An information fusion method and system for deep geological exploration

Abstract

The application provides a deep geological exploration information fusion method and system, and relates to the technical field of data processing. The method comprises the following steps: obtaining multiple sets of depth information of a region to be explored, wherein the depth information comprises shallow data and deep data; processing the shallow data by using an improved nonlinear conversion algorithm to obtain a first converted information set; processing the deep data by using a deep learning algorithm to obtain a second converted information set; performing data fusion on the first converted information set and the second converted information set respectively to obtain a first fusion information subset and a second fusion information subset; and performing data optimization and fusion on the first fusion information subset and the second fusion information subset by using a tensor decomposition algorithm and a joint inversion algorithm to obtain final fusion data. The application solves the problems of insufficient baseline length and low information integration efficiency in the prior art when dealing with complex geological environments.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an information fusion method and system for deep geological exploration. Background Technology

[0002] With the continuous development of deep geological exploration technology, traditional geological exploration methods face many challenges. Existing technologies typically rely on a single source of depth information, resulting in insufficiently comprehensive geological information and difficulty in meeting the exploration needs of complex geological environments. Especially in the case of multiple depth parallaxes, the linear transformation of traditional methods leads to a significant expansion of the number of bits in the information, affecting the efficiency and accuracy of data processing. In addition, existing demasking procedures have short baseline lengths when processing deep geological information, failing to obtain high-precision depth information and limiting the reliability of exploration results.

[0003] Existing patent CN113537276B discloses a method and system for fusing multi-depth information. It uses a nonlinear transformation method to process multi-depth information, avoiding the bit expansion problem caused by traditional linear transformation. Furthermore, it fully utilizes high-precision depth information through optical center conversion technology, significantly improving the accuracy of the fusion results. However, when dealing with complex geological environments, it still suffers from insufficient baseline length and low information integration efficiency. Summary of the Invention

[0004] In order to overcome the shortcomings of the existing technology, the purpose of this invention is to provide an information fusion method and system for deep geological exploration.

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

[0006] An information fusion method for deep geological exploration includes:

[0007] Acquire multiple sets of depth information for the area to be explored, including shallow and deep data;

[0008] The shallow data is processed using an improved nonlinear transformation algorithm to obtain a first transformation information set;

[0009] The deep data shown is processed using deep learning algorithms to obtain the second transformation information set;

[0010] The first conversion information set and the second conversion information set are fused separately to obtain a first fused information subset and a second fused information subset;

[0011] The first and second fused information subsets are optimized and fused using tensor decomposition and joint inverse algorithms to obtain the final fused data.

[0012] Preferably, obtaining multiple sets of depth information for the area to be explored includes:

[0013] Establish a 3D model of the area to be measured;

[0014] Exploration routes for different baselines are obtained based on the 3D model;

[0015] Based on the exploration route of different baselines, the corresponding sensor equipment is selected to obtain depth information.

[0016] Preferably, establishing the three-dimensional model corresponding to the area to be tested includes:

[0017] Acquire real-time and historical multi-source data;

[0018] The geological characteristics of the area to be measured are obtained based on the real-time multi-source data and the historical multi-source data.

[0019] The geological units to be constructed in the three-dimensional model are determined based on the geological features described.

[0020] A three-dimensional model was constructed based on the geological units and the Kriging interpolation method using geological modeling software.

[0021] Preferably, the multi-source data includes:

[0022] Seismic data, drilling data, electromagnetic data, and remote sensing imagery.

[0023] Preferably, obtaining the geological features of the area to be measured based on the real-time multi-source data and the historical multi-source data includes:

[0024] The real-time multi-source data is cleaned, standardized, and noise-reduced to obtain a preprocessed real-time dataset;

[0025] The historical multi-source data is normalized and missing values ​​are filled in using interpolation methods to obtain a preprocessed historical dataset.

[0026] The preprocessed real-time dataset is fused using an improved Kalman filter algorithm to obtain real-time fused geological information;

[0027] The preprocessed historical dataset was processed using geostatistical methods to obtain historical fused geological information;

[0028] Geological features are extracted from the real-time fused geological information based on machine learning algorithms to obtain the first initial geological features;

[0029] The geological features of the historical fused geological information are extracted using an improved time series analysis method to obtain the second initial geological features.

[0030] The first initial geological feature and the second initial geological feature are fused to obtain the geological features of the area to be measured.

[0031] Preferably, the expression for the real-time fused geological information is:

[0032] A= ;

[0033] Where A represents real-time fused geological information. Here is the state transition matrix. For state vectors, Here is the Kalman gain matrix. This is the observation vector.

[0034] Preferably, the calculation expression for the second initial geological feature is:

[0035] ;

[0036] in, This is the second initial geological feature. For the variational mode decomposition of the m-th mode, For the m-th modality, a bidirectional long short-term memory network, Let m be the spatiotemporal gating weights for the m-th mode. For spatial attention mechanisms, For historical geological information integration, M represents the total number of modes.

[0037] Preferably, the step of processing the shallow data using an improved nonlinear transformation algorithm to obtain a first transformation information set includes:

[0038] The shallow data is denoised using an adaptive wavelet threshold, and the denoised data is then segmented and normalized according to the baseline length to obtain the preprocessed shallow data.

[0039] The local nonlinear characteristics of the preprocessed shallow data are obtained using a polynomial fitting function.

[0040] The global geological structure of the preprocessed shallow data is obtained using a Gaussian kernel function.

[0041] The first transformation information set is determined based on the local nonlinear characteristics and the global geological structure.

[0042] An information fusion system for deep geological exploration, comprising:

[0043] The data acquisition module is used to acquire multiple sets of depth information of the area to be explored, including shallow data and deep data.

[0044] The first conversion module is used to process the shallow data using an improved nonlinear conversion algorithm to obtain a first conversion information set;

[0045] The second transformation module is used to process the deep data shown by the deep learning algorithm to obtain the second transformation information set.

[0046] The primary fusion module is used to perform data fusion on the first conversion information set and the second conversion information set respectively to obtain a first fused information subset and a second fused information subset;

[0047] The final fusion module is used to optimize and fuse the first and second fusion information subsets using tensor decomposition and joint inverse algorithms to obtain the final fused data.

[0048] The present invention discloses the following technical effects:

[0049] This invention provides an information fusion method and system for deep geological exploration. The method includes: acquiring multiple sets of depth information of the area to be explored, the depth information including shallow data and deep data; processing the shallow data using an improved nonlinear transformation algorithm to obtain a first transformed information set; processing the deep data using a deep learning algorithm to obtain a second transformed information set; fusing the first and second transformed information sets to obtain a first fused information subset and a second fused information subset; and optimizing and fusing the first and second fused information subsets using a tensor decomposition algorithm and a joint inverse algorithm to obtain the final fused data. This invention avoids the loss of deep information due to signal attenuation during traditional single-baseline fusion by distinguishing between the independent processing of shallow data (short baselines) and deep data (long baselines); it compresses the dimension of shallow data by 65% ​​using an improved nonlinear transformation algorithm to eliminate redundant information bit expansion; and in the final fusion stage, it uses a "tensor decomposition + joint inversion" algorithm to perform cross-scale alignment of shallow-deep data from different baselines, solving the spatial misalignment problem between long and short baseline data. Attached Figure Description

[0050] 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.

[0051] Figure 1 A flowchart of an information fusion method for deep geological exploration provided in an embodiment of the present invention;

[0052] Figure 2This is a schematic diagram of the structure of an information fusion system for deep geological exploration provided in an embodiment of the present invention.

[0053] Figure label:

[0054] 1-Data acquisition module, 2-First conversion module, 3-Second conversion module, 4-Primary fusion module, 5-Final fusion module. Detailed Implementation

[0055] 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.

[0056] 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.

[0057] like Figure 1 As shown, the present invention provides an information fusion method for deep geological exploration, comprising:

[0058] Step 100: Obtain multiple sets of depth information for the area to be explored, including shallow and deep data;

[0059] Step 200: Process the shallow data using an improved nonlinear transformation algorithm to obtain a first transformation information set;

[0060] Step 300: Process the deep data shown using a deep learning algorithm to obtain the second transformation information set;

[0061] Step 400: Perform data fusion on the first conversion information set and the second conversion information set respectively to obtain a first fused information subset and a second fused information subset;

[0062] Step 500: Optimize and fuse the first and second fused information subsets using tensor decomposition and joint inverse algorithms to obtain the final fused data.

[0063] Specifically, baseline length refers to the spacing between sensor arrays (such as seismic wave transmitters and receivers) or the coverage area of ​​the detection path.

[0064] Short baseline (<500 meters): High-frequency signals attenuate quickly, making it suitable for capturing high-resolution data in shallow layers (0-1000 meters), but its penetration depth is limited;

[0065] Long baseline (>2000 meters): Low-frequency signals have strong penetrating power and are suitable for detecting large-scale structures in deep layers (1000-5000 meters), but the resolution is low.

[0066] More specifically, shallow data: short baseline (200-meter) seismic wave data, used to identify shallow rock layer interfaces (such as the boundary between sandstone and shale at 0-800 meters); sensor selection: high-resolution seismic detectors (frequency range 50-200 Hz), to match the high-frequency signal requirements of short baselines.

[0067] Deep data: Long baseline (3000 m) electromagnetic sounding data, used to detect deep fault extensions (such as basement fractures at 2000-4500 m); Sensor selection: Low-frequency electromagnetic transmitter (frequency range 0.1-10 Hz), adapted to the low-frequency penetration requirements of long baselines.

[0068] Furthermore, obtaining multiple sets of depth information for the area to be explored includes:

[0069] Establish a 3D model of the area to be measured;

[0070] Exploration routes for different baselines are obtained based on the 3D model;

[0071] Based on the exploration route of different baselines, the corresponding sensor equipment is selected to obtain depth information.

[0072] Specifically, short baseline route generation (200-500m spacing):

[0073] Extract the three-dimensional coordinates of the shallow target area;

[0074] Perform Delaunay triangulation to generate an initial triangular mesh;

[0075] Insert new nodes in regions where the triangle side length is greater than 200m, and iterate to refine the mesh until the side length is less than or equal to 200m;

[0076] Generate parallel survey lines along the sides of the triangle to ensure a coverage density of ≥2 lines / km. 2 .

[0077] Long baseline route generation (1-3km spacing):

[0078] Calculate the structural continuity of each voxel in the 3D model (based on density gradient consistency).

[0079] Define pheromone update rules;

[0080] Ant colony iterative search for the optimal path (iteration count ≥ 100).

[0081] The path is smoothed using Bézier curves to avoid terrain obstacles.

[0082] Table 1 shows information on baselines of different lengths. Table 1 is as follows:

[0083] Table 1

[0084] Baseline type coordinate range Coverage target Measuring line density short baseline (1000m, 2000m) - (1500m, 2500m) Shallow caves and fissure zones 5 lines / km² Long baseline (3000m, 500m) → (3000m, 3500m) Basement fractures, deep magma chamber 1 line / 2 / km² Hybrid baseline along the short baseline grid diagonal Cross-scale construction verification 2 lines / km²

[0085] Furthermore, establishing the three-dimensional model corresponding to the area to be measured includes:

[0086] Acquire real-time and historical multi-source data;

[0087] The geological characteristics of the area to be measured are obtained based on the real-time multi-source data and the historical multi-source data.

[0088] The geological units to be constructed in the three-dimensional model are determined based on the geological features described.

[0089] A three-dimensional model was constructed based on the geological units and the Kriging interpolation method using geological modeling software.

[0090] Furthermore, multi-source data includes:

[0091] Seismic data, drilling data, electromagnetic data, and remote sensing imagery.

[0092] Furthermore, obtaining the geological features of the area to be measured based on the real-time multi-source data and the historical multi-source data includes:

[0093] The real-time multi-source data is cleaned, standardized, and noise-reduced to obtain a preprocessed real-time dataset;

[0094] The historical multi-source data is normalized and missing values ​​are filled in using interpolation methods to obtain a preprocessed historical dataset.

[0095] The preprocessed real-time dataset is fused using an improved Kalman filter algorithm to obtain real-time fused geological information;

[0096] The preprocessed historical dataset was processed using geostatistical methods to obtain historical fused geological information;

[0097] Geological features are extracted from the real-time fused geological information based on machine learning algorithms to obtain the first initial geological features;

[0098] The geological features of the historical fused geological information are extracted using an improved time series analysis method to obtain the second initial geological features.

[0099] The first initial geological feature and the second initial geological feature are fused to obtain the geological features of the area to be measured.

[0100] The expression for the real-time fused geological information is as follows:

[0101] A= ;

[0102] Where A represents real-time fused geological information. Here is the state transition matrix. For state vectors, Here is the Kalman gain matrix. The observation vector is the state vector, which includes the rock layer thickness, fault location, and physical properties. Both the state vector and the observation vector are obtained by processing pre-processed real-time multi-source data through a first mapping function and a second mapping function.

[0103] First mapping function:

[0104] Extracting geological structure parameters:

[0105] Rock layer thickness: verified by cross-validation of seismic wave reflection time series and drilling data;

[0106] Fault location: Combining electromagnetic anomaly gradient with fault line characteristics from remote sensing images;

[0107] Physical properties: Rock density and porosity were inverted based on seismic wave velocity (Vp / Vs);

[0108] Data dimensionality reduction: compressing multidimensional sensor data (such as 128-dimensional electromagnetic spectrum features) into a low-dimensional state space (such as an 8-dimensional vector).

[0109] Timing alignment: Compensating for time delays between different sensors (such as the difference in propagation speed between seismic waves and electromagnetic signals).

[0110] Second mapping function:

[0111] The preprocessed real-time multi-source data is mapped to the observation vector Y(t);

[0112] Specific functions:

[0113] Generate direct observations:

[0114] Seismic wave amplitude (0.5-80Hz frequency band);

[0115] Electromagnetic field strength (10^ -6 ~10^ -3 V / m);

[0116] Drill core density (2.5-3.2 g / cm³);

[0117] Dynamic noise suppression:

[0118] Eliminate the interference of surface mechanical vibration on seismic data (>20Hz high-frequency noise);

[0119] Suppress atmospheric ionospheric interference in electromagnetic data (daytime / nighttime mode adaptive).

[0120] The state transition matrix is ​​derived from the information in the three-dimensional model. Confirmed; the 3D model information here is determined by constructing an initial 3D model from historical data.

[0121] ;

[0122] in, This is the state transition matrix;

[0123] The state vector is determined by preprocessing the real-time multi-source dataset;

[0124] The Kalman gain matrix is ​​derived from the error covariance matrix. and weight matrix Sure;

[0125] ;

[0126] ;

[0127] in, Here is the Kalman gain matrix. Here is the state transition matrix. This represents the prediction error covariance matrix. The prediction error covariance matrix at the previous time step. for transpose, The process noise covariance matrix; The observation vector is determined by preprocessing the real-time multi-source dataset; after introducing the weight matrix, the Kalman gain matrix can dynamically adjust the contribution of different data sources (seismic, drilling, electromagnetic, remote sensing) to improve the accuracy of the fusion results. The uncertainty of the state vector is quantified, guiding the calculation of the Kalman gain matrix and updating it. The algorithm can dynamically adjust the confidence level between predicted and observed values, thereby improving the robustness of the fusion results.

[0128] =diag(σ S 2 (t),σ D 2 (t),σ E 2 (t),σ R 2 (t)); σ S 2 (t),σ D 2 (t),σE 2 (t),σ R 2 (t) represents the error variance of seismic data, drilling data, electromagnetic data, and remote sensing image data, respectively.

[0129] The `diag` function is used to convert vectors into diagonal matrices, define the error covariance matrix and weight matrix, and ensure independence between different data sources.

[0130] Furthermore, the calculation expression for the second initial geological feature is as follows:

[0131] ;

[0132] in, This is the second initial geological feature. For the variational mode decomposition of the m-th mode, For the m-th modality, a bidirectional long short-term memory network, Let m be the spatiotemporal gating weights for the m-th mode. For spatial attention mechanisms, For historical geological information integration, M represents the total number of modes.

[0133] Specifically, Multimodal Adaptive Decomposition (VMD) replaces the preset wavelet basis to improve the accuracy of signal decomposition; Bidirectional Temporal Modeling (BiLSTM) captures more complete temporal dependencies; Spatiotemporal Gating jointly optimizes temporal and spatial attention and dynamically adjusts modal weights; End-to-End Spatial Attention directly extracts geological spatial features and avoids information loss.

[0134] The process of using an improved nonlinear transformation algorithm to process the shallow data to obtain a first transformation information set includes:

[0135] The shallow data is denoised using an adaptive wavelet threshold, and the denoised data is then segmented and normalized according to the baseline length to obtain the preprocessed shallow data.

[0136] Specifically, adaptive wavelet denoising: dynamically adjusts the threshold to preserve details for high-frequency noise characteristics (such as ground vibration interference) in short-baseline shallow data;

[0137] Segmented normalization: Divide the data according to the baseline length (e.g., every 200 meters) to eliminate local biases caused by differences in sensor sensitivity.

[0138] The local nonlinear characteristics of the preprocessed shallow data are obtained using a polynomial fitting function.

[0139] Specifically, the matrix of preprocessed shallow data (n data points, d-dimensional features) is obtained. The data is divided into a preset number of grids according to spatial coordinates, and each sub-block is processed independently. A quadratic polynomial is used to fit each sub-block to obtain the fitting coefficient matrix of each sub-block. The local fitting quality is evaluated based on the fitting coefficient matrix. The local curvature and gradient direction are calculated through the quadratic term coefficients, and local nonlinear features are extracted based on the local curvature, gradient direction, and their fitting coefficient matrix.

[0140] The global geological structure of the preprocessed shallow data is obtained using a Gaussian kernel function.

[0141] Specifically, the Gaussian radial basis function (RBF) is used to calculate the similarity between data points. A global kernel matrix is ​​constructed based on the similarity between the data points. The global kernel matrix is ​​then subjected to eigenvalue decomposition. The eigenvector corresponding to the largest eigenvalue is used as the global structural basis vector to determine the global geological structure. The global geological structure includes branch faults and anticline axes.

[0142] The first transformation information set is determined based on the local nonlinear characteristics and the global geological structure.

[0143] Specifically, local nonlinear features and global geological structures are mapped to a unified coordinate system for fusion to obtain the first transformed information set.

[0144] like Figure 2 As shown, this embodiment also discloses an information fusion system for deep geological exploration, including:

[0145] Data acquisition module 1 is used to acquire multiple sets of depth information of the area to be explored, including shallow data and deep data;

[0146] The first conversion module 2 is used to process the shallow data using an improved nonlinear conversion algorithm to obtain a first conversion information set;

[0147] The second transformation module 3 is used to process the deep data shown by the deep learning algorithm to obtain the second transformation information set.

[0148] The primary fusion module 4 is used to perform data fusion on the first conversion information set and the second conversion information set respectively to obtain a first fused information subset and a second fused information subset;

[0149] The final fusion module 5 is used to optimize and fuse the first fusion information subset and the second fusion information subset using tensor decomposition algorithm and joint inverse algorithm to obtain the final fused data.

[0150] 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.

[0151] 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. An information fusion method for deep geological exploration, characterized in that, include: Acquire multiple sets of depth information for the area to be explored, including shallow and deep data; The shallow data is processed using an improved nonlinear transformation algorithm to obtain a first transformation information set; The deep data is processed using deep learning algorithms to obtain the second transformation information set; The first conversion information set and the second conversion information set are fused separately to obtain a first fused information subset and a second fused information subset; The first and second fused information subsets are optimized and fused using tensor decomposition and joint inversion algorithms to obtain the final fused data. The acquisition of multiple sets of depth information of the area to be explored includes: Establish a 3D model of the area to be measured; Exploration routes for different baselines are obtained based on the 3D model; Based on the exploration route of different baselines, select the corresponding sensor equipment to obtain depth information; The process of establishing a three-dimensional model corresponding to the region to be tested includes: Acquire real-time and historical multi-source data; The geological characteristics of the area to be measured are obtained based on the real-time multi-source data and the historical multi-source data. The geological units to be constructed in the three-dimensional model are determined based on the geological features described. A three-dimensional model was constructed based on geological modeling software, according to the geological units and the Kriging interpolation method. Multi-source data includes: Seismic data, drilling data, electromagnetic data, and remote sensing imagery; The process of obtaining the geological features of the area to be measured based on the real-time multi-source data and the historical multi-source data includes: The real-time multi-source data is cleaned, standardized, and noise-reduced to obtain a preprocessed real-time dataset; The historical multi-source data is normalized and missing values ​​are filled in using interpolation methods to obtain a preprocessed historical dataset. The preprocessed real-time dataset is fused using an improved Kalman filter algorithm to obtain real-time fused geological information; The preprocessed historical dataset was processed using geostatistical methods to obtain historical fused geological information; Geological features are extracted from the real-time fused geological information based on machine learning algorithms to obtain the first initial geological features; The geological features of the historical fused geological information are extracted using an improved time series analysis method to obtain the second initial geological features. The first initial geological feature and the second initial geological feature are fused to obtain the geological features of the area to be measured. The expression for the real-time fused geological information is: ； Where A represents real-time fused geological information. Here is the state transition matrix. For state vectors, Here is the Kalman gain matrix. For observation vectors; ； ； in, This represents the prediction error covariance matrix. The process noise covariance matrix; The observation vector is derived from a preprocessed real-time multi-source dataset. It is confirmed that, after introducing the weight matrix, the Kalman gain matrix can dynamically adjust the contributions of different data sources. The uncertainty of the state vector is quantified, guiding the calculation of the Kalman gain matrix and updating it. ; The state vector includes rock layer thickness, fault location, and physical properties; both the state vector and the observation vector are obtained by preprocessing real-time multi-source data through a first mapping function and a second mapping function. First mapping function: Extract geological structural parameters: rock layer thickness, fault location, physical properties, data dimensionality reduction, and time series alignment; The second mapping function maps the preprocessed real-time multi-source data into the observation vector Y(t); ; These represent the error variances of seismic data, drilling data, electromagnetic data, and remote sensing image data, respectively. The calculation expression for the second initial geological feature is as follows: ； in, This is the second initial geological feature. For the variational mode decomposition of the m-th mode, For the m-th modality, a bidirectional long short-term memory network, Let m be the spatiotemporal gating weights for the m-th mode. For spatial attention mechanisms, To integrate historical geological information, M represents the total number of modes; Multimodal adaptive decomposition replaces preset wavelet basis; bidirectional temporal modeling captures more complete temporal dependencies; spatiotemporal gating jointly optimizes temporal and spatial attention, dynamically adjusting modal weights; end-to-end spatial attention directly extracts geological spatial features; Shallow data: short baseline seismic wave data; Deep data: Long baseline electromagnetic bathymetry data; Short baseline route generation, spacing 200-500m: Extract the three-dimensional coordinates of the shallow target area; Perform Delaunay triangulation to generate an initial triangular mesh; Insert new nodes in regions where the triangle side length is greater than 200m, and iterate to refine the mesh until the side length is less than or equal to 200m; Generate parallel survey lines along the sides of the triangle to ensure a coverage density of ≥2 lines / km. 2 ; Long baseline route generation, spacing 1-3km: Calculate the structural continuity of each voxel in the 3D model; Define pheromone update rules; Ant colony iteratively searches for the optimal path, with ≥100 iterations; The path is smoothed using Bézier curves to avoid terrain obstacles; The process of using an improved nonlinear transformation algorithm to process the shallow data to obtain a first transformation information set includes: The shallow data is denoised using an adaptive wavelet threshold, and the denoised data is then segmented and normalized according to the baseline length to obtain the preprocessed shallow data. Adaptive wavelet denoising: dynamically adjusts the threshold to preserve details for high-frequency noise characteristics of short-baseline shallow data; Segmented normalization: Data is segmented according to the baseline length to eliminate local biases caused by differences in sensor sensitivity; The local nonlinear characteristics of the preprocessed shallow data are obtained using a polynomial fitting function. The global geological structure of the preprocessed shallow data is obtained using a Gaussian kernel function. The first transformation information set is determined based on the local nonlinear characteristics and the global geological structure; Specifically, local nonlinear features and global geological structures are mapped to a unified coordinate system for fusion to obtain the first transformed information set.

2. An information fusion system for deep geological exploration, characterized in that, include: The data acquisition module is used to acquire multiple sets of depth information of the area to be explored, including shallow data and deep data. The first conversion module is used to process the shallow data using an improved nonlinear conversion algorithm to obtain a first conversion information set; The second transformation module is used to process the deep data shown by the deep learning algorithm to obtain the second transformation information set. The primary fusion module is used to perform data fusion on the first conversion information set and the second conversion information set respectively to obtain a first fused information subset and a second fused information subset; The final fusion module is used to optimize and fuse the first fusion information subset and the second fusion information subset using tensor decomposition algorithm and joint inversion algorithm to obtain the final fusion data; Shallow data: short baseline seismic wave data; Deep data: Long baseline electromagnetic bathymetry data; Short baseline route generation, spacing 200-500m: Extract the three-dimensional coordinates of the shallow target area; Perform Delaunay triangulation to generate an initial triangular mesh; Insert new nodes in regions where the triangle side length is greater than 200m, and iterate to refine the mesh until the side length is less than or equal to 200m; Generate parallel survey lines along the sides of the triangle to ensure a coverage density of ≥2 lines / km. 2 ; Long baseline route generation, spacing 1-3km: Calculate the structural continuity of each voxel in the 3D model; Define pheromone update rules; Ant colony iteratively searches for the optimal path, with ≥100 iterations; The path is smoothed using Bézier curves to avoid terrain obstacles; The process of using an improved nonlinear transformation algorithm to process the shallow data to obtain a first transformation information set includes: The shallow data is denoised using an adaptive wavelet threshold, and the denoised data is then segmented and normalized according to the baseline length to obtain the preprocessed shallow data. Adaptive wavelet denoising: dynamically adjusts the threshold to preserve details for high-frequency noise characteristics of short-baseline shallow data; Segmented normalization: Data is segmented according to the baseline length to eliminate local biases caused by differences in sensor sensitivity; The local nonlinear characteristics of the preprocessed shallow data are obtained using a polynomial fitting function. The global geological structure of the preprocessed shallow data is obtained using a Gaussian kernel function. The first transformation information set is determined based on the local nonlinear characteristics and the global geological structure; Specifically, local nonlinear features and global geological structures are mapped to a unified coordinate system for fusion to obtain the first transformed information set.

Images