A Machine Learning-Based Intelligent Identification Method for Anomalous Geological Bodies in Mineral Exploration

By improving the federated graph convolution model and multi-scale three-dimensional residual convolutional network, and combining geological prior thresholds and connected component analysis, the problems of cross-mining right exploration data privacy and multi-scale feature fusion are solved, realizing high-precision and engineering-practical identification of anomalous geological bodies, which is suitable for mineral exploration scenarios with multiple mining rights zones.

CN123071718APending Publication Date: 2026-11-10张正常
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611242885.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-17
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Traditional technologies struggle to integrate multi-scale features that combine local fine geological attributes with regional macro-mineral control patterns. The spatial continuity and boundary positioning accuracy of anomaly identification results are insufficient, cross-mineral rights exploration data privacy is restricted, single-zone model training samples are insufficient and generalization ability is weak, identification results lack standardized hierarchical constraints and geological prior verification, and the engineering practicality of intelligent identification needs to be improved.

Method used

An improved federated graph convolutional recognition model is used for cross-mining rights collaborative training. Geological features are extracted by combining a multi-scale three-dimensional residual convolutional network. Geological prior constraint thresholds and connected component analysis are introduced to generate three-dimensional visualized anomaly geological body outline annotations and output a credibility assessment report.

Benefits of technology

While ensuring data privacy, it improves the accuracy and efficiency of identifying abnormal geological bodies, adapts to multi-mining rights zoning scenarios, and the identification results have good engineering applicability and hierarchical evaluation capabilities, significantly improving the automation level of intelligent identification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123071718A_ABST
    Figure CN123071718A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of mineral exploration and intelligent identification technology, and specifically relates to an intelligent identification method for anomalous geological bodies in mineral exploration based on machine learning. The method includes: acquiring multi-source geological exploration raw data and known anomalous geological body labeled sample datasets from the exploration area; constructing a three-dimensional grid geological feature dataset and extracting spatial multi-scale geological feature vectors; outputting a grid cell anomaly confidence distribution map using an improved federated graph convolutional machine learning identification model; dividing multi-level anomalous regions by combining geological prior constraint thresholds and generating three-dimensional anomalous geological body contour labels; and finally outputting the identification results and a confidence assessment report. This solves the problems of data privacy restrictions across mining rights, insufficient fusion of multi-scale geological features, and low accuracy and efficiency in identifying anomalous geological bodies.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of mineral exploration and intelligent identification technology, and specifically relates to an intelligent identification method for anomalous geological bodies in mineral exploration based on machine learning. Background Technology

[0002] As mineral exploration gradually advances towards deeper, concealed, and difficult-to-identify minerals, the industry has accumulated massive amounts of multi-source geological exploration data, covering multiple dimensions such as geophysics, geochemistry, geological mapping, and drilling verification. Utilizing artificial intelligence technology to achieve automated and intelligent identification of anomalous geological bodies has become a core development direction for improving mineral exploration efficiency, reducing exploration costs, and mitigating the subjective biases of human interpretation. As a core basis for delineating ore-forming target areas, the accuracy of anomalous geological bodies in identification and spatial positioning directly determines the effectiveness of subsequent exploration projects.

[0003] In related technologies, the identification of anomalous geological bodies largely relies on the human experience of geologists to comprehensively interpret multi-source data; some intelligent solutions use traditional machine learning algorithms or two-dimensional convolutional neural networks to extract anomalies from geological profiles or planar exploration data; a few three-dimensional recognition solutions use conventional three-dimensional convolutional networks to extract features and classify voxelized geological data.

[0004] However, traditional technologies struggle to integrate multi-scale features that combine local fine geological attributes with regional macro-mineral control patterns, resulting in insufficient spatial continuity and boundary positioning accuracy in anomaly identification results. Exploration data from different mining rights zones are constrained by ownership and confidentiality requirements, making it impossible to directly aggregate them for model training, leading to significant data silos and insufficient training samples and weak generalization ability for single-zone models. Furthermore, the identification results lack standardized hierarchical constraints and geological prior verification, making it difficult to directly support the hierarchical evaluation and engineering implementation of exploration target areas. The engineering practicality of intelligent identification needs to be improved. Summary of the Invention

[0005] This application provides a machine learning-based intelligent identification method for anomalous geological bodies in mineral exploration, which addresses issues such as limited privacy of cross-mining rights exploration data, low fusion of multi-scale geological features, and difficulty in balancing the accuracy and efficiency of anomalous geological body identification in related technologies.

[0006] The first aspect of this application provides a machine learning-based intelligent identification method for anomalous geological bodies in mineral exploration, comprising the following steps: acquiring multi-source geological exploration raw data of the exploration area and a dataset of labeled known anomalous geological bodies; The multi-source geological exploration raw data are processed to construct a three-dimensional grid geological feature dataset. The three-dimensional grid geological feature dataset is then input into a feature fusion network to extract spatial multi-scale geological feature vectors. The spatial multi-scale geological feature vectors and the labeled sample dataset of known anomalous geological bodies are input into the improved federated graph convolutional machine learning recognition model, and the output is a grid cell anomaly confidence distribution map. Based on the anomaly confidence distribution map and combined with geological prior constraint thresholds, multi-level anomaly regions are divided, connected anomaly blocks are extracted as candidate anomaly geological bodies, and three-dimensional visualization of anomaly geological body outline annotations are generated. Based on the outline annotation of the anomalous geological bodies and the feedback data from manual verification, the results of the anomalous geological body identification and a credibility assessment report are output.

[0007] Preferably, the spatial multi-scale geological feature vectors and the labeled sample dataset of known anomalous geological bodies are input into the improved federated graph convolutional machine learning recognition model, and the output grid cell anomaly confidence distribution map is generated, including: An improved federated graph convolutional recognition model is constructed, wherein the improved federated graph convolutional recognition model is divided into a local mining rights partitioning graph convolutional sub-module and a cloud-based federated aggregation module. Using each 3D grid cell as a graph node, an adjacency graph matrix is ​​constructed based on the grid spatial adjacency relationship. Spatial multi-scale geological feature vectors are used as node features, and the known anomalous geological body labeled sample dataset is used as local supervision labels. The local mining rights partition map convolution submodule is input, and the partition blocks complete local anomaly classification training. At the same time, the model gradient parameters are uploaded to the cloud federated aggregation module to perform global parameter updates on the improved federated graph convolution recognition model without transmitting the original geological exploration data and labeled samples. The updated improved federated graph convolution model is distributed to each mining rights partition, inference is performed on all voxel grids within the partition, the probability value of mineralization anomaly corresponding to each voxel grid is calculated, and a global grid cell anomaly confidence distribution map is generated.

[0008] Preferably, the improved federated graph convolutional node feature update formula is as follows:

[0009] in, For the first Features of layer grid node v after update; is the activation function; W is the node feature transformation weight; For node v, it is the set of adjacent meshes in space; The spatial association attention coefficient between node v and its neighboring node u; This is the spatial adjacency edge weight matrix; Let u be the feature vector of the l-th layer adjacent node; For grid node v Layer feature vectors.

[0010] Preferably, the step of dividing multi-level anomaly regions based on the anomaly confidence distribution map combined with geological prior constraint thresholds includes: Set three levels of anomaly confidence thresholds: first threshold τ1, second threshold τ2, and third threshold τ3, where τ1 < τ2 < τ3; Mesh cells with anomaly confidence greater than or equal to τ3 are classified as first-level anomaly regions, mesh cells with anomaly confidence greater than or equal to τ2 and less than τ3 are classified as second-level anomaly regions, and mesh cells with anomaly confidence greater than or equal to τ1 and less than τ2 are classified as third-level anomaly regions. Connectivity analysis is performed on the connected grid cells within each level of anomaly region to extract spatially continuous connected anomaly blocks, and isolated noisy grid cells are filtered out to obtain a set of candidate anomaly geological bodies.

[0011] Preferably, the feature extraction formula of the multi-scale three-dimensional residual convolutional fusion network is as follows:

[0012] in, This represents the multi-scale geological feature vector after voxel mesh fusion; Extracting local attribute features for shallow 3D convolution; To extract transition features for mid-layer dilated convolution; To extract regional ore-controlling features using deep, large-scale convolution; This is a cross-layer feature attention weighted fusion function.

[0013] Preferably, the multi-source geological exploration raw data is processed to construct a three-dimensional grid geological feature dataset, including: Construct a three-dimensional spatial grid division model for the exploration area; Based on the three-dimensional spatial grid partitioning model, the entire exploration area is divided into voxels, and the original geological exploration data from multiple sources is interpolated and mapped to each voxel grid cell. The multi-dimensional geological attribute data of each voxel grid cell are processed and the spatial attribute labels of strata, structures and rock masses are integrated to generate a three-dimensional grid geological feature dataset with spatial coordinates and multi-attribute features.

[0014] Preferably, the three-dimensional mesh geological feature dataset is input into a feature fusion network to extract spatial multi-scale geological feature vectors, including: Construct a multi-scale 3D residual convolutional fusion network; Based on the multi-scale three-dimensional residual convolutional fusion network, three-dimensional convolutional kernels with different expansion coefficients are used to extract shallow local geological attribute features, middle-layer structural transition features, and deep regional macro-mineral control features of grid units in layers. The shallow local geological attribute features, the middle structural transition features, and the deep regional macro-mineral control features are weighted and fused by the cross-layer attention fusion module to output a unified spatial multi-scale geological feature vector.

[0015] The second aspect of this application provides a machine learning-based intelligent identification system for anomalous geological bodies in mineral exploration, comprising: an acquisition module for acquiring multi-source geological exploration raw data of the exploration area and a dataset of labeled samples of known anomalous geological bodies; The extraction module is used to process the multi-source geological exploration raw data, construct a three-dimensional grid geological feature dataset, input the three-dimensional grid geological feature dataset into a feature fusion network, and extract spatial multi-scale geological feature vectors. The identification module is used to input the spatial multi-scale geological feature vectors and the labeled sample dataset of known anomalous geological bodies into the improved federated graph convolutional machine learning identification model, and output the grid cell anomaly confidence distribution map. The segmentation module is used to segment multi-level abnormal regions based on the anomaly confidence distribution map and geological prior constraint thresholds, extract connected abnormal blocks as candidate abnormal geological bodies, and generate three-dimensional visualized anomaly geological body outline annotations. The output module is used to output the results of the identification of the abnormal geological bodies and the credibility assessment report based on the outline annotation of the abnormal geological bodies and the feedback data of manual verification.

[0016] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the machine learning-based intelligent identification method for mineral exploration anomalies as described in the above embodiments.

[0017] The fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a machine learning-based intelligent identification method for mineral exploration anomalies as described in the above embodiments.

[0018] Therefore, this application has the following beneficial effects: (1) An improved federated graph convolutional recognition architecture is adopted. Local training is completed through the local mining rights partition map convolutional sub-module, and global parameter updates are achieved through the cloud federated aggregation module. During the training process, only the model gradient parameters are transmitted, and the original exploration data and labeled samples are not shared. Under the premise of ensuring cross-mining rights data privacy and ownership security, the full-domain samples are fully utilized to improve the generalization ability and recognition accuracy of the model, effectively solving the data silo problem in the exploration field.

[0019] (2) Construct a multi-scale three-dimensional residual convolutional fusion network. Extract shallow local attribute features, middle-layer structural transition features and deep regional mineralization features by using three-dimensional convolutional kernels with different expansion coefficients. Combine the cross-layer attention weighted fusion mechanism to achieve efficient fusion of multi-scale geological features, taking into account both micro-attribute details and macro-mineralization regularity, effectively improving the richness of the expression of anomalous geological body features and the positioning accuracy of the identification boundary.

[0020] (3) By introducing a multi-level confidence threshold and connected domain analysis mechanism based on geological prior constraints, the hierarchical delineation of abnormal areas and the filtering of isolated noise are realized. At the same time, a three-dimensional visualization of the outline of the abnormal geological body and a credibility assessment report are generated. The identification results can directly support the hierarchical evaluation and engineering deployment of the exploration target area, significantly improving the engineering applicability and implementation efficiency of the intelligent identification results.

[0021] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0022] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating a machine learning-based intelligent identification method for anomalous geological bodies in mineral exploration, according to an embodiment of this application. Figure 2 This is a schematic diagram of the structure of an intelligent identification system for anomalous geological bodies in mineral exploration based on machine learning, according to an embodiment of this application. Figure 3 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0023] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0024] The following describes an intelligent identification method for anomalous geological bodies in mineral exploration based on machine learning, referring to the accompanying drawings. Addressing the issues mentioned in the background art, such as privacy barriers in cross-mining-rights exploration data, insufficient fusion of multi-scale geological features, low accuracy in identifying anomalous geological bodies, and poor engineering adaptability, this application provides an intelligent identification method for anomalous geological bodies in mineral exploration based on machine learning. In this method, hierarchical extraction and adaptive fusion of global geological features are achieved through three-dimensional voxel partitioning and multi-scale residual convolutional networks. Cross-mining-rights collaborative training is completed using an improved federated graph convolutional architecture while protecting the privacy of the original exploration data. A multi-level threshold mechanism based on geological prior constraints is used to achieve hierarchical delineation of anomalous areas and noise filtering. Finally, a three-dimensional visualized outline of the anomalous geological body and a credibility assessment result are output. This method balances data privacy compliance with model recognition accuracy and can be adapted to application scenarios of different scales of exploration areas and multiple mining-rights zones, effectively improving the automation level and engineering practical value of anomalous geological body identification. Thus, it solves the problems of data silos restricting model performance, lack of multi-scale mineralization feature fusion, lack of hierarchical constraints on identification results, and insufficient engineering applicability in traditional technologies.

[0025] Specifically, Figure 1 This is a flowchart illustrating a machine learning-based intelligent identification method for anomalous geological bodies in mineral exploration, as provided in an embodiment of this application.

[0026] like Figure 1 As shown, the intelligent identification method for anomalous geological bodies in mineral exploration based on machine learning includes the following steps: In step S101, the original data of multi-source geological exploration in the exploration area and the labeled sample dataset of known anomalous geological bodies are obtained.

[0027] Among them, multi-source geological exploration raw data refers to multi-dimensional geological exploration results data covering the entire exploration area, including but not limited to geophysical exploration data (gravity, magnetic, electrical, seismic, etc.), geochemical exploration data (elemental content, anomaly zoning, etc.), geological mapping data (strata lithology, structural occurrence, rock mass distribution, etc.), and drilling verification data.

[0028] It is understood that the embodiments of this application provide comprehensive feature support for the identification of anomalous geological bodies by integrating multi-dimensional and multi-scale original exploration data. At the same time, verified known anomalous geological bodies are used as supervised annotation samples to provide a reliable training benchmark for machine learning models, ensuring that the model identification results conform to the geological laws of metallogenesis.

[0029] In this embodiment of the application, the known anomalous geological body annotation sample dataset includes spatial annotation information of ore bodies, mineralized bodies, alteration zones and ore-controlling structures verified by drilling, trenching and other engineering projects. The annotation content includes the spatial coordinates, geological attributes and mineralization type labels of the anomalous bodies.

[0030] In step S102, the original multi-source geological exploration data is processed to construct a three-dimensional grid geological feature dataset. The three-dimensional grid geological feature dataset is then input into a feature fusion network to extract spatial multi-scale geological feature vectors.

[0031] Among them, the three-dimensional grid geological feature dataset refers to a structured dataset in which each grid cell carries spatial coordinates and multi-dimensional geological attribute features after the exploration area is divided into three-dimensional voxels.

[0032] It is understood that the embodiments of this application achieve standardized mapping of multi-source heterogeneous exploration data through a unified three-dimensional spatial grid, which solves the problem of inconsistent spatial benchmarks and dimensions of data from different exploration methods, and provides standardized input for subsequent deep learning models.

[0033] In this embodiment of the application, the multi-source geological exploration raw data is processed to construct a three-dimensional grid geological feature dataset, specifically including: A three-dimensional spatial grid partitioning model for the exploration area is constructed; based on the three-dimensional spatial grid partitioning model, the entire exploration area is divided into voxels, and the original geological exploration data from multiple sources is interpolated and mapped to each voxel grid cell; the multi-dimensional geological attribute data of each voxel grid cell are processed, and the spatial attribute labels of strata, structures, and rock masses are integrated to generate a three-dimensional grid geological feature dataset with spatial coordinates and multi-attribute features.

[0034] Voxel partitioning refers to the process of discretizing the three-dimensional space of the exploration area into several regular cubic units according to a set three-dimensional grid size. The grid size can be flexibly adjusted according to the exploration accuracy requirements and computing resource allocation.

[0035] It is understood that the embodiments of this application use spatial interpolation algorithms to map discrete exploration point data into a unified three-dimensional voxel grid, while superimposing geological semantic tags such as strata, structures, and rock masses, so that each grid cell contains both quantified geophysical and geochemical attribute values ​​and geological genetic attribute information, thereby improving the richness of feature dimensions.

[0036] In this embodiment of the application, the three-dimensional mesh geological feature dataset is input into a feature fusion network to extract spatial multi-scale geological feature vectors, specifically including: A multi-scale three-dimensional residual convolutional fusion network is constructed. Based on the multi-scale three-dimensional residual convolutional fusion network, three-dimensional convolutional kernels with different dilation coefficients are used to extract shallow local geological attribute features, middle-layer structural transition features, and deep regional macro-mineral control features of grid units in layers. The shallow local geological attribute features, middle-layer structural transition features, and deep regional macro-mineral control features are weighted and fused through a cross-layer attention fusion module to output a unified spatial multi-scale geological feature vector.

[0037] Among them, the multi-scale three-dimensional residual convolutional fusion network refers to a deep learning network equipped with a residual connection structure and containing multiple layers of three-dimensional convolutions with different receptive fields, which can simultaneously capture geological features at different spatial scales.

[0038] It is understood that the embodiments of this application take into account both microscopic local anomaly details and macroscopic regional ore-controlling laws through a hierarchical convolutional architecture. Shallow small convolutional kernels capture fine attribute features such as lithology and elemental anomalies, mid-layer dilatational convolution captures mesoscale features such as tectonic zones and alteration transition zones, and deep large receptive field convolution captures large-scale ore-controlling features such as regional strata and rock mass distribution. Combined with a cross-layer attention mechanism, the weights of features at different scales are adaptively allocated to avoid recognition bias caused by single-scale features.

[0039] Specifically, the feature extraction formula of the multi-scale three-dimensional residual convolutional fusion network is as follows:

[0040] in, This represents the multi-scale geological feature vector after voxel mesh fusion; Extracting local attribute features for shallow 3D convolution; To extract transition features for mid-layer dilated convolution; To extract regional ore-controlling features using deep, large-scale convolution; It is a cross-layer feature attention weighted fusion function that can automatically assign fusion weights according to the contribution of features at different scales to mineralization anomalies.

[0041] In step S103, the spatial multi-scale geological feature vector and the known anomalous geological body labeled sample dataset are input into the improved federated graph convolutional machine learning recognition model, and the grid cell anomaly confidence distribution map is output.

[0042] Among them, the improved federated graph convolutional machine learning recognition model refers to an improved machine learning model that combines the privacy protection mechanism of federated learning with the spatial feature aggregation capability of graph convolution, which can complete multi-partition collaborative training and anomaly recognition without sharing the original data.

[0043] It is understood that the embodiments of this application rely on the federated learning architecture to solve the problem of cross-mining rights data silos, while using graph convolutional networks to adapt to the spatial correlation characteristics of geological bodies, thereby improving the spatial continuity and boundary accuracy of anomaly identification.

[0044] In this embodiment, spatial multi-scale geological feature vectors and a dataset of labeled known anomalous geological bodies are input into an improved federated graph convolutional machine learning recognition model to output a grid cell anomaly confidence distribution map, specifically including: An improved federated graph convolutional recognition model is constructed, wherein the improved federated graph convolutional recognition model is divided into a local mining rights partitioning graph convolutional sub-module and a cloud-based federated aggregation module. Using each 3D grid cell as a graph node, an adjacency graph matrix is ​​constructed based on the grid spatial adjacency relationship. Spatial multi-scale geological feature vectors are used as node features, and the known anomalous geological body labeled sample dataset is used as local supervision labels. The local mining rights partition map convolution submodule is input, and the partition blocks complete local anomaly classification training. At the same time, the model gradient parameters are uploaded to the cloud federated aggregation module to perform global parameter updates on the improved federated graph convolution recognition model without transmitting the original geological exploration data and labeled samples. The updated improved federated graph convolution model is distributed to each mining rights partition, inference is performed on all voxel grids within the partition, the probability value of mineralization anomaly corresponding to each voxel grid is calculated, and a global grid cell anomaly confidence distribution map is generated.

[0045] Among them, the adjacency graph matrix refers to the graph structure adjacency matrix constructed based on the spatial adjacency relationship of three-dimensional grid cells. Each grid cell has connecting edges with the surrounding grid cells that are adjacent to 6 faces, 12 edges, and 8 vertices. The adjacency range can be adjusted according to the geological spatial interaction law.

[0046] It is understood that the embodiments of this application adopt a federated learning paradigm of local training-cloud aggregation-inference distribution. Each mining rights partition only uploads the model gradient parameters, while the original exploration data and labeled samples are all stored locally. This not only ensures the privacy and ownership security of mining rights data, but also makes full use of the full-domain samples to improve the model's generalization ability and recognition accuracy, effectively solving the data silo problem in the exploration field.

[0047] Specifically, the improved federated graph convolutional node feature update formula is as follows:

[0048] in, For the first Features of layer grid node v after update; is the activation function; W is the node feature transformation weight; For node v, it is the set of adjacent meshes in space; The spatial association attention coefficient between node v and its neighboring node u; This is the spatial adjacency edge weight matrix; Let u be the feature vector of the l-th layer adjacent node; For grid node v Layer feature vectors.

[0049] It is understood that the embodiments of this application introduce spatial correlation attention coefficients and independent edge weight matrices in the graph convolution aggregation process, which breaks through the limitation of traditional graph convolution that equally weights all adjacent nodes. It can adaptively distinguish the mineralization contribution of adjacent grids in different directions and different geological interfaces, better fit the spatial non-uniformity law of mineralization, and improve the identification accuracy of abnormal boundaries.

[0050] In step S104, based on the anomaly confidence distribution map and combined with the geological prior constraint threshold, multi-level anomaly regions are divided, connected anomaly blocks are extracted as candidate anomaly geological bodies, and three-dimensional visualized anomaly geological body outline annotations are generated.

[0051] Among them, the anomaly confidence distribution map refers to the probability distribution map of mineralization anomalies covering the entire exploration area with a three-dimensional grid. The value of each voxel grid ranges from 0 to 1. The higher the value, the greater the probability that the grid belongs to an anomalous geological body.

[0052] It is understandable that the embodiments of this application combine prior geological knowledge to set hierarchical thresholds, replacing the binary division method of a single threshold. This is more in line with the engineering logic of gradually delineating target areas in mineral exploration. At the same time, isolated noise points are eliminated through connected component analysis, thereby improving the geological rationality of the identification results.

[0053] In this embodiment of the application, multi-level anomaly regions are divided based on the anomaly confidence distribution map combined with geological prior constraint thresholds, specifically including: Set three levels of anomaly confidence thresholds: First threshold Second threshold and the third threshold ,in ; Anomaly confidence level greater than or equal to The grid cells are divided into first-level anomaly regions, and anomaly confidence levels greater than or equal to are classified as such. and less than The grid cells are divided into secondary anomaly regions, and anomaly confidence levels greater than or equal to are classified as follows: and less than The grid cells are divided into three levels of anomaly regions; Connectivity analysis is performed on the connected grid cells within each level of anomaly region to extract spatially continuous connected anomaly blocks, and isolated noisy grid cells are filtered out to obtain a set of candidate anomaly geological bodies.

[0054] Among them, the confidence threshold of the three-level anomaly is set based on the known mineralization regularity and geological prior knowledge of the exploration area. The first-level anomaly area corresponds to the high-confidence mineralization target area, the second-level anomaly area corresponds to the mineralization potential area, and the third-level anomaly area corresponds to the mineralization prospective area.

[0055] It is understood that the embodiments of this application achieve gradient delineation of abnormal areas through hierarchical thresholds, which can directly support the hierarchical deployment of exploration projects; at the same time, spatial connectivity analysis is used to eliminate scattered and isolated noise grids, retaining spatially continuous abnormal blocks, making the identification results more consistent with the objective laws of the spatial continuous distribution of geological bodies, and reducing the interference of false anomalies.

[0056] Based on the extracted set of candidate anomalous geological bodies, a three-dimensional surface reconstruction algorithm is used to generate a closed contour surface for each anomalous block. Spatial coordinates and geological attribute annotations are then superimposed to form a three-dimensional visualized anomalous geological body contour annotation result.

[0057] In step S105, based on the outline annotation of the abnormal geological body and the feedback data from manual verification, the results of the abnormal geological body identification and the credibility assessment report are output.

[0058] Among them, the manual verification feedback data refers to the correction, annotation and evaluation data fed back by geological engineers after manually interpreting and engineering verifying the identified abnormal geological bodies.

[0059] It is understood that the embodiments of this application introduce a manual verification and feedback mechanism to form a closed-loop optimization process of intelligent identification, manual review and model iteration, which not only gives full play to the efficiency of machine learning, but also retains the experience and guidance of geological experts, and continuously improves the identification accuracy and geological rationality of the model.

[0060] In this embodiment, the credibility assessment report includes the spatial location, scale, classification, and attribute characteristics of the anomalous geological body, as well as quantitative assessment indicators such as the degree of overlap with known anomalous bodies, feature matching degree, and engineering verification suggestions, which can directly provide decision support for the selection of exploration target areas and engineering deployment.

[0061] The intelligent identification method for anomalous geological bodies in mineral exploration based on machine learning proposed in this application achieves standardized fusion and hierarchical feature extraction of multi-source geological data through 3D voxel subdivision and multi-scale residual convolutional networks, taking into account both local fine attributes and regional ore-controlling patterns. Relying on an improved federated graph convolutional architecture, cross-mining right collaborative training is completed while ensuring data privacy, improving model generalization ability and anomaly boundary identification accuracy. Combining multi-level thresholds and connected component filtering mechanisms based on geological priors, hierarchical delineation of anomalous areas and noise filtering are achieved, outputting 3D visualization results and a credibility assessment report that can directly support engineering deployment. This method is adaptable to complex exploration scenarios involving multiple mining rights partitions and multi-source data fusion, with identification efficiency several times higher than traditional manual interpretation, while also possessing good privacy compliance and engineering practicality. Therefore, it solves the problems of data silos restricting model performance, lack of multi-scale ore-forming feature fusion, lack of hierarchical constraints on identification results, and insufficient engineering applicability in traditional technologies.

[0062] The following is a specific example illustrating the intelligent identification method for anomalous geological bodies in mineral exploration based on machine learning. This example is applied to a cross-mining-rights anomalous geological body identification project in a copper-gold mining cluster. The mining cluster covers three independent mining-rights zones, with a total area of ​​approximately 120 km² and an exploration depth of 2000 m underground. Specifically, it includes: Step 1: Multi-source exploration data acquisition and annotation sample construction.

[0063] The study integrates multi-source exploration raw data from the mining area, including gravity, magnetic, induced polarization sounding, 1:10,000 geological mapping, soil geochemical measurements, and verification data from over 120 drilling sites. The geophysical data has a grid precision of 50m×50m, and the geochemical data has a sampling density of 16 points / km². Twenty-eight verified copper-gold orebodies and mineralization alteration zones were selected as known anomaly annotation samples and divided into training and test sets at a ratio of 7:3. The annotation content includes the three-dimensional spatial range of the anomaly, the mineralization type, and the ore-controlling structural attributes.

[0064] Step 2: Construction of 3D voxel mesh and multi-scale feature extraction.

[0065] A three-dimensional spatial coordinate system was established for the mineralized area. A 50m×50m×50m cubic grid was used to subdivide the entire area into voxel units, generating approximately 1.92 million three-dimensional voxel units. The discrete geophysical and geochemical data and geological attributes were mapped to each voxel unit using the Kriging interpolation algorithm. Semantic labels such as stratigraphy, lithology, fault structures, and intrusive bodies were integrated to construct a three-dimensional gridded geological feature dataset. A multi-scale three-dimensional residual convolutional fusion network was built. A 3×3×3 conventional convolution kernel was used in the shallow layer to extract local lithology and elemental anomaly features. A 3×3×3 dilatational convolution with a dilation coefficient of 2 was used in the middle layer to extract structural transition zone features. A 3×3×3 dilatational convolution with a dilation coefficient of 4 was used in the deep layer to extract regional ore-controlling pattern features. The three layers of features were weighted and fused using a cross-layer attention module, outputting a 128-dimensional spatial multi-scale geological feature vector. The average single-grid computation latency for feature extraction was less than 0.02ms.

[0066] Step 3: Improve the collaborative training of federated graph convolutional models.

[0067] A federated learning architecture consisting of one cloud aggregation node and three local mining rights nodes was constructed, with each mining rights node independently deploying its local graph convolutional submodule. Each voxel grid is used as a graph node, and an adjacency graph matrix is ​​constructed based on six-face adjacency relationships. 128-dimensional multi-scale features are used as node inputs, and known anomaly labels are used as supervision labels. Each mining rights node independently conducts local training with a batch size of 32 and a learning rate of 0.001. Every five training epochs, the model gradient parameters are encrypted and uploaded to the cloud aggregation module. The FedAvg algorithm is used to aggregate global parameters before distributing them to each local node. No original exploration data is transmitted throughout the process. The model converges after 50 iterations. The graph convolutional node aggregation process is improved by introducing spatial attention coefficients and edge weight matrices, assigning higher weights to adjacent grids on both sides of the fault zone, thereby enhancing the sensitivity of identifying ore-controlling anomalies.

[0068] Step 4: Multi-level anomaly region division and 3D contour generation.

[0069] Three levels of confidence thresholds were set based on the metallogenic regularity of the ore-bearing area: =0.3、 =0.6、 =0.8, the results of the global grid inference are classified into three levels: Level 1 anomaly areas correspond to high-confidence mineralization target areas, Level 2 corresponds to mineral exploration potential areas, and Level 3 corresponds to mineralization prospective areas. Three-dimensional connectivity analysis is performed on each level of anomaly area, and isolated noise blocks with fewer than 10 grid cells are removed, resulting in 21 connected anomaly blocks as candidate anomaly geological bodies. The moving cube algorithm is used to reconstruct the three-dimensional surface of each anomaly block, generating closed three-dimensional contour annotations with hierarchical attributes, with a contour spatial positioning error of less than 1 grid cell.

[0070] Step 5: Output and Iterative Optimization.

[0071] Geological engineers manually interpreted and verified 21 candidate anomalous geological bodies. Based on existing drilling data and mineralization patterns, they corrected three misjudged anomalies and supplemented two missed weak anomalies. The verification feedback data was added to the training sample set to fine-tune and optimize the model. Finally, they output an atlas of anomalous geological body identification results for the mineralized area and a credibility assessment report. Validated on the test set, the model in this embodiment achieved a 92.3% accuracy rate in identifying first-level anomalies, and its overall identification efficiency was approximately six times higher than traditional manual interpretation, directly supporting subsequent exploration engineering deployments.

[0072] In summary, the embodiments of this application construct a unified feature benchmark by integrating multi-source exploration data and standardizing voxel mapping, achieve adaptive fusion of hierarchical mineralization features by relying on multi-scale residual convolutional networks, complete cross-mining rights collaborative training by adopting an improved federated graph convolutional architecture under data privacy protection, and output engineered identification results by combining geological prior hierarchical thresholds and connected component screening, thus forming a full-process intelligent identification system covering data processing, feature extraction, model training, anomaly delineation, and result output.

[0073] Next, referring to the accompanying drawings, we describe the intelligent identification system for anomalous geological bodies in mineral exploration based on machine learning, according to an embodiment of this application.

[0074] Figure 2 This is a block diagram of an intelligent identification system for anomalous geological bodies in mineral exploration based on machine learning, according to an embodiment of this application.

[0075] like Figure 2 As shown, the intelligent identification system 10 for mineral exploration anomalies based on machine learning includes: an acquisition module 100, an extraction module 200, an identification module 300, a division module 400, and an output module 500.

[0076] Among them, the acquisition module 100 is used to acquire the original data of multi-source geological exploration in the exploration area and the labeled sample dataset of known anomalous geological bodies; The extraction module 200 is used to process the multi-source geological exploration raw data, construct a three-dimensional grid geological feature dataset, input the three-dimensional grid geological feature dataset into a feature fusion network, and extract spatial multi-scale geological feature vectors. The recognition module 300 is used to input the spatial multi-scale geological feature vector and the known anomalous geological body labeled sample dataset into the improved federated graph convolutional machine learning recognition model, and output the grid cell anomaly confidence distribution map; The segmentation module 400 is used to segment multi-level abnormal regions based on the anomaly confidence distribution map and geological prior constraint threshold, extract connected abnormal blocks as candidate abnormal geological bodies, and generate three-dimensional visualized anomaly geological body outline annotations. The output module 500 is used to output the abnormal geological body identification results and credibility assessment report based on the outline annotation of the abnormal geological body and the feedback data of manual verification.

[0077] It should be noted that the foregoing explanation of an embodiment of a machine learning-based intelligent identification method for mineral exploration anomalies also applies to a machine learning-based intelligent identification system for mineral exploration anomalies in this embodiment, and will not be repeated here.

[0078] This application proposes a machine learning-based intelligent identification system for anomalous geological bodies in mineral exploration. It constructs a unified feature dataset through multi-source data integration and 3D voxel standardization. Relying on a multi-scale residual convolutional network, it achieves hierarchical feature extraction and adaptive fusion. Utilizing an improved federated graph convolutional architecture, it completes cross-mining right collaborative identification while protecting data privacy. Combined with a geological priori hierarchical mechanism, it outputs engineered anomaly identification results and evaluation reports. This system is adaptable to complex exploration scenarios involving multiple mining rights and multiple scales, balancing identification efficiency, accuracy, and data security, providing reliable technical support for the intelligent upgrading of mineral exploration. Therefore, it solves the problems of data silos restricting model performance, lack of multi-scale mineralization feature fusion, lack of hierarchical constraints on identification results, and insufficient engineering applicability in traditional technologies.

[0079] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 301, the processor 302, and the computer program stored on the memory 301 and capable of running on the processor 302.

[0080] When the processor 302 executes the program, it implements the intelligent identification method for anomalous geological bodies in mineral exploration based on machine learning provided in the above embodiments.

[0081] Furthermore, electronic devices also include: Communication interface 303 is used for communication between memory 301 and processor 302.

[0082] The memory 301 is used to store computer programs that can run on the processor 302.

[0083] The memory 301 may include high-speed RAM (Random Access Memory) and may also include non-volatile memory, such as at least one disk storage.

[0084] If the memory 301, processor 302, and communication interface 303 are implemented independently, then the communication interface 303, memory 301, and processor 302 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0085] Optionally, in a specific implementation, if the memory 301, processor 302, and communication interface 303 are integrated on a single chip, then the memory 301, processor 302, and communication interface 303 can communicate with each other through an internal interface.

[0086] Processor 302 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement embodiments of this application.

[0087] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described machine learning-based intelligent identification method for mineral exploration anomalies.

[0088] Furthermore, this application also provides a computer program product, including a computer program or instructions, which, when executed, implement the above-mentioned intelligent identification method for mineral exploration anomalies based on machine learning.

[0089] In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0090] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0091] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0092] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0093] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium. When executed, the program includes one or a combination of the steps of the method embodiments.

[0094] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A machine learning-based intelligent identification method for anomalous geological bodies in mineral exploration, characterized in that, include: Acquire raw geological exploration data from multiple sources in the exploration area and a dataset of labeled known anomalous geological bodies. The multi-source geological exploration raw data are processed to construct a three-dimensional grid geological feature dataset. The three-dimensional grid geological feature dataset is then input into a feature fusion network to extract spatial multi-scale geological feature vectors. The spatial multi-scale geological feature vectors and the labeled sample dataset of known anomalous geological bodies are input into the improved federated graph convolutional machine learning recognition model, and the output is a grid cell anomaly confidence distribution map. Based on the anomaly confidence distribution map and combined with geological prior constraint thresholds, multi-level anomaly regions are divided, connected anomaly blocks are extracted as candidate anomaly geological bodies, and three-dimensional visualization of anomaly geological body outline annotations are generated. Based on the outline annotation of the anomalous geological bodies and the feedback data from manual verification, the results of the anomalous geological body identification and a credibility assessment report are output.

2. The method according to claim 1, characterized in that, The improved federated graph convolutional machine learning recognition model inputs spatial multi-scale geological feature vectors and a dataset of labeled known anomalous geological bodies, and outputs a grid cell anomaly confidence distribution map, including: An improved federated graph convolutional recognition model is constructed, wherein the improved federated graph convolutional recognition model is divided into a local mining rights partitioning graph convolutional sub-module and a cloud-based federated aggregation module; Using each 3D grid cell as a graph node, an adjacency graph matrix is ​​constructed based on the grid spatial adjacency relationship. Spatial multi-scale geological feature vectors are used as node features, and the known anomalous geological body labeled sample dataset is used as local supervision labels. The local mining rights partition map convolution submodule is input, and the partition blocks complete local anomaly classification training. At the same time, the model gradient parameters are uploaded to the cloud federated aggregation module to perform global parameter updates on the improved federated graph convolution recognition model without transmitting the original geological exploration data and labeled samples. The updated improved federated graph convolution model is distributed to each mining rights partition, inference is performed on all voxel grids within the partition, the probability value of mineralization anomaly corresponding to each voxel grid is calculated, and a global grid cell anomaly confidence distribution map is generated.

3. The method according to claim 1, characterized in that, The improved federated graph convolutional node feature update formula is as follows: in, For the first Features of layer grid node v after update; is the activation function; W is the node feature transformation weight; For node v, it is the set of adjacent meshes in space; The spatial association attention coefficient between node v and its neighboring node u; This is the spatial adjacency edge weight matrix; Let u be the feature vector of the l-th layer adjacent node; For grid node v Layer feature vectors.

4. The method according to claim 1, characterized in that, The process of dividing multi-level anomaly regions based on the anomaly confidence distribution map combined with geological prior constraint thresholds includes: Set three levels of anomaly confidence thresholds: First threshold Second threshold and the third threshold ,in ; Anomaly confidence level greater than or equal to The grid cells are divided into first-level anomaly regions, and anomaly confidence levels greater than or equal to are classified as such. and less than The grid cells are divided into secondary anomaly regions, and anomaly confidence levels greater than or equal to are classified as follows: and less than The grid cells are divided into three levels of anomaly regions; Connectivity analysis is performed on the connected grid cells within each level of anomaly region to extract spatially continuous connected anomaly blocks, and isolated noisy grid cells are filtered out to obtain a set of candidate anomaly geological bodies.

5. The method according to claim 1, characterized in that, The feature extraction formula of the multi-scale three-dimensional residual convolutional fusion network is as follows: in, This represents the multi-scale geological feature vector after voxel mesh fusion; Extracting local attribute features for shallow 3D convolution; To extract transition features for mid-layer dilated convolution; To extract regional ore-controlling features using deep, large-scale convolution; This is a cross-layer feature attention weighted fusion function.

6. The method according to claim 1, characterized in that, The multi-source geological exploration raw data are processed to construct a three-dimensional grid geological feature dataset, including: Construct a three-dimensional spatial grid division model for the exploration area; Based on the three-dimensional spatial grid partitioning model, the entire exploration area is divided into voxels, and the original geological exploration data from multiple sources is interpolated and mapped to each voxel grid cell. The multi-dimensional geological attribute data of each voxel grid cell are processed and the spatial attribute labels of strata, structures and rock masses are integrated to generate a three-dimensional grid geological feature dataset with spatial coordinates and multi-attribute features.

7. The method according to claim 1, characterized in that, The three-dimensional mesh geological feature dataset is input into a feature fusion network to extract spatial multi-scale geological feature vectors, including: Construct a multi-scale 3D residual convolutional fusion network; Based on the multi-scale three-dimensional residual convolutional fusion network, three-dimensional convolutional kernels with different expansion coefficients are used to extract shallow local geological attribute features, middle-layer structural transition features, and deep regional macro-mineral control features of grid units in layers. The shallow local geological attribute features, the middle structural transition features, and the deep regional macro-mineral control features are weighted and fused by the cross-layer attention fusion module to output a unified spatial multi-scale geological feature vector.

8. A machine learning-based intelligent identification system for anomalous geological bodies in mineral exploration, characterized in that, include: The acquisition module is used to acquire raw data from multi-source geological exploration in the exploration area and a dataset of labeled samples of known anomalous geological bodies. The extraction module is used to process the multi-source geological exploration raw data, construct a three-dimensional grid geological feature dataset, input the three-dimensional grid geological feature dataset into a feature fusion network, and extract spatial multi-scale geological feature vectors. The identification module is used to input the spatial multi-scale geological feature vectors and the labeled sample dataset of known anomalous geological bodies into the improved federated graph convolutional machine learning identification model, and output the grid cell anomaly confidence distribution map. The segmentation module is used to segment multi-level abnormal regions based on the anomaly confidence distribution map and geological prior constraint thresholds, extract connected abnormal blocks as candidate abnormal geological bodies, and generate three-dimensional visualized anomaly geological body outline annotations. The output module is used to output the results of the identification of the abnormal geological bodies and the credibility assessment report based on the outline annotation of the abnormal geological bodies and the feedback data of manual verification.

9. An electronic device, characterized in that, The invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the machine learning-based intelligent identification method for mineral exploration anomalies as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When a computer program or instruction is executed, it implements the intelligent identification method for mineral exploration anomalies based on machine learning as described in any one of claims 1-7.