A mine geological anomaly identification method and system based on machine learning

Abstract

This invention provides a machine learning-based method and system for identifying geological anomalies in mining areas, comprising: collecting and preprocessing multi-source geological data of a target mining area to construct a geological dataset; extracting multi-scale features from the geological dataset to obtain original data features, geological significance features, and spatial correlation features; constructing a geological anomaly identification model based on the original data features, geological significance features, and spatial correlation features using machine learning algorithms; and identifying geological anomalies in the mining area based on the geological anomaly identification model. This invention, through the deep integration of geology and artificial intelligence, not only achieves higher accuracy and efficiency in anomaly identification but also produces revolutionary technical effects in knowledge modeling, decision transparency, and exploration workflow optimization, providing a practical next-generation solution for intelligent mineral resource exploration.

Description

Technical Field

[0001] This invention belongs to the field of geological exploration technology, specifically relating to a method and system for identifying geological anomalies in mining areas based on machine learning. Background Technology

[0002] Traditional geological anomaly identification mainly relies on manual interpretation of geophysical, geochemical, and remote sensing data, which has the following drawbacks: 1. High subjectivity, with large differences in interpretation results among different experts; 2. Limited ability to comprehensively process multi-source data; 3. Difficulty in discovering complex nonlinear geological features; 4. Low efficiency, unable to meet the needs of rapid exploration in large-scale mining areas. Existing machine learning methods mostly use single algorithms or limited data types, which are not adaptable to the special geological conditions of mining areas and have limited identification accuracy. Summary of the Invention

[0003] This invention aims to address the shortcomings of existing technologies and provide an intelligent method that can automatically and accurately identify geological anomalies in mining areas. By fusing multi-source heterogeneous geological data and constructing an adaptive deep learning model, the accuracy and efficiency of geological anomaly identification are significantly improved.

[0004] To achieve the above objectives, the present invention provides the following solution: A machine learning-based method for identifying geological anomalies in mining areas, comprising: Collect and preprocess multi-source geological data from the target mining area to construct a geological dataset; Multi-scale feature extraction is performed on the geological dataset to obtain original data features, geological significance features, and spatial correlation features; Based on the characteristics of the original data, geological significance characteristics, and spatial correlation characteristics, a geological anomaly identification model is constructed using machine learning algorithms. Based on the geological anomaly identification model, the geological anomalies in the mining area are identified.

[0005] Preferably, the multi-source geological data includes geophysical data, geochemical data, remote sensing data, and geological background data; The geophysical data includes gravity, magnetic, and electromagnetic measurement data. The geochemical data includes surface and borehole element content data; The remote sensing data includes hyperspectral, multispectral, and radar remote sensing images; The geological background data includes stratigraphic information, structural information, and lithological information.

[0006] Preferably, the method for multi-scale feature extraction of the geological dataset includes: The preprocessed multi-source geological data is converted into spatial grid data with a unified coordinate system and resolution through a geologically constrained interpolation algorithm. The original numerical fields, gradients, and statistical distribution characteristics of each data source are directly extracted from the spatial grid data to obtain the original data characteristics; Based on the combination and ratio characteristics of elemental geochemical behavior, the alteration combination index of mineral spectral response, and the synergistic relationship characteristics obtained by quantifying the spatial coupling degree of multi-source anomalies, the regional metallogenic theoretical model is quantified into a computable index to obtain geologically significant characteristics. Based on the original data features and the geological significance features, a geologically weighted spatial autocorrelation analysis method is adopted. By integrating distance attenuation, geological unit similarity and tectonic orientation, a heterogeneous spatial weight matrix is ​​constructed to extract spatial correlation features that conform to geological laws.

[0007] Preferably, the method for obtaining the alteration combination index of mineral spectral response includes: based on the ore deposit alteration zoning model, calculating the spatial similarity between the single mineral alteration map extracted by remote sensing and the ideal alteration mineral combination pattern, and obtaining the alteration combination type matching degree and zoning position index of each pixel.

[0008] Preferably, the structure of the geological anomaly identification model includes: The multi-branch feature alignment input layer is used to align the original data features, geological significance features, and spatial correlation features to obtain an aligned multi-source feature map. A geological attention-guided feature fusion module is used to receive the multi-source feature map and perform adaptive weighted fusion of the multi-source feature map through an attention weight map generated by a dynamic feature fusion gating mechanism to obtain a fused feature map. The U-Net++ network layer is used to perform multi-scale feature abstraction and contextual information aggregation on the fused feature map using a dense skip connection structure to obtain a geological anomaly probability segmentation map.

[0009] The present invention also provides a machine learning-based geological anomaly identification system for mining areas, used to implement the method, comprising: The data acquisition module is used to collect and preprocess multi-source geological data from the target mining area to construct a geological dataset. The multi-scale feature extraction module is used to perform multi-scale feature extraction on the geological dataset to obtain original data features, geological significance features, and spatial correlation features. The model building module is used to construct a geological anomaly identification model based on the original data features, geological significance features, and spatial correlation features, using machine learning algorithms. The anomaly identification module is used to identify geological anomalies in the mining area based on the geological anomaly identification model.

[0010] Preferably, the multi-scale feature extraction module includes: The grid division unit is used to convert preprocessed multi-source geological data into spatial grid data with a unified coordinate system and resolution through a geologically constrained interpolation algorithm; The raw data feature extraction unit is used to directly extract the raw numerical field, gradient and statistical distribution features of each data source from the spatial grid data to obtain the raw data features; The geological significance feature extraction unit is used to quantify regional metallogenic theoretical models into computable indicators based on the combination and ratio characteristics of elemental geochemical behavior, the alteration combination index of mineral spectral response, and the synergistic relationship characteristics obtained by quantifying the spatial coupling degree of multi-source anomalies, thereby obtaining geological significance features. The spatial correlation feature extraction unit is used to extract spatial correlation features that conform to geological laws by using a geologically weighted spatial autocorrelation analysis method based on the original data features and the geological significance features, and by constructing a heterogeneous spatial weight matrix by fusing distance attenuation, geological unit similarity and tectonic orientation.

[0011] Preferably, in the model building module, the structure of the geological anomaly identification model includes: The multi-branch feature alignment input layer is used to align the original data features, geological significance features, and spatial correlation features to obtain an aligned multi-source feature map. A geological attention-guided feature fusion module is used to receive the multi-source feature map and perform adaptive weighted fusion of the multi-source feature map through an attention weight map generated by a dynamic feature fusion gating mechanism to obtain a fused feature map. The U-Net++ network layer is used to perform multi-scale feature abstraction and contextual information aggregation on the fused feature map using a dense skip connection structure to obtain a geological anomaly probability segmentation map.

[0012] Compared with existing technologies, the advantages of this invention are as follows: Traditional methods rely on expert interpretation, which is highly subjective and inconsistent. This invention quantifies regional metallogenic theories into "geological significance characteristics" and encodes geological spatial patterns into "spatial correlation characteristics," enabling the model to learn the deep geological patterns hidden in the data, rather than simple statistical patterns.

[0013] This method creatively solves the "black box" problem of machine learning models, which hinders their application in the geological field. By using quantitative features with clear geological meaning as input, the model's predictions can be directly correlated with specific geological elements (e.g., "high anomaly probability is mainly controlled by the combined anomalies in silicification alteration intensity and copper-molybdenum ratio"). Furthermore, the process of constructing "geologically significant features" is itself a digital accumulation and inheritance of expert knowledge, making exploration decisions no longer untraceable model outputs, but rather evidence-based and interactively discussed intelligent suggestions. Attached Figure Description

[0014] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are 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.

[0015] Figure 1 This is a flowchart of a machine learning-based method for identifying geological anomalies in mining areas, as described in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the method for multi-scale feature extraction from geological datasets according to an embodiment of the present invention. Figure 3 This is a structural diagram of a machine learning-based geological anomaly identification system for mining areas, as described in an embodiment of the present invention. Detailed Implementation

[0016] 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.

[0017] 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.

[0018] Example 1: like Figure 1 As shown, a machine learning-based method for identifying geological anomalies in mining areas includes: S1: Collect and preprocess multi-source geological data of the target mining area to construct a geological dataset; a further implementation method is that the multi-source geological data includes geophysical data, geochemical data, remote sensing data, and geological background data; geophysical data includes gravity, magnetic, and electromagnetic measurement data; geochemical data includes surface and borehole element content data; remote sensing data includes hyperspectral, multispectral, and radar remote sensing images; geological background data includes stratigraphic information, structural information, and lithological information.

[0019] In the data preprocessing stage, this invention introduces an intelligent data purification and fusion process guided by geological constraints. First, spatiotemporal benchmarks are unified for multi-source data. Gravity and magnetic data are corrected using variable density and variable magnetic susceptibility to eliminate the influence of regional background fields, and electromagnetic data undergoes deep focusing inversion to obtain electrical structure slices. For geochemical data, a geological unit-based zoning standardization method is adopted, dividing statistical units according to stratigraphic and lithological boundaries to calculate background values ​​and anomaly thresholds, thereby eliminating elemental abundance differences caused by lithology itself. Three-dimensional kriging interpolation is then used to fuse surface and borehole data to generate an elemental concentration volume model. In remote sensing data preprocessing, a hybrid pixel decomposition algorithm based on a typical alteration mineral spectral library of the mining area is developed for hyperspectral imagery to accurately extract alteration mineral abundance maps. For multispectral and radar data, topographic correction and multi-temporal fusion are performed to enhance structural and lithological identification information. Geological background data is structured into a computable knowledge graph, encoding stratigraphic contact relationships, structural sequences, and lithological coexistence combinations into spatial objects with topological attributes. Finally, through spatial registration technology guided by geological semantics, all the above data layers are unified into a variable resolution grid dominated by geological meaning—high resolution is used in structurally complex areas and around known mineral deposits, while low resolution is used in homogeneous areas, forming a three-dimensional geological data cube containing rich correlation information, laying the foundation for subsequent feature extraction that combines data quality and geological consistency.

[0020] S2: Perform multi-scale feature extraction on the geological dataset to obtain original data features, geological significance features, and spatial correlation features; such as... Figure 2 As shown.

[0021] A further implementation method for multi-scale feature extraction from geological datasets includes: S21: The preprocessed multi-source geological data is converted into spatial grid data with a unified coordinate system and resolution using a geologically constrained interpolation algorithm. In this process, an adaptive interpolation algorithm guided by geological boundaries is employed. This algorithm first identifies key geological boundaries, such as fault lines, rock mass contact surfaces, and stratigraphic unconformities, using these boundaries as insurmountable hard constraints during interpolation to ensure that the geophysical field or geochemical gradient exhibits discontinuous changes consistent with geological laws on both sides of the boundaries. Based on this, the local grid resolution is dynamically adjusted according to the spatial distribution density and geological complexity of the data: high-resolution grids are used in densely drilled areas, structural intersections, or known mineralization areas to capture fine anomaly features; while lower resolutions are used in areas with sparse data or uniform geological backgrounds to ensure computational efficiency and stability. For geochemical data, the interpolation process further introduces anisotropic parameters of elemental geochemical behavior, such as setting a larger search radius along the structural strike direction to reflect the directionality of hydrothermal migration. The final result is a three-dimensional grid data volume with enhanced geological consistency. Each grid node not only contains attribute values ​​after the fusion of multi-source data, but also includes the encoding of its geological unit and distance information to key geological boundaries, forming a standardized data base that combines numerical accuracy and geological semantics.

[0022] S22: Directly extract the original numerical fields, gradients, and statistical distribution characteristics of each data source from the spatial grid data to obtain the original data characteristics; S23: Based on the combination and ratio characteristics of elemental geochemical behavior, the alteration combination index of mineral spectral response, and the synergistic relationship characteristics obtained by quantifying the spatial coupling degree of multi-source anomalies, the regional metallogenic theoretical model is quantified into a computable index to obtain geologically significant characteristics. The process of quantifying the geochemical assemblage and zonal characteristics of elements includes: calculating the dynamic element coexistence matching degree and vertical zonal sequence pointer value for each sampling point based on the element coexistence law and vertical zonal theory; the vertical zonal sequence pointer value is obtained by encoding the regional standard element zonal sequence into a numerical sequence and calculating the weighted average position of the element combination of the sampling point in the sequence with the logarithm of the element concentration as the weight. Methods for obtaining alteration combination indices of mineral spectral responses include: based on the ore deposit alteration zoning model, calculating the spatial similarity between the remotely sensed single-mineral alteration map and the ideal alteration mineral combination pattern, and obtaining the alteration combination type matching degree and zoning position index of each pixel.

[0023] The process of quantifying the synergistic relationship of multiple anomalies includes: spatially overlaying geophysical local anomalies, geochemical composite anomalies, and remote sensing alteration anomalies; generating a multi-source synergistic anomaly intensity map by calculating the spatial coupling degree between each pair and the probability of joint occurrence of the three within a sliding window.

[0024] The above-mentioned quantitative indicators are integrated with geological environmental variables, including distance to a specific structure, lithological combination coding, and stratigraphic mineralization index. All the above-mentioned quantitative indicators are normalized through geological unit zoning, and finally a geological significance feature vector with clear geological interpretation significance is formed.

[0025] S24: Based on the characteristics of the original data and geological significance, a geologically weighted spatial autocorrelation analysis method is adopted. A heterogeneous spatial weight matrix is ​​constructed by integrating distance attenuation, geological unit similarity, and structural orientation to extract spatial correlation features that conform to geological laws. Specifically, in constructing the geologically weighted spatial autocorrelation analysis to extract spatial correlation features, this invention creatively designs a heterogeneous spatial weight matrix that integrates multi-dimensional geological constraints. The construction of this matrix is ​​not based on a traditional single distance attenuation function, but integrates three layers of geological logic: First, the distance attenuation term uses an anisotropic distance metric, with the attenuation rate along a known structural strike being lower than that perpendicular to the strike, reflecting the continuity of geological elements along structural extensions; second, a geological unit similarity factor is introduced, giving higher connection weights to two spatial units belonging to the same stratum, lithology, or known ore-bearing geological body, while significantly reducing weights when crossing important geological boundaries; finally, a structural orientation enhancement mechanism is embedded, assigning higher correlation weights to unit pairs located within the same linear structural zone or favorable ore-forming structural combination (such as fault intersection zones), even if the spatial distance is slightly greater. The improved local Moran's I index, calculated based on this weight matrix, can not only identify ordinary spatial clusters but also specifically detect anomalous spatial clustering patterns with clear geological significance, such as those "distributed along tectonic zones," "controlled by specific lithologies," or "limited to a certain stratigraphic unit." Furthermore, by analyzing changes in spatial autocorrelation values ​​through multi-scale sliding window analysis, the multi-level structural characteristics of geological anomalies, from local enrichment to regional distribution, can be further revealed, thus deeply uncovering the metallogenic geological laws hidden in spatial relationships.

[0026] S3: Based on the characteristics of the original data, geological significance characteristics, and spatial correlation characteristics, a geological anomaly identification model is constructed using machine learning algorithms; A further implementation method is that the structure of the geological anomaly identification model includes: The multi-branch feature alignment input layer aligns the original data features, geological significance features, and spatial correlation features to obtain an aligned multi-source feature map. The geological attention-guided feature fusion module receives the multi-source feature map and performs adaptive weighted fusion on the multi-source feature map through an attention weight map generated by a dynamic feature fusion gating mechanism to obtain a fused feature map. The U-Net++ network layer uses a dense skip connection structure to perform multi-scale feature abstraction and contextual information aggregation on the fused feature map to obtain a geological anomaly probability segmentation map.

[0027] Specifically, this invention provides a deep learning-based intelligent identification model architecture for geological anomalies in mining areas. Its core innovation lies in constructing a complete technical chain, from precise alignment of multi-source features and dynamic fusion guided by geological knowledge to multi-scale context-aware segmentation. The model first processes heterogeneous features through a multi-branch feature alignment input layer. This layer uses convolution, fully connected encoding, and spatial replication operations on the grid structure of the original data features, the structured vectors of geologically significant features, and the prior maps of spatially correlated features, transforming them to the same spatial dimension and resolution, forming a spatially strictly registered multi-source feature map stack. Subsequently, it enters a geological attention-guided feature fusion module. The innovation of this module lies in its dynamic gating mechanism, which is not driven by data-driven spontaneous learning but explicitly guided by geologically significant features. Geologically significant features containing mineralization knowledge are input into a lightweight sub-network, generating a set of spatially adaptive three-channel attention weight maps, corresponding to the importance of the original, geological, and spatial features respectively. Fusion is achieved through element-wise weighting. This process makes the model pay more attention to structural features near faults and rely more on geochemical indicators in altered areas. The fused unified feature map is input into the improved U-Net++ network layer. This network enhances multi-scale information flow with dense skip connections and introduces deformable convolutional units in the encoder. The kernel offsets of these units are initialized with structural orientation information from spatially correlated features, allowing the feature extraction process to adapt to the geometric shape of geological body boundaries. In the decoding path, a spatial-channel dual attention mechanism is designed to refine upsampled features using high-level semantic context information while suppressing irrelevant noise. The entire model is trained end-to-end using a hybrid loss function constrained by geological rules. In addition to the standard pixel-level segmentation loss, this loss function adds a boundary consistency loss term to force the model's predicted anomaly boundaries to be spatially compatible with known geological boundaries, and a feature reconstruction loss term, requiring the intermediate layer features to invert key geological significance indicators, thereby deeply embedding domain knowledge into the learning objective. The final model not only outputs a high-precision anomaly probability map but also provides pixel-by-pixel uncertainty estimation through Monte Carlo Dropout, forming a "probability-confidence" dual-map output, providing a quantitative basis for exploration decisions. This architecture achieves a deep integration of geological prior knowledge and data-driven learning, enabling the model to possess reasoning logic similar to that of an expert, and significantly improving the accuracy and interpretability of identifying hidden anomalies in complex geological contexts.

[0028] S4: Based on the geological anomaly identification model, complete the identification of geological anomalies in the mining area.

[0029] Example 2: like Figure 3 As shown, the present invention also provides a machine learning-based geological anomaly identification system for mining areas, used to implement the method of Embodiment 1, comprising: The data acquisition module is used to collect and preprocess multi-source geological data from the target mining area to construct a geological dataset. The multi-scale feature extraction module is used to extract multi-scale features from geological datasets to obtain original data features, geological significance features, and spatial correlation features. The model building module is used to construct a geological anomaly identification model based on the original data features, geological significance features, and spatial correlation features, using machine learning algorithms. The anomaly identification module is used to identify geological anomalies in the mining area based on the geological anomaly identification model.

[0030] A further implementation method includes a multi-scale feature extraction module comprising: The grid division unit is used to convert preprocessed multi-source geological data into spatial grid data with a unified coordinate system and resolution through a geologically constrained interpolation algorithm; The raw data feature extraction unit is used to directly extract the raw numerical fields, gradients, and statistical distribution features of each data source from the spatial grid data to obtain the raw data features. The geological significance feature extraction unit is used to quantify regional metallogenic theoretical models into computable indicators based on the combination and ratio characteristics of elemental geochemical behavior, the alteration combination index of mineral spectral response, and the synergistic relationship characteristics obtained by quantifying the spatial coupling degree of multi-source anomalies, thereby obtaining geological significance features. The spatial correlation feature extraction unit is used to extract spatial correlation features that conform to geological laws by using a geologically weighted spatial autocorrelation analysis method based on the original data features and geological significance features, and by integrating distance attenuation, geological unit similarity and tectonic orientation to construct a heterogeneous spatial weight matrix.

[0031] A further implementation method is that, in the model building module, the structure of the geological anomaly identification model includes: The multi-branch feature alignment input layer is used to align the original data features, geological significance features, and spatial correlation features to obtain an aligned multi-source feature map. The geological attention-guided feature fusion module receives multi-source feature maps and performs adaptive weighted fusion of the multi-source feature maps using an attention weight map generated by a dynamic feature fusion gating mechanism to obtain a fused feature map. The U-Net++ network layer is used to perform multi-scale feature abstraction and contextual information aggregation on the fused feature map using a dense skip connection structure to obtain a geological anomaly probability segmentation map.

[0032] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for identifying geological anomalies in mining areas based on machine learning, characterized in that, include: Collect and preprocess multi-source geological data from the target mining area to construct a geological dataset; Multi-scale feature extraction is performed on the geological dataset to obtain original data features, geological significance features, and spatial correlation features; Based on the characteristics of the original data, geological significance characteristics, and spatial correlation characteristics, a geological anomaly identification model is constructed using machine learning algorithms. Based on the geological anomaly identification model, the geological anomalies in the mining area are identified.

2. The method according to claim 1, characterized in that, The multi-source geological data includes geophysical data, geochemical data, remote sensing data, and geological background data; The geophysical data includes gravity, magnetic, and electromagnetic measurement data. The geochemical data includes surface and borehole element content data; The remote sensing data includes hyperspectral, multispectral, and radar remote sensing images; The geological background data includes stratigraphic information, structural information, and lithological information.

3. The method according to claim 1, characterized in that, Methods for multi-scale feature extraction of the geological dataset include: The preprocessed multi-source geological data is converted into spatial grid data with a unified coordinate system and resolution through a geologically constrained interpolation algorithm. The original numerical fields, gradients, and statistical distribution characteristics of each data source are directly extracted from the spatial grid data to obtain the original data characteristics; Based on the combination and ratio characteristics of elemental geochemical behavior, the alteration combination index of mineral spectral response, and the synergistic relationship characteristics obtained by quantifying the spatial coupling degree of multi-source anomalies, the regional metallogenic theoretical model is quantified into a computable index to obtain geologically significant characteristics. Based on the original data features and the geological significance features, a geologically weighted spatial autocorrelation analysis method is adopted. By integrating distance attenuation, geological unit similarity and tectonic orientation, a heterogeneous spatial weight matrix is ​​constructed to extract spatial correlation features that conform to geological laws.

4. The method according to claim 3, characterized in that, Methods for obtaining alteration combination indices of mineral spectral responses include: based on the ore deposit alteration zoning model, calculating the spatial similarity between the remotely sensed single-mineral alteration map and the ideal alteration mineral combination pattern, and obtaining the alteration combination type matching degree and zoning position index of each pixel.

5. The method according to claim 1, characterized in that, The structure of the geological anomaly identification model includes: The multi-branch feature alignment input layer is used to align the original data features, geological significance features, and spatial correlation features to obtain an aligned multi-source feature map. A geological attention-guided feature fusion module is used to receive the multi-source feature map and perform adaptive weighted fusion of the multi-source feature map through an attention weight map generated by a dynamic feature fusion gating mechanism to obtain a fused feature map. The U-Net++ network layer is used to perform multi-scale feature abstraction and contextual information aggregation on the fused feature map using a dense skip connection structure to obtain a geological anomaly probability segmentation map.

6. A machine learning-based geological anomaly identification system for mining areas, used to implement the method described in any one of claims 1-5, characterized in that, include: The data acquisition module is used to collect and preprocess multi-source geological data from the target mining area to construct a geological dataset. The multi-scale feature extraction module is used to perform multi-scale feature extraction on the geological dataset to obtain original data features, geological significance features, and spatial correlation features. The model building module is used to construct a geological anomaly identification model based on the original data features, geological significance features, and spatial correlation features, using machine learning algorithms. The anomaly identification module is used to identify geological anomalies in the mining area based on the geological anomaly identification model.

7. The system according to claim 6, characterized in that, The multi-scale feature extraction module includes: The grid division unit is used to convert preprocessed multi-source geological data into spatial grid data with a unified coordinate system and resolution through a geologically constrained interpolation algorithm; The raw data feature extraction unit is used to directly extract the raw numerical field, gradient and statistical distribution features of each data source from the spatial grid data to obtain the raw data features; The geological significance feature extraction unit is used to quantify regional metallogenic theoretical models into computable indicators based on the combination and ratio characteristics of elemental geochemical behavior, the alteration combination index of mineral spectral response, and the synergistic relationship characteristics obtained by quantifying the spatial coupling degree of multi-source anomalies, thereby obtaining geological significance features. The spatial correlation feature extraction unit is used to extract spatial correlation features that conform to geological laws by using a geologically weighted spatial autocorrelation analysis method based on the original data features and the geological significance features, and by constructing a heterogeneous spatial weight matrix by fusing distance attenuation, geological unit similarity and tectonic orientation.

8. The system according to claim 6, characterized in that, The geological anomaly identification model in the model construction module includes the following structure: The multi-branch feature alignment input layer is used to align the original data features, geological significance features, and spatial correlation features to obtain an aligned multi-source feature map. A geological attention-guided feature fusion module is used to receive the multi-source feature map and perform adaptive weighted fusion of the multi-source feature map through an attention weight map generated by a dynamic feature fusion gating mechanism to obtain a fused feature map. The U-Net++ network layer is used to perform multi-scale feature abstraction and contextual information aggregation on the fused feature map using a dense skip connection structure to obtain a geological anomaly probability segmentation map.

Images