape characteristics of the cave, assuming the center coordinates of the cave are (x...). c ,y c ,z c Given a point cloud with major axis a, minor axis b, and direction angle θ, the formula for calculating the point cloud coordinates (x, y, z) is:

[0021] x = x c +a·cos(φ)·cos(θ)-b·sin(φ)·sin(θ)

[0022] y = y c +a·cos(φ)·cos(θ)+b·sin(φ)·sin(θ)

[0023] z = z c +c·sin(φ)

[0024] Where φ is the parameter angle, 0≤φ<2π, and c is the height of the cave.

[0025] Point cloud data is converted into mesh data, and triangulation algorithms are used to generate triangular facets. The formula for triangulation is:

[0026]

[0027] Where v1, v2, and v3 are the vertex coordinates of the triangular facet.

[0028] Import the generated mesh data into 3D modeling software, render the 3D model in the software, and set the material and lighting effects;

[0029] Step S502: Through a visual interface and interactive tools, combined with expert experience and user input, the model is dynamically adjusted and optimized. The specific steps are as follows:

[0030] (1) Interactive interface design

[0031] The interface provides basic operations such as translation, rotation, and scaling, as well as advanced functions such as vertex editing and face adjustment, allowing interaction with the model via mouse, keyboard, or touch device;

[0032] (2) Model Adjustment

[0033] Vertex editing: Select a vertex in the model and adjust its position to optimize the model's shape. Assume the original coordinates of vertex vi are (x...). i ,y i ,z i The adjusted coordinates are (x...) i ′,y i ′,z i If ), then the vertex displacement vector is:

[0034] Δv i =(x i ′-x i ,y i ′-y i ,z i ′-z i )

[0035] Face adjustment: Select the triangular facet in the model and adjust its normal direction or area. Assuming the normal vector of the triangular facet is n, and the adjusted normal vector is n′, then the normal rotation matrix is:

[0036]

[0037] Where θ is the rotation angle;

[0038] (3) Real-time preview

[0039] During the adjustment process, the model changes are previewed in real time to ensure that the adjustment effect meets expectations. Layer detail technology is used to dynamically adjust the model details according to the viewing distance.

[0040] Step S503: The model accuracy is continuously optimized by iteratively adjusting model parameters, comparing with actual geological data, and using cross-validation methods. Error analysis and reliability assessment ensure the model's high accuracy and reliability. The cave model parameters include shape, size, volume, topology, physical properties, number and location of vertices and polygons. Specific steps are as follows:

[0041] (1) Model optimization

[0042] The Laplacian smoothing algorithm is used to smooth the model surface, assuming vertex v i If the set of neighboring vertices is N(i), then the smoothed vertex coordinates are:

[0043]

[0044] The model is simplified using an edge folding algorithm to reduce the number of triangular faces. Assume edge e... ij Connect vertex v i and v j The coordinates of the new vertex after folding are:

[0045]

[0046] (2) Model Validation

[0047] Check if the model's geometric properties match the original data. Assuming the model's volume is V and its surface area is S, the calculation formula is:

[0048]

[0049] Where T represents a triangular facet, the optimized model is visualized and compared with the original data to check the accuracy and realism of the model;

[0050] (3) Result Output

[0051] The optimized model is exported to common formats, including OBJ, STL, and FBX.

[0052] A rapid cave modeling system based on multi-source data includes:

[0053] The multi-source data acquisition and preprocessing module is used to collect multi-source data, including geological exploration data, remote sensing data, and UAV aerial photography data, and to preprocess the multi-source data.

[0054] The multi-source data fusion module is used to fuse preprocessed data using weighted fusion or principal component analysis (PCA) methods to form a comprehensive dataset of karst cave geological information.

[0055] The cave feature extraction module is used to extract shape, size, and location features from fused data using machine learning algorithms or deep learning models.

[0056] The 3D modeling and interactive optimization module is used to convert the extracted cave features into point cloud data or mesh data, and to construct a cave model using 3D modeling software; the initial model is edited by vertices and adjusted by facets through human-computer interaction tools, and the model shape is corrected by combining expert experience.

[0057] The model verification and iterative optimization module is used to compare the constructed karst cave model with actual geological data and calculate spatial error, morphological error and volume error. If the error exceeds the preset threshold, the model parameters are iteratively adjusted until the accuracy requirements are met.

[0058] Furthermore, the geological exploration data includes borehole data, seismic data, and geological profiles; the remote sensing data includes satellite imagery and radar data; and the UAV aerial photography data includes LiDAR point clouds and high-resolution imagery.

[0059] Furthermore, the multi-source data acquisition and preprocessing module preprocesses the multi-source data, specifically including:

[0060] Data cleaning: removing noise through filtering, interpolating and filling in missing data, and correcting erroneous data;

[0061] Standardized format: This includes format conversion and unifying the coordinate system;

[0062] Data calibration: Based on ground control point pairs, spatial alignment of multi-source data is performed to eliminate spatial biases caused by sensors or platforms; time series of multi-source data are aligned through spectrum analysis to ensure data consistency on the time axis;

[0063] Format conversion: Convert geological exploration data, remote sensing data, and LiDAR point cloud data into geospatial raster vector format.

[0064] Furthermore, the multi-source data fusion module employs weighted fusion on the preprocessed data, including dynamically allocating weights based on the accuracy and resolution of the data sources to generate a comprehensive dataset; the multi-source data fusion module also employs principal component analysis (PCA) to fuse the preprocessed data, including extracting principal components through covariance matrix decomposition, eliminating redundant information, and constructing a dimensionality-reduced dataset.

[0065] Furthermore, the 3D modeling and interaction optimization module is specifically used for:

[0066] By integrating, preprocessing, and extracting features from multi-source data, a high-precision karst cave model is constructed using 3D modeling technology. The extracted karst cave feature dataset includes shape features, size features, and location features. Shape features include the major axis, minor axis, and orientation angle of the karst cave; size features include the diameter or volume of the karst cave; and location features include the latitude and longitude coordinates or relative position of the karst cave. The feature data is then converted into point cloud data or mesh data supported by 3D modeling software.

[0067] Point cloud data is generated based on the location and shape characteristics of the cave, assuming the center coordinates of the cave are (x...). c ,y c ,z c Given a point cloud with major axis a, minor axis b, and direction angle θ, the formula for calculating the point cloud coordinates (x, y, z) is:

[0068] x = x c +a·cos(φ)·cos(θ)-b·sin(φ)·sin(θ)

[0069] y = y c +a·cos(φ)·cos(θ)+b·sin(φ)·sin(θ)

[0070] z = z c +c·sin(φ)

[0071] Where φ is the parameter angle, 0≤φ<2π, and c is the height of the cave.

[0072] Point cloud data is converted into mesh data, and triangulation algorithms are used to generate triangular facets. The formula for triangulation is:

[0073]

[0074] Where v1, v2, and v3 are the vertex coordinates of the triangular facet.

[0075] Import the generated mesh data into 3D modeling software, render the 3D model in the software, and set the material and lighting effects;

[0076] The model is dynamically adjusted and optimized using a visual interface and interactive tools, combined with expert experience and user input. The specific steps are as follows:

[0077] (1) Interactive interface design

[0078] The interface provides basic operations such as translation, rotation, and scaling, as well as advanced functions such as vertex editing and face adjustment, allowing interaction with the model via mouse, keyboard, or touch device;

[0079] (2) Model Adjustment

[0080] Vertex editing: Select a vertex in the model and adjust its position to optimize the model's shape. Assume the original coordinates of vertex vi are (x...). i ,y i ,z i The adjusted coordinates are (x...) i ′,y i ′,z i If ), then the vertex displacement vector is:

[0081] Δv i =(x i ′-x i ,y i ′-y i ,z i ′-z i )

[0082] Face adjustment: Select the triangular facet in the model and adjust its normal direction or area. Assuming the normal vector of the triangular facet is n, and the adjusted normal vector is n′, then the normal rotation matrix is:

[0083]

[0084] Where θ is the rotation angle;

[0085] (3) Real-time preview

[0086] During the adjustment process, the model changes are previewed in real time to ensure that the adjustment effect meets expectations. Layer detail technology is used to dynamically adjust the model details according to the viewing distance.

[0087] The model accuracy is continuously optimized by iteratively adjusting model parameters, comparing with actual geological data, and using cross-validation methods. Error analysis and reliability assessment ensure the model's high accuracy and reliability. The cave model parameters include shape, size, volume, topology, physical properties, number and location of vertices and polygons. Specific steps are as follows:

[0088] (1) Model optimization

[0089] The Laplacian smoothing algorithm is used to smooth the model surface, assuming vertex v i If the set of neighboring vertices is N(i), then the smoothed vertex coordinates are:

[0090]

[0091] The model is simplified using an edge folding algorithm to reduce the number of triangular faces. Assume edge e... ij Connect vertex v i and v j The coordinates of the new vertex after folding are:

[0092]

[0093] (2) Model Validation

[0094] Check if the model's geometric properties match the original data. Assuming the model's volume is V and its surface area is S, the calculation formula is:

[0095]

[0096] Where T represents a triangular facet, the optimized model is visualized and compared with the original data to check the accuracy and realism of the model;

[0097] (3) Result Output

[0098] The optimized model is exported to common formats, including OBJ, STL, and FBX.

[0099] The present invention has the following beneficial effects:

[0100] 1. This invention achieves efficient fusion of multi-source data, accurate extraction of cave features, and rapid construction, verification, and optimization of 3D models;

[0101] 2. By employing advanced data preprocessing technology, this invention solves the problems of format differences and coordinate system inconsistencies among multi-source data, achieving seamless data fusion;

[0102] 3. By utilizing machine learning algorithms and deep learning models, this invention can accurately extract key features of karst caves from complex geological data, thereby improving the accuracy and practicality of the model.

[0103] 4. This invention continuously optimizes and adjusts model parameters by comparing and verifying with actual geological data, thus ensuring the accuracy and reliability of the model.

[0104] The aforementioned advantages together constitute the core competitiveness of this invention in the field of cave modeling, providing strong technical support for cave geological exploration, mineral resource development and other fields. Attached Figure Description

[0105] Figure 1 This is a flowchart of a rapid cave modeling method based on multi-source data according to an embodiment of the present invention;

[0106] Figure 2 This is a flowchart of multi-source data acquisition and preprocessing in an embodiment of the present invention;

[0107] Figure 3 This is a flowchart of the 3D modeling and interactive optimization of the cave model in this embodiment of the invention;

[0108] Figure 4 This is a flowchart of model verification and iterative optimization in an embodiment of the present invention. Detailed Implementation

[0109] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0110] Please see Figure 1-4 This invention provides a method for rapid modeling of karst caves based on multi-source data, comprising the following steps:

[0111] Step S101: Multi-source data acquisition and preprocessing

[0112] Collect multi-source data such as geological exploration data, remote sensing data, and UAV aerial photography data, and perform preprocessing, including data cleaning, format standardization, and data calibration, to ensure data accuracy and consistency. Specifically, step S101 includes the following steps:

[0113] Step S201: Through multi-source data collection, comprehensively obtain information on the spatial distribution, morphological characteristics, and geological attributes of karst caves, providing reliable data for the accurate construction and verification of the model.

[0114] Specifically, geological exploration data, such as geological exploration reports, borehole data, seismic data, and gravity measurement data, are obtained through on-site exploration, laboratory analysis, and historical data processing. Remote sensing data, such as satellite imagery, aerial imagery, and radar data, are obtained through satellite receiving stations, aerial photography, and radar scanning. High-resolution cameras and LiDAR (Light Detection and Ranging) devices mounted on drones are used to acquire drone aerial photography data, such as LiDAR point clouds and high-resolution imagery, through drone flight missions.

[0115] Step S202: Perform data cleaning on the multi-source data. Specifically, remove noise by filtering, interpolate and fill in missing data, and correct erroneous data to ensure the integrity, consistency and accuracy of the data, providing high-quality data for subsequent model building and optimization.

[0116] Noise removal is performed on geological exploration data, remote sensing data, and UAV aerial photography data using filtering techniques, employing a Gaussian filtering method. The calculation formula is as follows:

[0117]

[0118] In the formula: G(x,y) is the weight value of the Gaussian filter at point (x,y); x and y are the coordinates of the current pixel (relative to the center of the filter); σ is the standard deviation of the Gaussian distribution, controlling the smoothness of the filter. The larger σ is, the smoother the filter. e is the natural constant, approximately equal to 2.71828. π is pi, approximately equal to 3.14159. For smaller datasets, the median filtering method can also be used, where the median of the neighborhood of each pixel is taken as the value of that pixel.

[0119] Missing values ​​in geological exploration data, remote sensing data, and UAV aerial photography data are interpolated. The linear interpolation formula is as follows:

[0120]

[0121] In the formula: y is the value of the interpolation point (the value to be calculated); x is the position of the interpolation point; x1 and x2 are the positions of the known data points; y1 and y2 are the values ​​of the known data points. The Kriging interpolation method is adopted, based on spatial autocorrelation, and uses a variogram function for interpolation.

[0122] Errors in geological exploration data, remote sensing data, and drone aerial photography data are corrected. The least squares method formula is used as follows:

[0123]

[0124] In the formula: y i This represents the actual observed value. pi The model predicts the value. N is the total number of data points. The ∑ (summation) symbol represents minimizing the sum of squared errors over all data points.

[0125] Step S203: Ensure data consistency and comparability by unifying the format of multi-source data (such as format conversion, coordinate system unification) and calibrating the data (such as time synchronization, spatial alignment and unit conversion).

[0126] The specific format conversions are as follows: Geological exploration data typically includes borehole data (point data), geological profiles (line data), and geological boundaries (area data), and the format may be Excel, CSV, or CAD files. Remote sensing data mainly includes satellite imagery and aerial imagery, and the format is mostly GeoTIFF, JPEG, or HDF. UAV aerial photography data includes high-resolution imagery and LiDAR point cloud data, and the format may be JPEG, LAS, or PLY. For geological exploration data, point data (such as borehole data) can be converted from CSV files to Shapefiles or GeoJSON using tools. For example, the Geopandas library in Python can be used to read a CSV file, specify the latitude and longitude fields, and export it as a Shapefile. Line data (such as geological profiles) and area data (such as geological boundaries) can be converted from CAD files to Shapefiles or GeoJSON using QGIS or ArcGIS. The conversion of remote sensing data mainly involves converting JPEG or HDF format to GeoTIFF. This process can be completed using the GDAL command-line tool or the "Raster->Conversion->Translate" tool in QGIS. For multi-band data (such as HDF), GDAL can be used to extract the desired bands and convert them to GeoTIFF. The conversion of drone aerial data includes image data and LiDAR point cloud data. Image data can be converted from JPEG to GeoTIFF using GDAL, while LiDAR point cloud data requires using PDAL or CloudCompare to convert LAS files to a raster format (such as GeoTIFF). For example, the PDAL command-line tool can be used to convert point cloud data to raster via a configuration file.

[0127] All data is converted to a unified coordinate system (e.g., WGS84, EPSG:4326), and the alignment and integrity of the data are verified in GIS software. Through these steps, geological exploration data, remote sensing data, and UAV aerial photography data can be successfully converted to a unified format, providing a reliable foundation for subsequent multi-source data analysis and modeling. The basic formula for unified conversion is:

[0128] x′=a+bx+cy+dxy

[0129] y′=e+fx+gy+hxy

[0130] In the formula: x, y are the coordinates in the original coordinate system. x′, y′ are the coordinates in the transformed coordinate system. a, b, c, d, e, f, g, h are transformation coefficients, calculated using control points.

[0131] Geometric calibration of data at different resolutions is a crucial step in ensuring spatial alignment of multi-source data. Geometric calibration eliminates spatial biases caused by differences in sensors, platforms, or acquisition conditions, enabling precise matching of data at different resolutions within the same coordinate system. Commonly used geometric calibration methods include polynomial calibration, triangulated network (TIN) calibration, and least squares matching. The polynomial calibration formula for geometric calibration of data at different resolutions is as follows:

[0132] x′=a0+a1x+a2y+a3x 2 +a4xy+a5y 2

[0133] y′=b0+b1x+b2y+b3x 2 +b4xy+b5y 2

[0134] In the formula: x, y are the coordinates in the original image. x′, y′ are the coordinates in the corrected image. a0, a1, a2, a3, a4, a5 and b0, b1, b2, b3, b4, b5 are polynomial coefficients, calculated using control points.

[0135] The steps of geometric calibration are as follows: First, collect control points (GCPs). These points can be ground markers with known coordinates or feature points extracted from a high-resolution reference image. The quantity and quality of control points directly affect the calibration accuracy and they typically need to be evenly distributed throughout the image. Second, establish a geometric calibration model using the control points. For example, in polynomial calibration, the coordinates of the control points are used to calculate polynomial coefficients, mapping the pixel coordinates of the original image to the reference coordinate system. Finally, apply the calibration model to resample the image, generating the calibrated image. The resampling method can be nearest neighbor, bilinear interpolation, or cubic convolution interpolation; the specific choice depends on the data resolution and application requirements.

[0136] Taking remote sensing imagery as an example, the steps for polynomial correction using the GDAL tool are as follows: First, prepare a control point file (e.g., CSV format), containing the original image coordinates and reference coordinates. Then, use the GDAL gdalwarp command to perform the correction.

[0137] Time series calibration was performed on data collected from multiple sources at different times, including geological exploration data, remote sensing data, and UAV aerial photography data, using the following Fourier transform formula:

[0138]

[0139] In the formula: F(k) is the k-th frequency component in the frequency domain. f(n) is the n-th sampling point in the time domain. N is the total number of sampling points. k is the frequency index (k = 0, 1, 2, ..., N-1). I is the imaginary unit, satisfying i² = -1. E is the natural constant, approximately equal to 2.71828. π is pi, approximately equal to 3.14159.

[0140] Fourier transforms were performed on geological exploration data, remote sensing data, and UAV aerial data to extract their respective spectral characteristics. Significant temporal periodic components in each data set were identified by analyzing the amplitude and phase information of the spectra. Based on the spectral analysis results, the time offset between different data sets was calculated. For example, if there is a phase difference between remote sensing data and UAV aerial data at a certain frequency component, the time offset can be calculated using the relationship between phase difference and frequency. For geological exploration data, if its spectral characteristics are delayed compared to other data, this can also be quantified using phase information. These time offsets were used to perform time correction on the data, ensuring alignment of multi-source data on the time axis. For example, the time series of UAV aerial data was shifted forward or backward to align with the time characteristics of remote sensing data. After time correction, the results were verified. By recalculating the spectral characteristics of the corrected data, it was checked whether the phase difference between different data sets was significantly reduced or eliminated. Furthermore, the consistency and reliability of the corrected data can be verified in conjunction with practical application scenarios, such as geological activity monitoring or ecological environment assessment. Ultimately, the corrected multi-source data will be used for comprehensive analysis to provide more accurate time series support for geological exploration, environmental monitoring, and other fields.

[0141] Step S102: Using advanced data fusion algorithms, the preprocessed multi-source data are fused to form a comprehensive karst cave geological information dataset.

[0142] Specifically, preprocessed geological exploration data, remote sensing data, and UAV aerial photography data are fused together. Weighted fusion is a simple and effective data fusion method that assigns weights to different data sources, comprehensively considering the importance of each data source to generate a fused result. The specific steps are as follows:

[0143] 1. Data source feature extraction

[0144] Feature values ​​are extracted from the preprocessed geological exploration data, remote sensing data, and UAV aerial photography data. For example, features such as depth and density of underground structures can be extracted from geological exploration data, surface indices and temperature can be extracted from remote sensing data, and surface texture and elevation can be extracted from UAV aerial photography data. Let's assume the extracted feature values ​​are x1, x2, and x3, respectively.

[0145] 2. Weight Allocation

[0146] Based on the accuracy, resolution, or expert experience of the data source, assign weights w1, w2, and w3 to each data source, such that w1 + w2 + w3 = 1. For example, geological exploration data has high accuracy in underground structure information and can be assigned a larger weight (e.g., w1 = 0.5); remote sensing data has advantages in land cover information and can be assigned a medium weight (e.g., w2 = 0.3); and UAV aerial photography data has high resolution in surface details and can be assigned a smaller weight (e.g., w3 = 0.2).

[0147] 3. Weighted fusion calculation

[0148] Calculate the fused eigenvalues ​​y using the weighted fusion formula:

[0149] y = w1·x1 + w2·x2 + w3·x3

[0150] For example, if x1 = 10 (geological exploration data feature value), x2 = 8 (remote sensing data feature value), and x3 = 6 (UAV aerial photography data feature value), then the fusion result is:

[0151] y=0.5·10+0.3·8+0.2·6=5+2.4+1.2=8.6

[0152] 4. Result Generation

[0153] All feature values ​​are weighted and fused to generate a comprehensive dataset of karst cave geological information. This method is simple and intuitive, and suitable for scenarios where the data source features are relatively clear and the weights are easy to determine.

[0154] Principal Component Analysis (PCA) is a commonly used dimensionality reduction method that can extract the main features from multi-source data, eliminate redundant information, and retain key information. The specific steps are as follows:

[0155] 1. Data Matrix Construction

[0156] The preprocessed geological exploration data, remote sensing data, and UAV aerial photography data are used to construct a data matrix X, where each column represents a feature vector from one data source. Assuming there are n samples and m features are extracted from each data source, the dimension of data matrix X is n×3m.

[0157] 2. Covariance matrix calculation

[0158] Calculate the covariance matrix C of the data matrix X:

[0159] C = (1 / n)XTX

[0160] The covariance matrix reflects the correlation between different features.

[0161] 3. Eigenvalue decomposition

[0162] The covariance matrix C is decomposed into eigenvalues ​​and eigenvectors. The eigenvalues ​​represent the magnitude of the variance of each principal component, and the eigenvectors represent the direction of each principal component.

[0163] 4. Principal Component Selection

[0164] Based on the magnitude of the eigenvalues, select the first k principal components (usually those with a cumulative variance contribution rate reaching a certain threshold, such as 95%). Construct a projection matrix P, whose column vectors are the first k eigenvectors.

[0165] 5. Data dimensionality reduction and fusion

[0166] Projecting the original data matrix X onto a lower-dimensional space yields the dimensionality-reduced fused dataset Y:

[0167] Y = X·P

[0168] For example, if the original data matrix X has a dimension of 100×9 (100 samples, 9 features), and the first 3 principal components are selected, then the dimension of the projection matrix P is 9×3, and the dimension of the dimensionality-reduced dataset Y is 100×3.

[0169] 6. Result Generation

[0170] The dimensionality-reduced dataset Y is the fused comprehensive karst cave geological information dataset. PCA can effectively reduce data dimensionality while retaining the main features of multi-source data, making it suitable for high-dimensional data fusion scenarios.

[0171] Cross-validation is performed on the fused dataset to evaluate its accuracy and reliability. The dataset is divided into training and test sets. The model is trained on the training set and validated on the test set. For example, K-fold cross-validation can be used, dividing the dataset into K subsets, using each subset sequentially as the test set and the remaining subsets as the training set, repeating this process K times to calculate the model's average performance metrics (such as accuracy, recall, F1 score, etc.). Cross-validation comprehensively evaluates the stability and generalization ability of the fused result. For example, in a karst cave geological information dataset, if the model's performance metrics fluctuate little across different subsets, it indicates that the fused result has high reliability.

[0172] The fused cave geological information is compared with field survey data to further verify its accuracy. For example, the distribution, depth, and morphology of caves in the fused results are compared item by item with known field survey data, and consistency indices (such as mean square error and correlation coefficient) are calculated. If the fused results are highly consistent with the field data in key features, the fusion algorithm is effective; if there are significant differences, the reasons need to be analyzed, which may be due to unreasonable weight allocation of the data sources or improper parameter settings of the fusion algorithm. For example, if the cave depth in the fused results deviates significantly from the field data, it may be necessary to adjust the weights of the geological exploration data or optimize the fusion algorithm.

[0173] Based on the validation results, the parameters or weights of the fusion algorithm are adjusted to further optimize the fusion effect. For example, in the weighted fusion method, if the weight of a certain data source is too high, leading to a large deviation in the fusion result, the weights can be reallocated to better meet the actual needs. In the PCA fusion method, if important features are lost after dimensionality reduction, the number of principal components can be increased or the feature extraction method can be adjusted. Furthermore, machine learning methods (such as grid search or Bayesian optimization) can be combined to automatically find the optimal parameter combination. Through multiple iterative optimizations, a high-quality comprehensive karst cave geological information dataset is finally formed, providing reliable data support for applications such as geological exploration and disaster early warning. For example, the optimized dataset can more accurately predict the distribution of karst caves.

[0174] Step S103: Using machine learning algorithms or deep learning models, extract the shape, size, and location features of the caves from the fused dataset.

[0175] Deep learning models are used to extract features from karst caves. Deep learning models, especially convolutional neural networks (CNNs), are widely used in spatial data (such as remote sensing images and geological data) analysis due to their powerful feature extraction capabilities. For the task of karst cave location extraction, CNNs can automatically learn spatial features from multi-source fused data and output accurate location features.

[0176] The input data is a fused dataset, including geological exploration data, remote sensing data, and UAV aerial photography data. This data can be represented as a multidimensional matrix, for example: geological exploration data containing features such as depth and density; remote sensing data containing features such as terrain texture; and UAV aerial photography data containing features such as elevation and texture. The multi-source data is integrated into a multidimensional tensor, for example, with a shape of n×h×w×c, where n is the number of samples, h and w are spatial dimensions (such as height and width), and c is the number of feature channels.

[0177] (1) Design of a convolutional neural network (CNN) architecture for cave data analysis:

[0178] 1. In the input layer, multidimensional tensors are received as input.

[0179] 2. In the convolutional layer, local features are extracted using convolution kernels. The convolution operation formula is:

[0180]

[0181] Where wab is the convolution kernel weight, b is the bias term, and σ is the activation function (such as ReLU).

[0182] 3. In the pooling layer, reduce the feature map dimensionality using max pooling or average pooling. The max pooling formula is: y ij =max(x i·s,j·s ,...,x i·s+k-1,j·s+k-1 )

[0183] Where s is the step size and k is the pooling window size.

[0184] 4. In the fully connected layer, the features extracted by the convolutional and pooling layers are mapped to the output space.

[0185] 5. In the output layer, output the location characteristics of the cave, such as latitude and longitude coordinates or relative position.

[0186] The mean squared error (MSE) is used as the loss function to measure the difference between the predicted and actual locations:

[0187]

[0188] Where yi is the actual location and yi is the predicted location.

[0189] The Adam optimization algorithm is used to update model parameters, combined with a learning rate decay strategy to improve training efficiency. Overfitting is prevented using Dropout or L2 regularization.

[0190] Use root mean square error (RMSE) or mean absolute error (MAE) to evaluate model performance:

[0191]

[0192] Specifically, a dataset of geological information about karst caves in a certain region contains 1000 samples. Model training: 80% of the data was used as the training set and 20% as the test set to train a CNN model. The model achieved an RMSE of 0.05 and a MAE of 0.03 on the test set, indicating high prediction accuracy.

[0193] (2) Feature integration and result output

[0194] 1. Feature integration method

[0195] Feature integration involves unifying the extracted features of cave shape, size, and location to generate a complete dataset of key cave features. Integration methods include: concatenating different feature vectors column-wise to generate a comprehensive feature matrix; and assigning weights based on feature importance to generate weighted feature vectors.

[0196] 2. Feature Database Construction

[0197] Design a relational database (such as MySQL) or a non-relational database (such as MongoDB) to store cave feature data. Create a table named `cave_features` containing fields for `shape`, `size`, and `location`. Import the integrated feature data into the database for subsequent querying and analysis.

[0198] 3. Results visualization

[0199] Use tools such as Matplotlib, Seaborn, or Tableau for data visualization. Use 3D plots to show the shape, size, and location of caves. Use heatmaps to show the distribution density of caves. In 3D plots, the X and Y axes represent the location of caves, the Z axis represents the depth of caves, and color represents the size of caves.

[0200] Specifically, a geological information dataset of karst caves in a certain region contains 500 samples. The extracted shape, size, and location features of the caves are integrated into a comprehensive feature dataset. The integrated data is then imported into a MySQL database, where the table `cave_features` contains the fields `shape`, `size`, and `location`.

[0201] Step S104: Using the extracted cave features, construct a 3D model of the cave using 3D modeling software or platform. Simultaneously, fine-tune and optimize the model through human-computer interaction to ensure its accuracy and realism. Step S104 specifically includes the following steps:

[0202] Step S501: By integrating, preprocessing and extracting features from multi-source data, a high-precision cave model is constructed using 3D modeling technology.

[0203] The extracted cave feature dataset includes shape, size, and location features. Shape features include the cave's major axis, minor axis, and orientation angle. Size features include the cave's diameter or volume. Location features include the cave's latitude and longitude coordinates or relative position. The feature data is then converted to a format supported by 3D modeling software, such as point cloud data or mesh data.

[0204] Point cloud data is generated based on the location and shape characteristics of the cave. Assume the center coordinates of the cave are (x...).c ,y c ,z c Given a point cloud with major axis a, minor axis b, and direction angle θ, the formula for calculating the point cloud coordinates (x, y, z) is:

[0205] x = x c +a·cos(φ)·cos(θ)-b·sin(φ)·sin(θ)

[0206] y = y c +a·cos(φ)·cos(θ)+b·sin(φ)·sin(θ)

[0207] z = z c +c·sin(φ)

[0208] Where φ is the parameter angle (0≤φ<2π), and c is the height of the cave.

[0209] Point cloud data is converted into mesh data, and triangulation algorithms (such as Delaunay triangulation) are used to generate triangular facets. The formula for triangulation is:

[0210]

[0211] Where v1, v2, and v3 are the vertex coordinates of the triangular facet.

[0212] Import the generated mesh data using 3D modeling software (such as Blender, AutoCAD, MeshLab) or platforms (such as Unity, Unreal Engine). Render the 3D model in the software, setting materials and lighting effects to make the model more realistic.

[0213] Step S502: Dynamically adjust and optimize the model by using a visual interface and interactive tools (such as 3D editing, parameter adjustment, and real-time feedback) combined with expert experience and user input.

[0214] 1. Interactive Interface Design

[0215] The interface provides basic operations such as translation, rotation, and scaling, as well as advanced functions such as vertex editing and face adjustment. Users can interact with the model using a mouse, keyboard, or touch device.

[0216] 2. Model Adjustment

[0217] Vertex Editing: Select vertices in the model and adjust their positions to optimize the model's shape. Assume the original coordinates of vertex vi are (x...). i ,y i ,z i The adjusted coordinates are (x...) i ′,y i′,z i If ), then the vertex displacement vector is:

[0218] Δv i =(x i ′-x i ,y i ′-y i ,z i ′-z i )

[0219] Face adjustment: Select the triangular facet in the model and adjust its normal direction or area. Assuming the normal vector of the triangular facet is n, and the adjusted normal vector is n′, then the normal rotation matrix is:

[0220]

[0221] Where θ is the rotation angle.

[0222] 3. Real-time preview

[0223] During the adjustment process, model changes are previewed in real time to ensure that the adjustment effect meets expectations. Level of Detail (LOD) technology is used to dynamically adjust model details based on the viewing distance, improving the smoothness of interaction.

[0224] Step S503: Continuously optimize the model accuracy by iteratively adjusting model parameters, comparing with actual geological data, and using cross-validation methods, and ensure the high accuracy and reliability of the model through error analysis and reliability assessment.

[0225] The parameters of a cave model include shape, size, volume, topology, physical properties, number and location of vertices and polymorphs.

[0226] 1. Model Optimization

[0227] The Laplacian smoothing algorithm is used to smooth the model surface. Assuming the set of neighboring vertices of vertex vi is N(i), the smoothed vertex coordinates are:

[0228]

[0229] The model is simplified using an edge folding algorithm, reducing the number of triangular faces. Assume edge e... ij Connect vertex v i and v j The coordinates of the new vertex after folding are:

[0230]

[0231] 2. Model Validation

[0232] Check that the model's geometric properties (such as volume and surface area) match the original data. Assuming the model's volume is V and its surface area is S, the calculation formula is:

[0233]

[0234] Where T represents a triangular facet. The optimized model is then visualized and compared with the original data to check the model's accuracy and realism.

[0235] 3. Result Output

[0236] Export the optimized model to common formats (such as OBJ, STL, FBX) for easy subsequent application.

[0237] Step S105: Compare and verify the constructed karst cave model with actual geological data. Optimize and adjust the model based on the verification results to improve its accuracy and reliability. Step S105 specifically includes:

[0238] Step S601: By calculating indicators such as spatial error, morphological error and volumetric error, comprehensively evaluate the consistency between the model and the actual geological data.

[0239] Define indicators for comparing models with actual geological data, including spatial errors in spatial location, morphological errors in morphological features, and volumetric errors in volume.

[0240] Use Euclidean distance to calculate the spatial error between model points and actual geological data points:

[0241]

[0242] Among them, (x m ,y m ,z m (x) represents the coordinates of the model point. g ,y g ,z g () represents the coordinates of the actual geological data points.

[0243] Calculate shape error using curvature or surface normal vector difference:

[0244] E f =||n m -n g ||

[0245] Where, n m Let n be the surface normal vector of the model. g This is the surface normal vector of the actual geological data.

[0246] Volumetric error between the computational model and actual geological data:

[0247] E v =|V m -V g |

[0248] Among them, V m V is the model volume. g This represents the actual volume of geological data.

[0249] S602: By analyzing the error distribution characteristics and adjusting the model parameters, an iterative optimization method is used to gradually improve the model accuracy until the preset error threshold is met.

[0250] Draw an error distribution map and analyze the spatial distribution characteristics of the errors. Identify areas with large errors and determine the key areas for optimization.

[0251] Based on the error analysis results, adjust the model parameters. Optimize the model point locations using the least squares method:

[0252]

[0253] The surface morphology of the model is adjusted using a curvature optimization algorithm; the model scale is adjusted based on volume error.

[0254] Repeat the comparison, analysis, and adjustment process until the model accuracy meets the requirements. Set an iteration termination condition, for example:

[0255] max(E sp E f E v )<ε

[0256] Where ε is a preset error threshold.

[0257] S603: Through cross-validation and uncertainty analysis, the generalization ability and reliability of the model are comprehensively evaluated to ensure that the optimized model has high accuracy and stability in practical applications.

[0258] Geological data is divided into training and validation sets, used for model optimization and validation respectively, to evaluate the model's generalization ability. The confidence interval or uncertainty range of the model is calculated to assess its reliability. For example:

[0259]

[0260] Among them, E i Let be the i-th error value, and E be the average error.

[0261] S604: In the model output and application phase, the optimized model is output in a standardized format (such as a mesh model or point cloud data) and applied to actual engineering or geological analysis. At the same time, feedback information is collected to further optimize model performance.

[0262] During the model output stage, the optimized cave model is converted into a usable standardized format, such as a mesh model or point cloud data, to ensure broad compatibility and applicability across different software platforms and tools. Mesh models describe the surface morphology of the cave using triangular or quadrilateral facets, suitable for visualization, simulation, and analysis; point cloud data records the spatial distribution of the cave as discrete points, suitable for high-precision reconstruction and detailed processing. By selecting an appropriate output format, it is ensured that the model meets the specific needs of subsequent engineering or geological analysis.

[0263] In the model application phase, the output karst cave model is deployed in actual engineering or geological analysis, such as karst cave stability assessment, groundwater flow simulation, tunnel design, or geological hazard prediction. By integrating with real-world scenarios, the model's performance in practical applications is validated, and relevant data and feedback are collected. For example, in tunnel engineering, the model is used to predict the impact of karst caves on construction; in geological analysis, the model is used to study the formation mechanisms and evolution patterns of karst caves. The goal of this phase is to ensure that the model not only possesses high theoretical accuracy but also demonstrates its value in practical applications.

[0264] During the feedback and optimization phase, the model is further adjusted and optimized based on feedback information from practical applications. For example, if the model's predictions in a certain area deviate significantly from actual observations, this can be corrected by recalibrating parameters or introducing new data sources. Through continuous iterative optimization, the model's accuracy and reliability are improved, enabling it to better serve practical engineering and geological research. This process reflects the dynamic and continuous improvement characteristics of model building, with the ultimate goal of achieving a high degree of alignment between the model and actual needs.

[0265] This invention also provides a rapid cave modeling system based on multi-source data, comprising:

[0266] The multi-source data acquisition and preprocessing module is used to collect multi-source data, including geological exploration data, remote sensing data, and UAV aerial photography data, and to preprocess the multi-source data.

[0267] The multi-source data fusion module is used to fuse preprocessed data using weighted fusion or principal component analysis (PCA) methods to form a comprehensive dataset of karst cave geological information.

[0268] The cave feature extraction module is used to extract shape, size, and location features from fused data using machine learning algorithms or deep learning models.

[0269] The 3D modeling and interactive optimization module is used to convert the extracted cave features into point cloud data or mesh data, and to construct a cave model using 3D modeling software; the initial model is edited by vertices and adjusted by facets through human-computer interaction tools, and the model shape is corrected by combining expert experience.

[0270] The model verification and iterative optimization module is used to compare the constructed karst cave model with actual geological data and calculate spatial error, morphological error and volume error. If the error exceeds the preset threshold, the model parameters are iteratively adjusted until the accuracy requirements are met.

[0271] This invention significantly improves the efficiency and accuracy of karst cave geological research by organically combining multi-source data fusion, intelligent analysis, and 3D modeling technologies. First, it integrates multimodal data from geological exploration, remote sensing, and UAV aerial photography, breaking through the spatial and resolution limitations of single data sources and comprehensively capturing the complex features of karst cave morphology and geological environment. Second, it utilizes machine learning / deep learning algorithms to automate the extraction of karst cave features, replacing traditional manual analysis methods, reducing subjective bias, and accelerating the processing flow. Third, through a closed-loop mechanism encompassing data calibration, human-machine collaborative optimization, and iterative verification, it ensures a high degree of consistency between the 3D model and actual geological data, effectively overcoming the poor generalization of traditional empirical models. Finally, it reduces the risks and costs of field operations based on UAV aerial photography and remote sensing technologies, and expands application scenarios by combining intelligent modeling tools, significantly improving the detection capabilities and model universality of complex karst cave systems, providing highly reliable and cost-effective technical solutions for fields such as geological engineering and ecological protection.

[0272] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for fast modeling of a cave based on multi-source data, characterized in that, Comprise the following steps: Step S101: Collecting multi-source data including geological exploration data, remote sensing data and unmanned aerial vehicle aerial photography data, preprocessing the multi-source data; Step S102: Fusing the preprocessed data by using weighted fusion or principal component analysis method to form a comprehensive karst cave geological information data set: Step S103: Using machine learning algorithm or deep learning model to extract shape feature, size feature and position feature from the fused data; Step S104: Converting the extracted karst cave features into point cloud data or grid data, and using three-dimensional modeling software to construct a karst cave model; Through human-computer interaction tool, vertex editing and face adjustment are carried out on the initial model, and the model form is corrected combined with expert experience; Step S105: Comparing the constructed karst cave model with the actual geological data, calculating the spatial error, shape error and volume error; If the error exceeds the preset threshold, the model parameters are iteratively adjusted until the accuracy requirement is met; Step S104 specifically comprises: Step S501: Through integration, preprocessing and feature extraction of multi-source data, a high-precision karst cave model is constructed by using three-dimensional modeling technology, wherein the extracted karst cave feature data set includes shape feature, size feature and position feature; The shape feature includes the long axis, short axis and direction angle of the karst cave; The size feature includes the diameter or volume of the karst cave; The position feature includes the latitude and longitude coordinates or relative position of the karst cave; The feature data is converted into point cloud data or grid data supported by three-dimensional modeling software; Wherein the point cloud data is generated according to the position and shape characteristics of the cave, assuming that the center coordinates of the cave are (x c , y c , z c ), the long axis is a, the short axis is b, and the direction angle is θ, and the calculation formula of the point cloud coordinates (x, y, z) is: ； ； ； wherein is the parameter angle, c is the height of the cave; Convert the point cloud data into grid data, use the triangulation algorithm to generate triangular patches, and the area formula of the triangular patch is: ； Wherein, v1, v2, v3 are the vertex coordinates of the triangular patch, Import the generated grid data into the three-dimensional modeling software, render the three-dimensional model in the software, and set the material and lighting effects; Step S502: Through the visual interface and interactive tool, combined with expert experience and user input, the model is dynamically adjusted and optimized, and the specific steps are as follows: (1) Interactive interface design The interface provides basic operations such as translation, rotation and scaling, as well as advanced functions such as vertex editing and face adjustment, and interacts with the model through mouse, keyboard or touch device; (2) Model adjustment Vertex editing: select a vertex in the model, adjust its position to optimize the model shape, assuming the original coordinates of vertex vi are (x i , y i , z i ), and the adjusted coordinates are (x i ′, y i ′, z i ), then the vertex displacement vector is: ； Face adjustment: Select the triangular patch in the model, adjust its normal direction or area, assume that the normal vector of the triangular patch is n, and the adjusted normal vector is n', then the normal rotation matrix is: ； Wherein, θ is the rotation angle; (3) Real-time preview Real-time preview of model changes during adjustment to ensure that the adjustment effect meets the expectations, and use level-of-detail technology to dynamically adjust model details according to viewing distance; Step S503: Through iterative adjustment of model parameters, comparison of actual geological data and cross-validation method, the model accuracy is continuously optimized, and the high precision and reliability of the model are ensured through error analysis and reliability evaluation, the model parameters include shape, size, volume, topological structure, physical property, vertex and number of deformation, position; The specific steps are as follows: (1) Model optimization The model surface is smoothed using a Laplacian smoothing algorithm, assuming that the neighbor vertex set of vertex v i is N(i), then the smoothed vertex coordinates are: ； Using edge collapse algorithm to simplify the model, reduce the number of triangular facets, assuming edge e ij Connect vertex v i And v j , the coordinates of the new vertex after folding are: ； (2) Model verification Check whether the geometric properties of the model are consistent with the original data, assuming that the volume of the model is V and the surface area is S, the calculation formula is: ； ； Where T is a triangular facet, the optimized model is compared with the original data for visualization, and the accuracy and authenticity of the model are checked; (3) Result output Export the optimized model into a common format, including OBJ, STL, FBX.

2. The method of claim 1, wherein, The geological exploration data includes drilling data, seismic data and geological profile, the remote sensing data includes satellite image and radar data, and the unmanned aerial vehicle aerial photography data includes LiDAR point cloud and high-resolution image.

3. The method of claim 2, wherein, The preprocessing of the multi-source data includes: Data cleaning: remove noise by filtering method, interpolate missing data, and correct error data; Format unification: including format conversion, coordinate system unification, format conversion is to convert geological exploration data, remote sensing data and unmanned aerial vehicle aerial photography data into geographic spatial raster vector format; Data calibration: based on ground control point, spatial alignment of multi-source data, eliminate spatial deviation caused by sensor or platform; through spectrum analysis, align the time sequence of multi-source data to ensure the consistency of data on time axis.

4. The method of claim 1, wherein the weighted fusion includes dynamically assigning weights according to the accuracy and resolution of the data sources to generate a comprehensive data set; The principal component analysis method is used for fusion, including extracting principal components by covariance matrix decomposition, eliminating redundant information and constructing dimensionality reduction data set. It includes:

5. A multi-source data based rapid modeling system for a cave, characterized in that, A multi-source data acquisition and preprocessing module is used to collect multi-source data including geological exploration data, remote sensing data and unmanned aerial vehicle aerial photography data, and to preprocess the multi-source data; A multi-source data fusion module is used to fuse the preprocessed data using weighted fusion or principal component analysis method to form a comprehensive karst cave geological information data set: A karst cave feature extraction module is used to extract shape features, size features and location features from the fused data using machine learning algorithms or deep learning models; A three-dimensional modeling and interactive optimization module is used to convert the extracted karst cave features into point cloud data or mesh data, and to construct a karst cave model using three-dimensional modeling software; the initial model is edited and adjusted through human-computer interaction tools, and the model shape is corrected combined with expert experience; A model verification and iterative optimization module is used to compare the constructed karst cave model with the actual geological data, calculate the spatial error, shape error and volume error, and if the error exceeds the preset threshold, iteratively adjust the model parameters until the accuracy requirement is met; The three-dimensional modeling and interactive optimization module is specifically used for: Through integration, preprocessing and feature extraction of multi-source data, a high-precision karst cave model is constructed using three-dimensional modeling technology, wherein the extracted karst cave feature data set includes shape features, size features and location features; Shape features include the major axis, minor axis and direction angle of the karst cave; Size features include the diameter or volume of the karst cave; The location feature includes the latitude and longitude coordinates or relative position of the karst cave; the feature data is converted into point cloud data or mesh data supported by three-dimensional modeling software; ​ Point cloud data is generated based on the location and shape characteristics of the cave, assuming the center coordinates of the cave are (x...). c , y c , z c Given a point cloud with major axis a, minor axis b, and direction angle θ, the formula for calculating the point cloud coordinates (x, y, z) is: ； ； ； wherein is the parameter angle, c is the height of the cave; Convert the point cloud data into mesh data, use the triangulation algorithm to generate triangular facets, and the area formula of the triangular facets is: ； where v1, v2, v3 are the vertex coordinates of the triangular facet, Import the generated mesh data into a three-dimensional modeling software, render the three-dimensional model in the software, and set the material and lighting effects; Through the visualization interface and interactive tools, combined with expert experience and user input, dynamically adjust and optimize the model, the specific steps are as follows: (1) Interactive interface design The interface provides basic operations such as translation, rotation, and scaling, as well as advanced functions such as vertex editing and facet adjustment, and interacts with the model through the mouse, keyboard, or touch device; (2) Model adjustment Vertex editing: select a vertex in the model, adjust its position to optimize the model shape, assuming the original coordinates of vertex vi are (x i , y i , z i ), and the adjusted coordinates are (x i ′, y i ′, z i ), then the vertex displacement vector is: ； Facet adjustment: Select the triangular facet in the model and adjust its normal direction or area. Assuming that the normal vector of the triangular facet is n and the adjusted normal vector is n', the normal rotation matrix is: ； where θ is the rotation angle; (3) Real-time preview Real-time preview of model changes during adjustment to ensure that the adjustment effect meets expectations. Use level-of-detail technology to dynamically adjust model details based on viewing distance; Iteratively adjust model parameters, compare actual geological data, and use cross-validation methods to continuously optimize model accuracy. The model parameters include shape, size, volume, topological structure, physical properties, vertex and number of deformations, and position. The specific steps are as follows: (1) Model optimization The model surface is smoothed using a Laplacian smoothing algorithm, assuming that the neighbor vertex set of vertex v i is N(i), then the smoothed vertex coordinates are: ； Using edge collapse algorithm to simplify the model, reduce the number of triangular facets, assuming edge e ij Connect vertex v i And v j , the coordinates of the new vertex after folding are: ； (2) Model verification Check if the geometric properties of the model are consistent with the original data. Assuming that the volume of the model is V and the surface area is S, the calculation formula is: ； ； where T is the triangular facet. Compare the optimized model with the original data visually to check the accuracy and authenticity of the model; (3) Result output Export the optimized model in common formats, including OBJ, STL, and FBX.

6. The system of claim 5, wherein, The geological exploration data includes drilling data, seismic data, and geological profiles. The remote sensing data includes satellite images and radar data. The unmanned aerial vehicle aerial data includes LiDAR point cloud and high-resolution images.

7. The system of claim 6, wherein, The multi-source data acquisition and preprocessing module preprocesses the multi-source data, specifically including: Data cleaning: Remove noise through filtering methods, interpolate missing data, and correct error data; Format unification: including format conversion and coordinate system unification; Data calibration: Align multi-source data in space based on ground control points to eliminate spatial deviations caused by sensors or platforms. Align the time series of multi-source data through spectral analysis to ensure consistency on the time axis; Format conversion: Convert geological exploration data, remote sensing data, and LiDARd point cloud data into geographic spatial raster and vector formats.

8. The system of claim 5, wherein, The multi-source data fusion module uses weighted fusion on the preprocessed data, including dynamically assigning weights based on the accuracy and resolution of the data sources to generate a comprehensive data set. The multi-source data fusion module uses principal component analysis to fuse the preprocessed data, including extracting principal components through covariance matrix decomposition to eliminate redundant information and construct a dimensionality reduction data set.

Images