A method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data

Abstract

This invention belongs to the field of geological exploration technology and provides a method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data. The method includes: data acquisition, construction of a high-resolution multispectral aboveground 3D model, construction of an original underground model, optimization of the original underground model, construction of a regional full-space model, regional segmentation and data fusion, and calculation of the soil and rock mass data analysis model. This invention effectively utilizes borehole data and basic geological data by fusing multiple acquired data sources, thereby improving the estimation accuracy of soil and rock mass data. By performing spatial filling processing on the original underground model, the data density of the underground model is increased, providing more refined data for model estimation and avoiding noise interference in the results due to low data volume. Through the fusion of upper and lower models and regional segmentation masks, the regional feature analysis effect is improved, the boundary extraction speed and accuracy are increased, and the estimation accuracy of conventional soil and rock mass data is enhanced.

Description

Technical Field

[0001] This invention relates to the field of geological exploration technology, and in particular to a method for analyzing rock and soil mass data based on multi-source satellite data and geoscientific data. Background Technology

[0002] Currently, soil and rock mass data in engineering geology mainly relies on field investigations and tests, resulting in a scarcity of large-area engineering geological data. Methods for rapidly acquiring large-area engineering geological soil and rock mass data are urgently needed for applications such as intelligent site selection, off-road navigation, and emergency rescue command. Researchers, in their published study on the application of data fusion technology in railway geological remote sensing interpretation, mentioned a method using multi-source data fusion for geological remote sensing interpretation. This method is primarily applied to the interpretation of adverse geological bodies in engineering geology.

[0003] However, existing technologies mainly rely on field surveys, and methods for rapidly acquiring large-area data on soil and rock masses primarily depend on remote sensing interpretation, without effectively utilizing geological data, borehole data, etc. Moreover, the main targets are adverse geological bodies, and there are no methods for refining basic soil and rock mass data. Existing methods lack a process for integrating above-ground and underground models, often analyzing the above-ground model alone, resulting in unreliable analysis results. In addition, existing methods rely solely on researchers' subjective analysis after constructing regional models to obtain conclusions, leading to low reliability and inefficiency. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data, thereby solving the problems of existing technologies such as lack of borehole data and basic geological data utilization, low data density, lack of physical model pre-analysis, and high model computational load.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data includes:

[0007] High-resolution panchromatic satellite remote sensing image data, multispectral satellite remote sensing image data, DEM data, borehole data, and basic geological data of the target analysis area are collected; the basic geological data includes: geological plan map, geological profile map, stratigraphic and lithological information, and geological structural information.

[0008] By fusing the high-resolution panchromatic satellite remote sensing image data, the multispectral satellite remote sensing image data, and the DEM data, a high-resolution multispectral ground three-dimensional model is obtained.

[0009] By integrating the borehole data and the basic geological data, an original subsurface model is obtained;

[0010] The original underground model is subjected to space filling processing to obtain a three-dimensional model of the underground geological body;

[0011] The high-resolution multispectral above-ground 3D model and the underground geological body 3D model are registered and fused to obtain a regional full-space model;

[0012] The full-space model of the region is segmented to obtain a three-dimensional segmentation mask. The full-space model of the region and the three-dimensional segmentation mask are then fused to obtain the data to be analyzed.

[0013] The data to be analyzed is input into a pre-trained soil and rock mass data analysis model for analysis, and the analysis results of the target soil and rock mass data are obtained.

[0014] Preferably, high-resolution panchromatic satellite remote sensing image data, multispectral satellite remote sensing image data, hyperspectral satellite remote sensing image data, DEM data, borehole data, and basic geological data of the target analysis area are collected, including:

[0015] The boundary of the target analysis area is defined by satellite imagery and topographic mapping to obtain the area range;

[0016] The area is divided into regular vertical and horizontal grids to obtain the area grid;

[0017] Using GPS, the grid intersections of the area grid within the target analysis area are marked, and the line segments formed by the grid intersections located in the same row or column are determined as a longitudinal or transverse edge.

[0018] Preferably, the acquisition of high-resolution panchromatic satellite remote sensing image data, multispectral satellite remote sensing image data, hyperspectral satellite remote sensing image data, DEM data, borehole data, and basic geological data of the target analysis area also includes:

[0019] For each of the aforementioned longitudinal and transverse sides, stratigraphic outcrop observations, structural identification, overburden investigations, hydrogeological phenomenon recordings, and geophysical data collection are conducted to obtain the geological profile map.

[0020] Drilling was performed at each of the grid intersections to obtain samples;

[0021] Core and soil samples were analyzed from the collected samples to obtain basic physical parameters;

[0022] Geological data were collected at each grid intersection using geophysical logging and in-situ testing techniques to obtain formation parameters as well as mechanical and hydrological parameters.

[0023] The borehole data is obtained by integrating the basic physical parameters, the formation parameters, and the mechanical and hydrological parameters.

[0024] Preferably, the high-resolution panchromatic satellite remote sensing image data, the multispectral satellite remote sensing image data, and the DEM data are fused to obtain a high-resolution multispectral ground three-dimensional model, including:

[0025] Radiometric correction, geometric fine correction, and image registration are performed on the high-resolution panchromatic satellite remote sensing image data and the multispectral satellite remote sensing image data, respectively.

[0026] The correlation coefficient is calculated between each multispectral band of the multispectral satellite remote sensing image data and the high-resolution panchromatic satellite remote sensing image data; the formula for calculating the correlation coefficient is as follows: ;in, The i-th correlation coefficient; This represents the function for calculating the correlation coefficient. The high-resolution panchromatic satellite remote sensing image data; For the i-th multispectral band;

[0027] The multispectral bands are weighted and summed using the correlation coefficient as the weight to obtain a simulated panchromatic image.

[0028] The difference between the high-resolution panchromatic satellite remote sensing image data and the simulated panchromatic image is calculated to obtain the original difference data. The original difference data is then subjected to three times the standard deviation threshold filtering and interpolation to obtain the optimized difference data.

[0029] The high-frequency detail components are obtained by calculating the difference between the multispectral bands and their projection components on the simulated panchromatic image.

[0030] The optimized difference data is proportionally replaced with the high-frequency detail components to obtain a high-resolution multispectral image; the expression for the high-resolution multispectral image is: ;in, This refers to the i-th optimized band data of the high-resolution multispectral image; For projection formulas; The simulated panchromatic image; This is the scaling factor; The optimized difference data.

[0031] Preferably, the high-resolution panchromatic satellite remote sensing image data, the multispectral satellite remote sensing image data, and the DEM data are fused to obtain a high-resolution multispectral ground three-dimensional model, which further includes:

[0032] The DEM data was converted into a digital terrain model using GIS tools.

[0033] The high-resolution multispectral image is mapped onto the surface of the digital terrain model to obtain the high-resolution multispectral ground 3D model.

[0034] Preferably, the borehole data and the basic geological data are fused to obtain the original subsurface model, including:

[0035] A grid model is constructed based on the depth data of the regional grid and the geological profile, and the resolution of the grid model is set.

[0036] A spatial coordinate system is obtained by taking the top left corner of the mesh model as the origin, the two sides where the origin is located as the X and Y axes, and the vertical depth as the Z axis;

[0037] The borehole data and the basic geological data are injected into the grid model according to the coordinate scaling ratio based on the spatial coordinate system to obtain the original underground model.

[0038] Preferably, the original underground model is subjected to space-filling processing to obtain a three-dimensional model of the underground geological body, including:

[0039] The original underground model is divided into several cuboids; the attribute data of the vertical edges of the cuboids correspond to the borehole data; the attribute data of the vertical lateral surfaces of the cuboids correspond to the basic geological data.

[0040] Map the cuboid into a unit cube;

[0041] The drilling data is fitted using the unit cube to obtain the edge segment function;

[0042] The boundary surface function is obtained by fitting the basic geological data to the unit cube;

[0043] The governing equations are constructed based on the unit cube and the boundary surface function; the expression of the governing equations is: ;in, ; ; ; ; For attribute functions; , , These are the horizontal axis, vertical axis, and vertical axis of the unit cube, respectively; , , , The boundary surface functions corresponding to the front, back, left, and right sides of the unit cube, respectively;

[0044] Starting with the edge segment function, the governing equations are discretized and iteratively solved using the finite difference method to obtain the original solution results.

[0045] Tensor product spline fitting and inverse mapping are performed on the original solution to obtain the optimized solution;

[0046] When the difference in the local first derivative of the optimized solution on both sides of the cuboid in the vertical direction is greater than the preset abnormal difference threshold, a second solution is performed on the two cuboids corresponding to the difference in the local first derivative.

[0047] Preferably, the high-resolution multispectral above-ground 3D model and the underground geological body 3D model are registered and fused to obtain a regional full-space model, including:

[0048] The spatial coordinate system of the underground geological body 3D model is converted into the global coordinate system of the high-resolution multispectral aboveground 3D model using a coordinate transformation algorithm;

[0049] Extract the location data of the grid intersections of the three-dimensional model of the underground geological body;

[0050] By matching the location data of the grid intersection points in the underground geological 3D model with the location data in the high-resolution multispectral above-ground 3D model, a mapping pair is obtained;

[0051] A rigid transformation model is constructed based on the mapping pair, and the rigid transformation model is used to match the three-dimensional model of the underground geological body to the high-resolution multispectral above-ground three-dimensional model to obtain the full-space model of the region; the expression of the rigid transformation model is: ;in, This refers to the original location data of the three-dimensional model of the underground geological body; The mapped location data; It is a rotation matrix; It is a translation vector.

[0052] Preferably, the full-space model of the region is segmented to obtain a three-dimensional segmentation mask. The full-space model of the region and the three-dimensional segmentation mask are then fused to obtain the data to be analyzed, including:

[0053] Each node in the full-space model of the region where the borehole is located is determined as an initial seed point; the initial seed point includes soil and rock property data.

[0054] The attribute difference data between the initial seed point and its neighboring points is calculated using the Euclidean distance formula to obtain the attribute distance value;

[0055] Add neighboring points whose attribute distance value is less than the out-of-type point threshold to the same region point set of the initial seed point;

[0056] The newly added neighboring point in the same region point set is determined as the new initial seed point, and the process returns to the step "using the Euclidean distance formula to calculate the attribute difference data between the initial seed point and the neighboring points to obtain the attribute distance value" for iterative growth, so as to obtain the same region point set after propagation is completed.

[0057] Remove the set of points in the same region whose number of points is less than the minimum value of the region size;

[0058] The points corresponding to the removed points in the same region are labeled and numbered. The points in the full space model of the region that are removed from the same region are assigned a value of 0 to obtain the three-dimensional segmentation mask.

[0059] Preferably, the training process of the soil and rock mass data analysis model includes:

[0060] A three-dimensional convolutional neural network is selected, and several classification output heads are added to the three-dimensional convolutional neural network;

[0061] Set target outputs and assign similar target outputs to the same classification output head; the target outputs include: rock mass and soil boundary line, rock mass and soil type, rock mass fracture degree, rock mass weathering degree, rock mass overlying soil thickness, rock mass overlying soil type, soil deposition type, and soil underlying bedrock type.

[0062] Sub-loss functions are constructed based on Dice loss, cross-entropy loss, and mean squared error loss, and all sub-loss functions are weighted and fused to obtain the total loss function;

[0063] Several sets of data to be analyzed and the actual measurement results corresponding to the data to be analyzed are to be acquired in advance;

[0064] The three-dimensional convolutional neural network is iteratively trained using the total loss function based on the pre-acquired data to be analyzed.

[0065] When the proportion of the classification output heads with loss values ​​less than a preset loss threshold exceeds a preset freeze ratio, the network layer parameters of the three-dimensional convolutional neural network are frozen, and the classification output heads with loss values ​​greater than the preset loss threshold are iteratively trained using the sub-loss function to obtain the trained soil and rock mass data analysis model.

[0066] The present invention discloses the following technical effects:

[0067] This invention provides a method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data. By fusing multiple acquired data sources, it overcomes the shortcomings of existing technologies that lack borehole data and basic geological data, enabling the construction and analysis of three-dimensional models of underground geological bodies. By performing spatial filling processing on the original underground model, it addresses the deficiency of low data density in conventionally constructed physical models, achieving single-body data filling and summative analysis of multi-body boundary data. Through the fusion of upper and lower models and regional segmentation masks, it solves the problems of lack of physical model pre-analysis and high model computational load in conventional technologies, enabling the fusion analysis of upper and lower models and the pre-extraction of regional information. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0069] Figure 1 A schematic diagram of the rock and soil mass data analysis process based on multi-source satellite data and geoscience data provided in an embodiment of the present invention;

[0070] Figure 2 This is a schematic diagram of the process for acquiring borehole data and basic geological data provided in an embodiment of the present invention;

[0071] Figure 3 This is a schematic diagram illustrating the process of constructing a high-resolution multispectral terrestrial 3D model according to an embodiment of the present invention.

[0072] Figure 4 This is a schematic diagram of the original underground model construction process provided in an embodiment of the present invention;

[0073] Figure 5 This is a schematic diagram illustrating the process of constructing a three-dimensional model of an underground geological body, as provided in an embodiment of the present invention. Detailed Implementation

[0074] 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, and 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.

[0075] The purpose of this invention is to provide a method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data, which solves the problems of existing technologies such as lack of borehole data and basic geological data utilization, low data density, lack of physical model pre-analysis, and high model calculation volume.

[0076] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0077] Figure 1 This is a schematic diagram of the rock and soil mass data analysis process based on multi-source satellite data and geoscience data provided in an embodiment of the present invention, as shown below. Figure 1 As shown, this invention provides a method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data, including:

[0078] Step 100: Collect high-resolution panchromatic satellite remote sensing image data, multispectral satellite remote sensing image data, DEM data, borehole data, and basic geological data for the target analysis area; the basic geological data includes: geological plan map, geological profile map, stratigraphic and lithological information, and geological structural information;

[0079] Step 200: Fuse the high-resolution panchromatic satellite remote sensing image data, the multispectral satellite remote sensing image data, and the DEM data to obtain a high-resolution multispectral ground three-dimensional model;

[0080] Step 300: Merge the borehole data and the basic geological data to obtain the original subsurface model;

[0081] Step 400: Perform space filling processing on the original underground model to obtain a three-dimensional model of the underground geological body;

[0082] Step 500: Register and fuse the high-resolution multispectral above-ground 3D model and the underground geological body 3D model to obtain a regional full-space model;

[0083] Step 600: Perform region segmentation on the full-space model of the region to obtain a three-dimensional segmentation mask, and fuse the full-space model of the region and the three-dimensional segmentation mask to obtain the data to be analyzed;

[0084] Step 100: Input the data to be analyzed into the pre-trained soil and rock mass data analysis model for analysis to obtain the target soil and rock mass data analysis results.

[0085] refer to Figure 2 High-resolution panchromatic satellite remote sensing imagery, multispectral satellite remote sensing imagery, hyperspectral satellite remote sensing imagery, DEM data, borehole data, and basic geological data of the target analysis area were collected, including:

[0086] Step 101: Define the boundary of the target analysis area using satellite imagery and topographic mapping to obtain the area range;

[0087] Step 102: Divide the area into regular vertical and horizontal grids to obtain the area grid;

[0088] Step 103: Using GPS, mark the grid intersections of the area grid within the target analysis area, and determine the line segment formed by the grid intersections located in the same row or column as a vertical or horizontal edge.

[0089] refer to Figure 2 The data collected includes high-resolution panchromatic satellite remote sensing imagery, multispectral satellite remote sensing imagery, hyperspectral satellite remote sensing imagery, DEM data, borehole data, and basic geological data for the target analysis area.

[0090] Step 104: Conduct stratigraphic outcrop observation, structural identification, overburden investigation, hydrogeological phenomenon recording, and geophysical data acquisition for the area where each of the aforementioned longitudinal and transverse sides is located, and obtain the geological profile map;

[0091] Step 105: Drill at each intersection of the grid to obtain samples;

[0092] Step 106: Analyze the collected samples, including rock cores and soil samples, to obtain basic physical parameters;

[0093] Step 107: Geological data are collected at each grid intersection using geophysical logging and in-situ testing techniques to obtain formation parameters as well as mechanical and hydrological parameters;

[0094] Step 108: Integrate the basic physical parameters, the formation parameters, and the mechanical and hydrological parameters to obtain the borehole data.

[0095] refer to Figure 3 By fusing the high-resolution panchromatic satellite remote sensing data, the multispectral satellite remote sensing data, and the DEM data, a high-resolution multispectral ground three-dimensional model is obtained, including:

[0096] Step 201: Perform radiometric correction, geometric fine correction, and image registration on the high-resolution panchromatic satellite remote sensing image data and the multispectral satellite remote sensing image data, respectively;

[0097] Step 202: Calculate the correlation coefficient between each multispectral band of the multispectral satellite remote sensing image data and the high-resolution panchromatic satellite remote sensing image data; the formula for calculating the correlation coefficient is: ;in, The i-th correlation coefficient; This represents the function for calculating the correlation coefficient. The high-resolution panchromatic satellite remote sensing image data; For the i-th multispectral band;

[0098] Step 203: Weight the multispectral bands using the correlation coefficient as the weight to obtain a simulated panchromatic image;

[0099] Step 204: Calculate the difference between the high-resolution panchromatic satellite remote sensing image data and the simulated panchromatic image to obtain the original difference data. Perform three-times standard deviation threshold filtering and interpolation on the original difference data to obtain optimized difference data.

[0100] Step 205: Calculate the difference between the multispectral band and its projection component on the simulated panchromatic image to obtain the high-frequency detail component;

[0101] Step 206: Replace the high-frequency detail components with the optimized difference data proportionally to obtain a high-resolution multispectral image; the expression for the high-resolution multispectral image is: ;in, This refers to the i-th optimized band data of the high-resolution multispectral image; For projection formulas; The simulated panchromatic image; This is the scaling factor; The optimized difference data.

[0102] refer to Figure 3 By fusing the high-resolution panchromatic satellite remote sensing data, the multispectral satellite remote sensing data, and the DEM data, a high-resolution multispectral ground three-dimensional model is obtained, which also includes:

[0103] Step 207: Use GIS tools to convert the DEM data into a digital terrain model;

[0104] Step 208: Map the high-resolution multispectral image onto the surface of the digital terrain model to obtain the high-resolution multispectral ground 3D model.

[0105] refer to Figure 4 By integrating the borehole data and the basic geological data, an original subsurface model is obtained, including:

[0106] Step 301: Construct a grid model based on the regional grid and the depth data of the geological profile, and set the resolution of the grid model;

[0107] Step 302: Using the top left corner of the mesh model as the origin, the two sides containing the origin as the X and Y axes, and the vertical depth as the Z axis, a spatial coordinate system is obtained;

[0108] Step 303: According to the spatial coordinate system, inject the borehole data and the basic geological data into the grid model according to the coordinate scaling ratio to obtain the original underground model.

[0109] refer to Figure 5 The original underground model is subjected to space filling processing to obtain a three-dimensional model of the underground geological body, including:

[0110] Step 401: Divide the original underground model into several cuboids; the attribute data of the vertical edges of the cuboids correspond to the borehole data; the attribute data of the vertical sides of the cuboids correspond to the basic geological data.

[0111] Step 402: Map the cuboid into a unit cube;

[0112] Step 403: Fit the borehole data to the unit cube to obtain the edge segment function;

[0113] Step 404: Fit the basic geological data to the unit cube to obtain the boundary surface function;

[0114] Step 405: Construct the governing equations based on the unit cube and the boundary surface function; the expression of the governing equations is: ;in, ; ; ; ; For attribute functions; , , These are the horizontal axis, vertical axis, and vertical axis of the unit cube, respectively; , , , The boundary surface functions corresponding to the front, back, left, and right sides of the unit cube, respectively;

[0115] Step 406: Starting with the edge segment function, discretize and iteratively solve the governing equation using the finite difference method to obtain the original solution result;

[0116] Step 407: Perform tensor product spline fitting and inverse mapping on the original solution results to obtain optimized solution results;

[0117] Step 408: When the difference between the local first derivatives of the optimized solution on both sides of the cuboid in the vertical direction is greater than the preset abnormal difference threshold, perform a second solution on the two cuboids corresponding to the difference between the local first derivatives.

[0118] Specifically, the high-resolution multispectral above-ground 3D model and the underground geological body 3D model are registered and fused to obtain a regional full-space model, including:

[0119] The spatial coordinate system of the underground geological body 3D model is converted into the global coordinate system of the high-resolution multispectral aboveground 3D model using a coordinate transformation algorithm;

[0120] Extract the location data of the grid intersections of the three-dimensional model of the underground geological body;

[0121] By matching the location data of the grid intersection points in the underground geological 3D model with the location data in the high-resolution multispectral above-ground 3D model, a mapping pair is obtained;

[0122] A rigid transformation model is constructed based on the mapping pair, and the rigid transformation model is used to match the three-dimensional model of the underground geological body to the high-resolution multispectral above-ground three-dimensional model to obtain the full-space model of the region; the expression of the rigid transformation model is: ;in, This refers to the original location data of the three-dimensional model of the underground geological body; The mapped location data; It is a rotation matrix; It is a translation vector.

[0123] Further, the full-space model of the region is segmented to obtain a three-dimensional segmentation mask. The full-space model of the region and the three-dimensional segmentation mask are then fused to obtain the data to be analyzed, including:

[0124] Each node in the full-space model of the region where the borehole is located is determined as an initial seed point; the initial seed point includes soil and rock property data.

[0125] The attribute difference data between the initial seed point and its neighboring points is calculated using the Euclidean distance formula to obtain the attribute distance value;

[0126] Add neighboring points whose attribute distance value is less than the out-of-type point threshold to the same region point set of the initial seed point;

[0127] The newly added neighboring point in the same region point set is determined as the new initial seed point, and the process returns to the step "using the Euclidean distance formula to calculate the attribute difference data between the initial seed point and the neighboring points to obtain the attribute distance value" for iterative growth, so as to obtain the same region point set after propagation is completed.

[0128] Remove the set of points in the same region whose number of points is less than the minimum value of the region size;

[0129] The points corresponding to the removed points in the same region are labeled and numbered. The points in the full space model of the region that are removed from the same region are assigned a value of 0 to obtain the three-dimensional segmentation mask.

[0130] Preferably, the training process of the soil and rock mass data analysis model includes:

[0131] A three-dimensional convolutional neural network is selected, and several classification output heads are added to the three-dimensional convolutional neural network;

[0132] Set target outputs and assign similar target outputs to the same classification output head; the target outputs include: rock mass and soil boundary line, rock mass and soil type, rock mass fracture degree, rock mass weathering degree, rock mass overlying soil thickness, rock mass overlying soil type, soil deposition type, and soil underlying bedrock type.

[0133] Sub-loss functions are constructed based on Dice loss, cross-entropy loss, and mean squared error loss, and all sub-loss functions are weighted and fused to obtain the total loss function;

[0134] Several sets of data to be analyzed and the actual measurement results corresponding to the data to be analyzed are to be acquired in advance;

[0135] The three-dimensional convolutional neural network is iteratively trained using the total loss function based on the pre-acquired data to be analyzed.

[0136] When the proportion of the classification output heads with loss values ​​less than a preset loss threshold exceeds a preset freeze ratio, the network layer parameters of the three-dimensional convolutional neural network are frozen, and the classification output heads with loss values ​​greater than the preset loss threshold are iteratively trained using the sub-loss function to obtain the trained soil and rock mass data analysis model.

[0137] Specifically, the Gaofen-2 satellite, a civilian optical remote sensing satellite with a spatial resolution better than 1 meter, is equipped with two high-resolution 1-meter panchromatic and 4-meter multispectral cameras. These cameras can provide high-resolution panchromatic satellite remote sensing image data with a resolution of 1 meter and multispectral satellite remote sensing image data with a resolution of 4 meters for this embodiment. Gaofen-7 can generate stereo images and can provide high-precision DEM data for this embodiment.

[0138] Preferably, the geographical boundaries of the target analysis area are determined by satellite imagery and topographic mapping technology, and the area is clearly defined. The defined area is divided into regular vertical and horizontal grids to form regional grids, which facilitates the standardization and systematization of subsequent data collection. Using GPS positioning technology, the intersection points of each regional grid are marked within the target analysis area. The intersection points of grids located in the same row or column are connected to form line segments, which are defined as vertical and horizontal edges, providing a positioning reference for subsequent data collection.

[0139] Furthermore, multi-dimensional field observations and data collection were conducted in the areas where each longitudinal and transverse edge was located to obtain basic geological data; geological structures within the area were identified; the type, thickness, distribution range, and engineering properties of the surface overburden were investigated; the location, quantity, and quality of groundwater outcrops, as well as the recharge relationship between surface water bodies and groundwater, were recorded; and geophysical exploration techniques were used to obtain information on underground geological structures and lithological interfaces to assist in the construction of geological profile maps.

[0140] Furthermore, drilling is conducted at the intersections of each grid to extract core samples and soil samples. These samples are then analyzed in the laboratory to obtain basic physical parameters. Additionally, resistivity logging and sonic logging are used to measure the formation resistivity and sonic velocity. Finally, the basic physical parameters, formation parameters, mechanical parameters, and hydrological parameters from these intersections are integrated to form borehole data.

[0141] Specifically, a simulated panchromatic image is constructed, and then a weighted sum is performed on the simulated panchromatic image to generate a simulated panchromatic image that has the highest correlation with the panchromatic image. It is usually determined based on the correlation coefficient between the multispectral bands and the panchromatic image, using the following formula:

[0142]

[0143] Calculate panchromatic image Compared with simulated panchromatic images The difference:

[0144]

[0145] This difference contains high-frequency details in the panchromatic image that were not captured by the multispectral bands.

[0146] High-frequency detail injection into multispectral bands, for each multispectral band Perform the following procedures:

[0147] Calculate the band in Projected components in the direction;

[0148] The high-frequency detail components of this band are separated by orthogonalization calculations.

[0149] Difference of panchromatic images Inject high-frequency components from each band proportionally to generate new bands:

[0150]

[0151] Processed multispectral bands By combining these images, a high-resolution multispectral image is obtained, whose spatial resolution is consistent with that of the panchromatic image.

[0152] Furthermore, the preprocessed high-resolution multispectral imagery is spatially registered with the DEM data. Using the 3D modeling capabilities of GIS or remote sensing software, the model is generated through the following steps:

[0153] Generate digital terrain models based on DEM;

[0154] High-resolution multispectral imagery is used as texture data and mapped onto the DTM surface to give the terrain realistic color and ground feature details;

[0155] By specifying the geographic coordinate system and the projected coordinate system, a high-resolution multispectral 3D model of the ground is obtained.

[0156] Optionally, a grid model is constructed based on the depth data of the regional grid and geological profile, and the resolution of the grid model is set; a spatial coordinate system is obtained with the upper left corner of the grid model as the origin, the two sides where the origin is located as the X-axis and Y-axis, and the vertical depth as the Z-axis; based on the spatial coordinate system, the borehole data and basic geological data are injected into the grid model according to the coordinate scaling ratio to obtain the original underground model.

[0157] Furthermore, the division of three-dimensional space:

[0158] Line segments extending along the Z-axis are located at uniform grid nodes in the XY plane. , recorded as ;

[0159] Uniformly distributed along the X direction, the plane is numbered as follows ,correspond ;

[0160] Uniformly distributed along the Y direction, the plane is numbered as follows ,correspond ;

[0161] Definition of a cuboid: a rectangular prism is formed by adjacent planes. , and height range Composition, denoted as Its four vertical edges are: , , , ;

[0162] The known attribute data is ;

[0163] Boundary condition constraints:

[0164] cuboid All four sides (front, back, left, and right) belong to the vertical plane, and their attribute values ​​are directly provided by the plane nodes;

[0165] The attribute values ​​of the four vertical edges are provided by the line segment nodes.

[0166] Specifically, the cuboid Mapping to a unit cube, defining local coordinates Side attribute function: front side is The rear side is The left side is The right side is Construct the governing equations:

[0167]

[0168] Boundary conditions: ; ; ; .

[0169] The above governing equations are discretized using the finite difference method, solved, and the solution is smoothed using the tensor product spline interpolation method.

[0170] When the difference in the local first derivative of the optimized solution on both sides of the cuboid in the vertical direction is greater than the preset abnormal difference threshold, a second solution is performed on the two cuboids corresponding to the local first derivative difference to finally obtain the three-dimensional model of the underground geological body.

[0171] Preferably, a coordinate transformation algorithm is used to convert the local coordinate system of the underground model into a global coordinate system consistent with that of the above-ground model. The borehole locations have a clear spatial correspondence in both the above-ground and underground models. Surface coordinates are extracted from the borehole data as common control points for both the above-ground and underground models. A rigid transformation model is used to adjust the underground model so that the projections of the borehole control points in the above-ground model are aligned with the coordinates of the underground model, as shown in the following formula:

[0172]

[0173] The final result is a full-space model of the region.

[0174] Further, firstly, each borehole location in the region model is identified as an initial seed point, and its soil and rock property data is extracted. Next, the attribute differences between these initial seed points and their neighboring points are calculated using the Euclidean distance formula, yielding attribute distance values. If the attribute distance value of a neighboring point is lower than a set threshold, it is added to the same region point set of the initial seed points. Then, the most recently added neighboring point is set as a new initial seed point, and the process returns to the previous steps, continuing the calculation and iteration until the expansion of this same region point set is complete. Next, points with fewer points than the minimum volume of the region are removed. Finally, the remaining same region point set is numbered, and points not in this set are assigned a value of 0, thus generating a 3D segmentation mask.

[0175] Preferably, this embodiment selects a three-dimensional convolutional neural network (hereinafter referred to as 3D-CNN) as the basic training architecture for the geotechnical data analysis model. The input data consists of a three-dimensional region full-space model and a three-dimensional segmentation mask, and the target output includes classification and regression tasks. Several sub-loss functions are constructed based on Dice loss, cross-entropy loss, and mean squared error loss, and then these sub-loss functions are weighted together to form a total loss function; training samples are pre-collected and labeled with actual target attribute data or image mask data; a 3D-CNN multi-task network is constructed and the weights are initialized; training data is input into the 3D-CNN multi-task network in batches to alternately optimize the loss of each task. When the loss value of the classification output head is lower than a preset threshold and exceeds a certain proportion, the parameters of the network layer are frozen, and then the classification output heads with loss values ​​higher than the threshold are trained again using the sub-loss functions, finally completing the training of the geotechnical data analysis model.

[0176] The beneficial effects of this invention are as follows:

[0177] This invention improves the estimation accuracy of soil and rock mass data by integrating multiple types of acquired data, effectively utilizing borehole data and basic geological data; it increases the data density of the underground model by performing space filling processing on the original underground model, providing more refined data for model estimation and avoiding noise interference to the results due to low data volume; and it improves the regional feature analysis effect, increases the speed and accuracy of boundary extraction, and improves the estimation accuracy of conventional soil and rock mass data by fusing upper and lower models and regional segmentation masks.

[0178] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0179] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data, characterized in that, include: High-resolution panchromatic satellite remote sensing image data, multispectral satellite remote sensing image data, DEM data, borehole data, and basic geological data of the target analysis area were collected. The basic geological data includes: geological plan, geological profile, stratigraphic and lithological information, and geological structural information; By fusing the high-resolution panchromatic satellite remote sensing image data, the multispectral satellite remote sensing image data, and the DEM data, a high-resolution multispectral ground three-dimensional model is obtained. By integrating the borehole data and the basic geological data, an original subsurface model is obtained; The original underground model is subjected to space filling processing to obtain a three-dimensional model of the underground geological body; The high-resolution multispectral above-ground 3D model and the underground geological body 3D model are registered and fused to obtain a regional full-space model; The full-space model of the region is segmented to obtain a three-dimensional segmentation mask. The full-space model of the region and the three-dimensional segmentation mask are then fused to obtain the data to be analyzed. The data to be analyzed is input into a pre-trained rock and soil data analysis model for analysis to obtain the target rock and soil data analysis results; The training process of the soil and rock mass data analysis model includes: A three-dimensional convolutional neural network is selected, and several classification output heads are added to the three-dimensional convolutional neural network; Set target outputs and assign similar target outputs to the same classification output head; the target outputs include: rock mass and soil boundary line, rock mass and soil type, rock mass fracture degree, rock mass weathering degree, rock mass overlying soil thickness, rock mass overlying soil type, soil deposition type, and soil underlying bedrock type. Sub-loss functions are constructed based on Dice loss, cross-entropy loss, and mean squared error loss, and all sub-loss functions are weighted and fused to obtain the total loss function; Several sets of data to be analyzed and the actual measurement results corresponding to the data to be analyzed are to be acquired in advance; The three-dimensional convolutional neural network is iteratively trained using the total loss function based on the pre-acquired data to be analyzed. When the proportion of the classification output heads with loss values ​​less than a preset loss threshold exceeds a preset freeze ratio, the network layer parameters of the three-dimensional convolutional neural network are frozen, and the classification output heads with loss values ​​greater than the preset loss threshold are iteratively trained using the sub-loss function to obtain the trained soil and rock mass data analysis model.

2. The method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data according to claim 1, characterized in that, High-resolution panchromatic satellite remote sensing imagery, multispectral satellite remote sensing imagery, hyperspectral satellite remote sensing imagery, DEM data, borehole data, and basic geological data of the target analysis area were collected, including: The boundary of the target analysis area is defined by satellite imagery and topographic mapping to obtain the area range; The area is divided into regular vertical and horizontal grids to obtain the area grid; Using GPS, the grid intersections of the area grid within the target analysis area are marked, and the line segments formed by the grid intersections located in the same row or column are determined as a longitudinal or transverse edge.

3. The method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data according to claim 2, characterized in that, The data collected includes high-resolution panchromatic satellite remote sensing imagery, multispectral satellite remote sensing imagery, hyperspectral satellite remote sensing imagery, DEM data, borehole data, and basic geological data for the target analysis area. For each of the aforementioned longitudinal and transverse sides, stratigraphic outcrop observations, structural identification, overburden investigations, hydrogeological phenomenon recordings, and geophysical data collection are conducted to obtain the geological profile map. Drilling was performed at each of the grid intersections to obtain samples; Core and soil samples were analyzed from the collected samples to obtain basic physical parameters; Geological data were collected at each grid intersection using geophysical logging and in-situ testing techniques to obtain formation parameters as well as mechanical and hydrological parameters. The borehole data is obtained by integrating the basic physical parameters, the formation parameters, and the mechanical and hydrological parameters.

4. The method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data according to claim 3, characterized in that, By fusing the high-resolution panchromatic satellite remote sensing image data, the multispectral satellite remote sensing image data, and the DEM data, a high-resolution multispectral 3D ground model is obtained, including: Radiometric correction, geometric fine correction, and image registration are performed on the high-resolution panchromatic satellite remote sensing image data and the multispectral satellite remote sensing image data, respectively. The correlation coefficient is calculated between each multispectral band of the multispectral satellite remote sensing image data and the high-resolution panchromatic satellite remote sensing image data; the formula for calculating the correlation coefficient is as follows: ;in, The i-th correlation coefficient; This represents the function for calculating the correlation coefficient. The high-resolution panchromatic satellite remote sensing image data; For the i-th multispectral band; The multispectral bands are weighted and summed using the correlation coefficient as the weight to obtain a simulated panchromatic image. The difference between the high-resolution panchromatic satellite remote sensing image data and the simulated panchromatic image is calculated to obtain the original difference data. The original difference data is then subjected to three times the standard deviation threshold filtering and interpolation to obtain the optimized difference data. The high-frequency detail component is obtained by calculating the difference between the multispectral band and the projection component of the multispectral band on the simulated panchromatic image. The optimized difference data is proportionally replaced with the high-frequency detail components to obtain a high-resolution multispectral image; the expression for the high-resolution multispectral image is: ;in, This refers to the i-th optimized band data of the high-resolution multispectral image; For projection formula; The simulated panchromatic image; This is the scaling factor; The optimized difference data.

5. The method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data according to claim 4, characterized in that, By fusing the high-resolution panchromatic satellite remote sensing data, the multispectral satellite remote sensing data, and the DEM data, a high-resolution multispectral 3D ground model is obtained, which also includes: The DEM data was converted into a digital terrain model using GIS tools. The high-resolution multispectral image is mapped onto the surface of the digital terrain model to obtain the high-resolution multispectral ground 3D model.

6. The method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data according to claim 5, characterized in that, By integrating the borehole data and the basic geological data, an original subsurface model is obtained, including: A grid model is constructed based on the depth data of the regional grid and the geological profile, and the resolution of the grid model is set. A spatial coordinate system is obtained by taking the upper left corner of the mesh model as the origin, the two sides where the origin is located as the X and Y axes, and the vertical depth as the Z axis; The borehole data and the basic geological data are injected into the grid model according to the coordinate scaling ratio based on the spatial coordinate system to obtain the original underground model.

7. The method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data according to claim 6, characterized in that, The original underground model is subjected to space filling processing to obtain a three-dimensional model of the underground geological body, including: The original underground model is divided into several cuboids; the attribute data of the vertical edges of the cuboids correspond to the borehole data; the attribute data of the vertical lateral surfaces of the cuboids correspond to the basic geological data. Map the cuboid into a unit cube; The borehole data is fitted using the unit cube to obtain the edge segment function; The boundary surface function is obtained by fitting the basic geological data to the unit cube; The governing equations are constructed based on the unit cube and the boundary surface function; the expression of the governing equations is: ;in, ; ; ; ; For attribute functions; , , These are the horizontal axis, vertical axis, and vertical axis of the unit cube, respectively; , , , The boundary surface functions corresponding to the front, back, left, and right sides of the unit cube, respectively; Starting with the edge segment function, the governing equations are discretized and iteratively solved using the finite difference method to obtain the original solution results. Tensor product spline fitting and inverse mapping are performed on the original solution to obtain the optimized solution; When the difference in the local first derivative of the optimized solution on both sides of the cuboid in the vertical direction is greater than the preset abnormal difference threshold, a second solution is performed on the two cuboids corresponding to the difference in the local first derivative.

8. The method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data according to claim 7, characterized in that, The high-resolution multispectral aboveground 3D model and the underground geological body 3D model are registered and fused to obtain a regional full-space model, including: The spatial coordinate system of the underground geological body 3D model is converted into the global coordinate system of the high-resolution multispectral aboveground 3D model using a coordinate transformation algorithm; Extract the location data of the grid intersections of the three-dimensional model of the underground geological body; By matching the location data of the grid intersection points in the underground geological 3D model with the location data in the high-resolution multispectral above-ground 3D model, a mapping pair is obtained; A rigid transformation model is constructed based on the mapping pair, and the rigid transformation model is used to match the three-dimensional model of the underground geological body to the high-resolution multispectral above-ground three-dimensional model to obtain the full-space model of the region; the expression of the rigid transformation model is: ;in, This refers to the original location data of the three-dimensional model of the underground geological body; The mapped location data; It is a rotation matrix; It is a translation vector.

9. The method for analyzing soil and rock mass data based on multi-source satellite data and geoscientific data according to claim 8, characterized in that, The full-space model of the region is segmented to obtain a 3D segmentation mask. The full-space model of the region and the 3D segmentation mask are then fused to obtain the data to be analyzed, including: Each node in the full-space model of the region where the borehole is located is determined as an initial seed point; the initial seed point includes soil and rock property data. The attribute difference data between the initial seed point and its neighboring points is calculated using the Euclidean distance formula to obtain the attribute distance value; Add neighboring points whose attribute distance value is less than the out-of-type point threshold to the same region point set of the initial seed point; The newly added neighboring point in the same region point set is determined as the new initial seed point, and the process returns to the step "using the Euclidean distance formula to calculate the attribute difference data between the initial seed point and the neighboring points to obtain the attribute distance value" for iterative growth, so as to obtain the same region point set after propagation is completed. Remove the set of points in the same region whose number of points is less than the minimum value of the region size; The points corresponding to the removed points in the same region are labeled and numbered. The points in the full space model of the region that are removed from the same region are assigned a value of 0 to obtain the three-dimensional segmentation mask.

Images