Metallogenic prediction method and device based on neural network, equipment and storage medium

Abstract

The present disclosure provides a mineralization prediction method and device based on a neural network, equipment and a storage medium. The method comprises: acquiring a multi-modal geological image of a target area; using a pre-trained mineralization prediction model to determine one-dimensional multi-modal data corresponding to a center point and two-dimensional multi-modal data of a neighborhood of the center point from the multi-modal geological image; extracting local attribute features from the one-dimensional multi-modal data and spatial context features from the two-dimensional multi-modal data; performing adaptive fusion processing on the local attribute features and the spatial context features to obtain target fusion features; determining a mineralization probability value of the center point based on the target fusion features; and determining a mineralization prediction result of the target area according to the mineralization probability values of all center points in the target area combined with an integration strategy. In this way, the mineralization possibility of each center point can be more accurately evaluated, and the accuracy of the mineralization prediction result of the target area is improved.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to a method, apparatus, device and storage medium for mineralization prediction based on neural networks. Background Technology

[0002] With continuous economic development, the demand for mineral resources is increasing, and regional metallogenic prediction is a crucial part of geological prospecting. However, directly predicting whether a specific location is a mineral deposit can lead to inaccurate predictions for the entire region.

[0003] In view of this, how to improve the accuracy of mineralization prediction results in the region has become an urgent technical problem to be solved. Summary of the Invention

[0004] In view of this, the purpose of this disclosure is to propose a method, apparatus, device and storage medium for mineralization prediction based on neural networks to solve or partially solve the above-mentioned technical problems.

[0005] To achieve the above objectives, the first aspect of this disclosure proposes a neural network-based mineralization prediction method, the method comprising:

[0006] Acquire multimodal geological images of the target area;

[0007] Using a pre-trained mineralization prediction model, one-dimensional multimodal data corresponding to the center point and two-dimensional multimodal data of the neighborhood of the center point are determined from the multimodal geological image.

[0008] Local attribute features are extracted from the one-dimensional multimodal data, and spatial context features are extracted from the two-dimensional multimodal data;

[0009] The target fusion feature is obtained by fusing the local attribute features and the spatial context features.

[0010] Based on the target fusion features, the mineralization probability value of the center point is determined, and the mineralization prediction result of the target area is determined based on the mineralization probability values ​​of all center points in the target area.

[0011] Based on the same inventive concept, a second aspect of this disclosure proposes a neural network-based mineralization prediction device, comprising:

[0012] The acquisition module is configured to acquire multimodal geological images of the target area;

[0013] The geological data determination module is configured to use a pre-trained mineralization prediction model to determine the one-dimensional multimodal data corresponding to the center point and the two-dimensional multimodal data of the neighborhood of the center point from the multimodal geological image.

[0014] The feature extraction module is configured to extract local attribute features from the one-dimensional multimodal data and spatial context features from the two-dimensional multimodal data.

[0015] The feature fusion module is configured to fuse the local attribute features and the spatial context features to obtain the target fused features;

[0016] The mineralization prediction module is configured to determine the mineralization probability value of the center point based on the target fusion features, and to determine the mineralization prediction result of the target area based on the mineralization probability values ​​of all center points in the target area.

[0017] Based on the same inventive concept, a third aspect of this disclosure proposes an electronic device including a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor implements the method described above when executing the computer program.

[0018] Based on the same inventive concept, a fourth aspect of this disclosure provides a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to perform the methods described above.

[0019] As can be seen from the above, the present disclosure provides a neural network-based mineralization prediction method, apparatus, equipment, and storage medium. The method involves acquiring multimodal geological images of the target area. Using a pre-trained mineralization prediction model, one-dimensional multimodal data corresponding to the center point and two-dimensional multimodal data of the center point's neighborhood are determined from the multimodal geological images. The one-dimensional multimodal data can cover detailed local geological attributes corresponding to the center point, while the two-dimensional multimodal data can cover a wider range of spatial geological information in the center point's neighborhood. Both the one-dimensional and two-dimensional multimodal data have rich data dimensions, enabling accurate mineralization prediction. Local attribute features are extracted from the one-dimensional multimodal data, and spatial context features are extracted from the two-dimensional multimodal data. The local attribute features accurately capture the geological attribute details of the center point location, while the spatial context features fully consider the geological spatial relationships of the surrounding neighborhood. The local attribute features and spatial context features are fused to obtain target fused features. These target fused features contain detailed local information and consider overall spatial relationships, thus reflecting the relationship between geological data and mineralization more comprehensively and accurately. The mineralization probability value of the center point is determined based on the target fusion characteristics, and the mineralization prediction result of the target area is determined based on the mineralization probability values ​​of all center points in the target area. In this way, the mineralization potential of each center point can be assessed more accurately, and the accuracy of the mineralization prediction result of the target area can be improved. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this disclosure or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a neural network-based mineralization prediction method according to an embodiment of the present disclosure;

[0022] Figure 2 This is a schematic diagram of an attribute-space convolutional neural network according to an embodiment of the present disclosure;

[0023] Figure 3 This is a schematic diagram of a mineralization prediction model according to an embodiment of the present disclosure;

[0024] Figure 4 This is a flowchart of a regional mineralization prediction method based on attribute-spatial convolutional neural networks according to an embodiment of the present disclosure;

[0025] Figure 5 This is a schematic diagram of the structure of a neural network-based mineralization prediction device according to an embodiment of the present disclosure;

[0026] Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0028] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this disclosure should have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms "first," "second," and similar terms used in the embodiments of this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0029] As mentioned above, how to improve the accuracy of mineralization prediction results within a region has become an important research question.

[0030] Based on the above description, such as Figure 1 As shown in this embodiment, the mineralization prediction method based on neural networks includes:

[0031] Step 101: Obtain multimodal geological images of the target area.

[0032] In practice, the target area is the region where mineralization prediction needs to be performed. Multimodal geological images can be multimodal geographic-physical-chemical-remote sensing images.

[0033] For example, if the target area is a high vegetation cover area, collect 22 multimodal geological images at a scale of 1:250,000 covering the target area. These multimodal geological images can be in MapGIS format. Collect aeromagnetic data with a spatial resolution of 500m covering the target area. These aeromagnetic data can be in .GRD format. Collect stream sediment measurement data covering the target area, including latitude and longitude and 39 elements or oxides. These stream sediment measurement data are stored in .xlsx files. Collect 11 Landsat images covering the target area, including Landsat 4, 5, 7, and 9 data. Collect gold mineralization data. These gold mineralization data are in .shp format.

[0034] Twenty-two 1:250,000 multimodal geological images in MapGIS format were converted from .wp, .wl, and .wt files to .shp files. At the junctions of multiple multimodal geological images, identical lithological units were merged and lithology was unified. Lithology and geological age attributes were then sequentially linked, and the unified data were encoded. Vector-to-raster conversion yielded lithology and geological age raster data with an EPSG:3857 coordinate system and a spatial resolution of 100m. Fault lines were extracted, and after buffer stacking, fault raster data with an EPSG:3857 coordinate system and a spatial resolution of 100m was obtained. Aeromagnetic data with an EPSG:3857 coordinate system and a spatial resolution of 100m was obtained through format conversion, reprojection, and resampling. Data cleaning was then performed. The data was washed, and two elements with large data gaps were removed and missing values ​​were filled in. For each element or oxide, raster data of that element or oxide was obtained by converting latitude and longitude to point vectors and interpolating the point vector space. This process was repeated for 37 elements, resulting in 37 raster data with EPSG:3857 coordinate system and 100m spatial resolution. After radiometric calibration, atmospheric correction, orthorectification, and georegistration of 11 remote sensing images, a total of 6 bands (red, green, blue, near-infrared, shortwave infrared 1, and shortwave infrared 2) were extracted from each image. Image mosaicking was then performed to generate a six-band remote sensing image covering the target area. The gold mine point data was reprojected and converted into a .shp file in the EPSG:3857 coordinate system.

[0035] Using vector data of the target area, lithology, geological age, faults, aeromagnetic data, 37 elements or oxides, and 6-band remote sensing images are cropped sequentially. The cropped raster data are then resampled sequentially to obtain raster data with the same number of rows and columns. After stacking, a mineralization prediction dataset is obtained, which has a total of 47 channels.

[0036] Step 102: Using a pre-trained mineralization prediction model, determine the one-dimensional multimodal data corresponding to the center point and the two-dimensional multimodal data of the center point's neighborhood from the multimodal geological image.

[0037] In practice, the center point is the location point for mineralization prediction. The center point can be a mineralized point or a non-mineralized point.

[0038] Mineralization prediction models can be obtained by training an Attribute Spatial Convolutional Neural Network (ASCNN). The mineralization prediction model includes: a first extraction module (local attribute information extraction module), a second extraction module (neighborhood spatial information extraction module), a one-dimensional convolutional layer (attribute convolution module), a two-dimensional convolutional layer (spatial convolution module), an adaptive feature fusion module, and a classification prediction module.

[0039] Using the first extraction module (local attribute information extraction module), multiple first geological parameters corresponding to the center point are determined from the multimodal geological image, and the multiple first geological parameters are converted into one-dimensional multimodal data corresponding to the center point.

[0040] Using the second extraction module (neighborhood spatial information extraction module), various second geological parameters of the neighborhood of the center point are determined from the multimodal geological image, and the various second geological parameters are converted into two-dimensional multimodal data of the neighborhood of the center point.

[0041] Step 103: Extract local attribute features from the one-dimensional multimodal data and extract spatial context features from the two-dimensional multimodal data.

[0042] In practice, Figure 2 This is a schematic diagram of an attribute-spatial convolutional neural network according to an embodiment of this disclosure. Figure 2As shown, multimodal geological images can be derived from multi-source geological data. A one-dimensional convolutional layer (attribute convolution module) is used to extract local attribute features from the one-dimensional multimodal data corresponding to the center point. These local attribute features can be high-dimensional attributes ranging from 64 to 128 to 256 to 512. A two-dimensional convolutional layer (spatial convolution module) is then used to extract spatial context features from the two-dimensional multimodal data. These spatial context features can be high-dimensional spatial embeddings ranging from 12×12×64 to 6×6×128 to 3×3×256 to 1×1×512.

[0043] Step 104: Perform adaptive fusion processing on the local attribute features and the spatial context features to obtain the target fusion features.

[0044] In specific implementation, such as Figure 2 As shown, an adaptive feature fusion module is used to fuse local attribute features and spatial context features through a fully connected Softmax layer to obtain the target fused feature, which can be a 1024×1×1 feature. In this way, by fusing local attribute features and spatial context features, the complementarity of the two features can be enhanced.

[0045] Step 105: Determine the mineralization probability value of the center point based on the target fusion features, and determine the mineralization prediction result of the target area based on the mineralization probability values ​​of all center points in the target area combined with the integration strategy.

[0046] In practice, based on the target fusion features, a classification prediction result is obtained to determine whether the center point is a mineral deposit. This classification prediction result is then converted into a mineralization probability value for the center point. A sliding window is used to determine the mineralization probability values ​​of all center points in the target area, and a mineralization probability map of the target area is determined based on these values. Finally, an ensemble strategy is used to weight and process multiple mineralization probability maps output by multiple mineralization prediction models to obtain the mineralization prediction result for the target area.

[0047] To address the challenges of small sample sizes, class imbalance, and noise interference, this disclosure employs a sample cross-partitioning and multi-model ensemble strategy based on the Bagging concept. This strategy performs pixel-level weighted averaging of the independent predictions from multiple models, significantly improving the robustness and stability of the ensemble model. This disclosure is innovative in both its network structure and ensemble strategy, significantly enhancing model robustness and prediction accuracy, ensuring stable and reliable mineralization prediction results, and exhibiting strong regional applicability and mineral type universality. It is suitable for identifying and delineating favorable mineralization areas for different deposit types.

[0048] Through the above embodiments, multimodal geological images of the target area are obtained. Using a pre-trained mineralization prediction model, one-dimensional multimodal data corresponding to the center point and two-dimensional multimodal data of the center point's neighborhood are determined from the multimodal geological images. The one-dimensional multimodal data can cover detailed local geological attributes corresponding to the center point, while the two-dimensional multimodal data can cover a wider range of spatial geological information in the center point's neighborhood. Both one-dimensional and two-dimensional multimodal data have rich data dimensions, enabling accurate mineralization prediction. Local attribute features are extracted from the one-dimensional multimodal data, and spatial context features are extracted from the two-dimensional multimodal data. Local attribute features accurately capture the geological attribute details of the center point location, while spatial context features fully consider the geological spatial relationships of the surrounding neighborhood. The local attribute features and spatial context features are fused to obtain target fusion features. Target fusion features contain detailed local information and consider overall spatial relationships, more comprehensively and accurately reflecting the relationship between geological data and mineralization. The mineralization probability value of the center point is determined based on the target fusion features, and the mineralization prediction result of the target area is determined based on the mineralization probability values ​​of all center points in the target area. This allows for a more accurate assessment of the mineralization potential of each central point, improving the accuracy of mineralization prediction results for the target area.

[0049] In some embodiments, step 102 includes:

[0050] Step 1021: Determine the center point for mineralization prediction from the target area.

[0051] In practice, the center point is the location point for mineralization prediction. Specifically, mineralized points and non-mineralized points are used as the center points.

[0052] Step 1022: Using the first extraction module in the pre-trained mineralization prediction model, determine multiple first geological parameters corresponding to the center point from the multimodal geological image, and convert the multiple first geological parameters into one-dimensional multimodal data corresponding to the center point.

[0053] In specific implementation, the first extraction module can be a local attribute information extraction module. This module extracts multiple primary geological parameters corresponding to the center point from the multimodal geological image. These parameters include: lithology, structure, geological age, aeromagnetic properties, geoelectricity, stream sediments, geophysical properties, geochemical properties, and remote sensing spectral properties corresponding to the center point. A one-dimensional vector (c) is extracted from these parameters and used as the one-dimensional multimodal data corresponding to the center point. Here, c represents the number of data types or dimensions in the mineralization prediction dataset, corresponding to the attribute information of mineralized and non-mineralized points.

[0054] Step 1023: Using the second extraction module in the pre-trained mineralization prediction model, determine multiple second geological parameters of the neighborhood of the center point from the multimodal geological image, and convert the multiple second geological parameters into two-dimensional multimodal data of the neighborhood of the center point.

[0055] In specific implementation, the second extraction module can be a neighborhood spatial information extraction module. This module extracts various second geological parameters from the neighborhood of the center point in the multimodal geological image. These parameters include lithology, structure, geological age, aeromagnetic properties, geoelectricity, stream sediments, geophysical properties, geochemical properties, and remote sensing spectral properties within a n-pixel neighborhood of the center point. These second geological parameters are then combined to form a spatial feature matrix, which can be an n×n×c multi-source spatial feature matrix (height×width×channels, where n is an odd number). This spatial feature matrix serves as the two-dimensional multimodal data of the center point's neighborhood. The neighborhood range n of the mineralization point covers both the ore-forming body and the non-mineralized background, and the channel size c has the same dimension as the regional mineralization prediction dataset.

[0056] The above scheme determines the center point for mineralization prediction within the target area. Using the first extraction module of a pre-trained mineralization prediction model, multiple first geological parameters corresponding to the center point are determined from the multimodal geological image, and these parameters are converted into one-dimensional multimodal data corresponding to the center point. Using the second extraction module of the same model, multiple second geological parameters in the neighborhood of the center point are determined from the multimodal geological image, and these parameters are converted into two-dimensional multimodal data of the center point's neighborhood. Thus, the one-dimensional multimodal data can cover the detailed local geological attributes corresponding to the center point, while the two-dimensional multimodal data can cover a wider range of spatial geological information in the center point's neighborhood. Both the one-dimensional and two-dimensional multimodal data possess rich data dimensions, enabling accurate mineralization prediction.

[0057] In some embodiments, step 103 includes:

[0058] Step 1031: Using a one-dimensional convolutional layer in a pre-trained mineralization prediction model, extract the first geological feature from the one-dimensional multimodal data, and enhance the first geological feature through a channel attention module to obtain local attribute features.

[0059] In practical implementation, the one-dimensional convolutional layer can be an attribute convolutional module (Attribute CNN), used to mine attribute features of both mineral and non-mineral points. The one-dimensional convolutional layer can be at least three layers of a 1D CNN, where each layer has a kernel size of 1 and a stride of 1. Each one-dimensional convolutional layer is followed by a layer normalization layer to increase model training stability. Each layer normalization layer is followed by a ReLU activation function to increase the model's non-linear expressive power. Specifically, using one-dimensional convolutional layers, normalization layers, and the ReLU activation function, the one-dimensional multimodal data (local attribute vectors) is mapped layer by layer to (64→128→256→512) to extract high-dimensional attribute features as the first geological feature.

[0060] In addition, a channel attention module is added to enhance the weight allocation of key attribute features. Specifically, the channel attention module enhances the first geological feature to obtain local attribute features.

[0061] Step 1032: Using the two-dimensional convolutional layer in the pre-trained mineralization prediction model, extract the second geological feature from the two-dimensional multimodal data, and enhance the second geological feature through the spatial attention module to obtain the spatial context feature.

[0062] In practical implementation, the two-dimensional convolutional layer can be a spatial convolutional CNN. This two-dimensional convolutional layer (spatial convolutional module) is used to model the spatial structure information of the neighborhoods of both ore deposits and non-ore deposits. The two-dimensional convolutional layer can be at least three layers of a 2D CNN, where the kernel size is 3, and the kernel sizes are combined in combinations of (1×1, 3×3, 5×5, 7×7) to achieve multi-scale feature fusion. The stride is 1, and the padding is 1, with multiple pixels used to ensure multi-scale convolution. Each two-dimensional convolutional layer is followed by a batch normalization layer to increase model training stability. Each batch normalization layer is followed by a ReLU activation function to increase the model's non-linear expressive power. The ReLU activation function is followed by a max pooling layer (kernel size = 2, stride = 2). Specifically, by using two-dimensional convolutional layers, batch normalization layers, ReLU activation functions, and max pooling layers, the two-dimensional multimodal data (neighborhood spatial information) is gradually reduced in dimensionality and spatial features are extracted to obtain a high-dimensional spatial embedding (12×12×64→6×6×128→3×3×256→1×1×512), and the high-dimensional spatial embedding is used as the second geological feature.

[0063] In addition, residual connections or skip connections between convolutional layers improve the stability of deep training. A spatial attention module is also added to highlight the importance of mineralization-related pixels in the neighborhood. Specifically, the spatial attention module enhances the second geological feature to obtain spatial context features.

[0064] The above scheme utilizes a one-dimensional convolutional layer in a pre-trained mineralization prediction model to extract the first geological feature from one-dimensional multimodal data. This first geological feature is then enhanced using a channel attention module to obtain local attribute features. Similarly, a two-dimensional convolutional layer in the same pre-trained model is used to extract the second geological feature from two-dimensional multimodal data. This second geological feature is then enhanced using a spatial attention module to obtain spatial context features. In this way, the local attribute features accurately capture the geological attribute details of the center point location, while the spatial context features fully consider the geological spatial relationships of the surrounding neighborhood.

[0065] In some embodiments, step 104 includes:

[0066] Step 1041: Adaptively fuse the local attribute features and the spatial context features based on the channel attention mechanism to obtain the first fused feature.

[0067] In practice, global average pooling is used to determine the first-channel descriptor of the local attribute features and the second-channel descriptor of the spatial context features. The first dynamic weights corresponding to the local attribute features are determined based on the first-channel descriptors, and the second dynamic weights corresponding to the spatial context features are determined based on the second-channel descriptors. Based on the first and second dynamic weights, the local attribute features and the spatial context features are fused to obtain the first fused feature.

[0068] Step 1042: The local attribute features and the spatial context features are spliced ​​together to obtain spliced ​​features, and the spliced ​​features are processed by a gating unit to obtain a second fused feature.

[0069] In practice, the local attribute features and spatial context features are concatenated to obtain the concatenated features. ,in, For splicing features, For splicing processing, For local attribute features, This refers to spatial context features.

[0070] The second fused feature is obtained by processing the splicing features using a gating unit. Specifically, in the input gate, ,in, The first output parameter of the input gate. For activation function, The weight matrix represents the concatenated features of the input gate pairs. For splicing features, Let the input gate be the weight matrix of the hidden state at the previous time step. This is the hidden state from the previous moment. This is the bias vector for the input gate. In the forget gate, ,in, This is the second output parameter of the forget gate. For activation function, This is the weight matrix for the concatenated features of the forgetting gate. For splicing features, Let the forget gate be the weight matrix of the hidden state at the previous time step. This is the hidden state from the previous moment. This is the bias vector for the forget gate. In the output gate, ,in, This is the third output parameter of the output gate. For activation function, This is the weight matrix for the concatenated features of the output gate pairs. For splicing features, The output gate is the weight matrix of the hidden state at the previous time step. This is the hidden state from the previous moment. This is the bias vector for the output gate. ,in, This represents the current candidate hidden state. The hyperbolic tangent activation function is used. To calculate the weight matrix of the concatenated features when calculating candidate hidden states, For splicing features, This is the weight matrix for the hidden state at the previous time step after adjustment via the forget gate. This is the second output parameter of the forget gate. This is the hidden state from the previous moment. The bias vector calculated for the candidate hidden state. ,in, The current hidden state. The first output parameter of the input gate. This is the hidden state from the previous moment. This represents the candidate hidden state at the current moment. ,in, This is the second fusion feature. This is the third output parameter of the output gate. This represents the current hidden state.

[0071] In this way, the fusion ratio of local attribute features and spatial context features can be dynamically adjusted through nonlinear transformation to adapt to complex geological conditions.

[0072] Step 1043: Determine the target fusion feature based on the first fusion feature and the second fusion feature.

[0073] In practice, the first fusion feature and the second fusion feature are weighted to obtain the target fusion feature.

[0074] In this way, the adaptive feature fusion module adopts an adaptive fusion method with channel attention mechanism to adaptively fuse high-dimensional local attribute features (512 dimensions) and high-dimensional spatial context features (1×1×512) to form a 1024-dimensional target fusion feature.

[0075] The above scheme uses a channel attention mechanism to adaptively fuse local attribute features and spatial context features to obtain a first fused feature. The local attribute features and spatial context features are then stitched together to obtain a stitched feature, which is further processed using gating units to obtain a second fused feature. The target fused feature is then determined based on the first and second fused features. In this way, the target fused feature contains detailed local information while considering overall spatial relationships, enabling a more comprehensive and accurate reflection of the relationship between geological data and mineralization.

[0076] In some embodiments, step 1041 includes:

[0077] Step 1041A: Determine the first channel descriptor of the local attribute feature and the second channel descriptor of the spatial context feature.

[0078] In practice, the first channel descriptor of the local attribute features is determined. ,in, This is the first channel descriptor. For global average pooling, These are local attribute features. The second-channel descriptor determines the spatial context features. ,in, For the second channel descriptor, For global average pooling, This refers to spatial context features.

[0079] Step 1041B: Determine the first dynamic weight corresponding to the local attribute feature based on the first channel descriptor, and determine the second dynamic weight corresponding to the spatial context feature based on the second channel descriptor.

[0080] In practice, the first dynamic weight corresponding to the local attribute features is determined based on the first channel descriptor. ,in, As the first dynamic weight, It is the Sigmoid activation function. It is the ReLU activation function. This is the first channel descriptor. This is the first layer weight matrix. This is the weight matrix for the second layer. This is the first layer bias vector. This is the bias vector for the second layer.

[0081] The second dynamic weights corresponding to the spatial context features are determined based on the second channel descriptor. ,in, As the second dynamic weight, It is the Sigmoid activation function. It is the ReLU activation function. For the second channel descriptor, This is the first layer weight matrix. This is the weight matrix for the second layer. This is the first layer bias vector. This is the bias vector for the second layer.

[0082] Step 1041C: Based on the first dynamic weight and the second dynamic weight, the local attribute features and the spatial context features are fused to obtain the first fused feature.

[0083] In practice, the first fused feature is obtained by fusing local attribute features and spatial context features based on the first dynamic weight and the second dynamic weight. ,in, The first fusion feature, As the first dynamic weight, For local attribute features, As the second dynamic weight, This refers to spatial context features.

[0084] The above scheme determines a first-channel descriptor for local attribute features and a second-channel descriptor for spatial context features. Based on the first-channel descriptor, a first dynamic weight is determined for the local attribute features, and based on the second-channel descriptor, a second dynamic weight is determined for the spatial context features. Based on the first and second dynamic weights, the local attribute features and spatial context features are fused to obtain a first fused feature. Thus, by adaptively fusing local attribute features and spatial context features using a channel attention mechanism to obtain the first fused feature, the resulting first fused feature is more accurate.

[0085] In some embodiments, step 105 includes:

[0086] Step 1051: Based on the target fusion features, predict whether the center point is a mineral deposit to obtain a classification prediction result, and convert the classification prediction result into the mineralization probability value of the center point.

[0087] In practice, the fully connected layer in the classification prediction module is used to perform binary classification on the target fusion features to obtain the classification prediction result. The softmax is then used to map the classification prediction result to the probability space of [0,1] to obtain the mineralization probability value of the center point.

[0088] Step 1052: Use a sliding window to determine the mineralization probability value of all center points in the target area, and determine the mineralization probability map of the target area based on the mineralization probability values ​​of all center points.

[0089] In practice, a pixel-by-pixel window sliding method is used to process the mineralization probability values ​​of all center points in the target area to obtain a mineralization probability map of the target area. For example, if a certain area contains 29,134,336 sample sequences, each with a different spatial location of its center point, these 29,134,336 sample sequences are sequentially input into five independent mineralization prediction models to obtain five independent mineralization probability values ​​for the center points. Each independent mineralization probability value contains the mineralization probability values ​​for the 29,134,336 center points. Based on the spatial location of the center points of the sample sequences, the mineralization probability values ​​of each independent center point are combined into a single mineralization probability map of the target area, resulting in five independent mineralization probability maps of the target area. Each independent mineralization probability map of the target area has 5,912 rows and 4,928 columns.

[0090] Step 1053: Weight the multiple mineralization probability maps output by multiple mineralization prediction models to obtain the mineralization prediction results for the target area.

[0091] In practice, the multi-model integration strategy uses an independent sampling method based on the Bagging concept to construct multiple independent mineralization prediction models.

[0092] The mineralization probability maps output by each mineralization prediction model are fused using strategies such as pixel-level weighted averaging, adaptive weighted fusion based on validation set performance, and dynamic weight allocation based on attention mechanisms to obtain the final mineralization prediction result. This fusion strategy of multiple mineralization prediction models can alleviate the small sample size problem, data imbalance, and noise interference, thereby improving the generalization ability of the mineralization prediction models.

[0093] The mineralization prediction results of the target area are divided into mineralization potential zones. Among them, the area with a mineralization probability value of [0.9, 1.0] is divided into high potential zone, the area with a mineralization probability value of [0.5, 0.9) is divided into medium potential zone, the area with a mineralization probability value of [0.2, 0.5) is divided into low potential zone, and the area with a mineralization probability value of [0.0, 0.2) is divided into background value zone.

[0094] The accuracy of mineralization prediction results was evaluated by using the spatial location of known mineralization points and the spatial overlap of different potential areas. The evaluation results showed that among the 66 known mineralization points, 14 were located in high-potential areas, 48 ​​were located in medium-potential areas, and 4 were located in low-potential areas.

[0095] The above scheme predicts whether a center point is a mineral deposit based on target fusion features, obtaining classification prediction results. These prediction results are then converted into mineralization probability values ​​for the center points. A sliding window is used to determine the mineralization probability values ​​for all center points in the target area, and a mineralization probability map for the target area is determined based on these values. The multiple mineralization probability maps output by various mineralization prediction models are weighted to obtain the mineralization prediction results for the target area. This approach allows for a more accurate assessment of the mineralization potential of each center point, improving the accuracy of the mineralization prediction results for the target area.

[0096] In some embodiments, the pre-training process of the mineralization prediction model includes:

[0097] Step 102A: Obtain sample geological data of the target area, and determine positive and negative samples from the sample geological data.

[0098] In practice, the target area is divided into a predetermined number of grids based on the spatial locations of positive and negative samples in the geological data. The predetermined number of grids can be six, with each grid containing at least 10 positive samples and at least 10 negative samples.

[0099] Step 102B: Divide the positive samples into first training data and first test data according to a preset ratio, and divide the negative samples into second training data and second test data according to a preset ratio.

[0100] In practice, according to a preset example, the positive samples in each spatial grid are divided into first training data and first test data, where the preset ratio can be 8:2. Similarly, according to a preset example, the negative samples in each spatial grid are divided into second training data and second test data, where the preset ratio can also be 8:2.

[0101] Step 102C: Use the first training data and the second training data as training sample data, and use the first test data and the second test data as test sample data.

[0102] In practice, the first and second training data of the preset grid are merged to obtain training sample data (training set). The first and second test data of the preset grid are merged to obtain test sample data (test set).

[0103] Step 102D: Input the training sample data into the initial prediction model, and use the initial prediction model to determine the mineralization prediction map based on the training sample data.

[0104] In practice, there can be multiple initial prediction models. Each initial prediction model is trained to obtain a corresponding mineralization prediction model, meaning there are also multiple mineralization prediction models.

[0105] Figure 3 This is a schematic diagram of a mineralization prediction model according to an embodiment of this disclosure. Figure 3 As shown, the mineralization prediction model can consist of five relatively independent models, thus outputting five mineralization probability maps. These five probability maps can then be weighted to obtain the mineralization prediction result for the target area. Each mineralization prediction model is trained using sample data containing both positive and negative samples. The sample data includes training sample data (training set) and test sample data (test set).

[0106] The training sample data is input into the initial prediction model, and the initial prediction model is used to perform forward propagation on the training sample data to obtain the mineralization prediction map. The mineralization prediction map is obtained by predicting the training sample data using the initial prediction model.

[0107] Step 102E: Determine the cross-entropy loss function based on the actual mineralization labels corresponding to the mineralization prediction map and the training sample data, and update the model parameters of the initial prediction model according to the cross-entropy loss function to obtain the updated prediction model.

[0108] In practice, the cross-entropy loss function is determined based on the actual mineralization labels corresponding to the mineralization prediction map and training sample data. The gradient of each layer parameter in the initial prediction model is calculated through backpropagation. The optimizer Adam is used to update the model parameters of the initial prediction model according to the cross-entropy loss function to obtain the updated prediction model.

[0109] Step 102F: Validate the updated prediction model using the test sample data. In response to determining that the updated prediction model has passed validation, adopt the updated prediction model as the mineralization prediction model.

[0110] In practice, test sample data is input into the updated prediction model, and the F1-score values ​​of the training and test sample data are compared to evaluate the generalization ability of the updated prediction model. During model training, the changes in the F1-score values ​​of the training and test sample data are stored in the training log and visualized to analyze the training status of the updated prediction model, detect overfitting and underfitting, adjust the hyperparameters of the updated prediction model, and set dropout in the fully connected layers to help select the optimal updated prediction model as the mineralization prediction model.

[0111] In addition to the above, after obtaining the updated prediction model, the process also includes: recording the number of training iterations; and, in response to determining that the number of training iterations has reached a preset threshold, using the tested updated prediction model as the mineralization prediction model. For example, if the preset threshold is 100 iterations, when the number of training iterations reaches 100, the updated prediction model gradually converges, and the tested updated prediction model is used as the mineralization prediction model.

[0112] The above scheme acquires sample geological data of the target area and identifies positive and negative samples from this data. Positive samples are divided into first training data and first test data according to a preset ratio, and negative samples are divided into second training data and second test data according to a preset ratio. The first and second training data are used as training samples, and the first and second test data are used as test samples, making the training and test data more comprehensive and accurate. The training sample data is input into the initial prediction model, which then determines a mineralization prediction map based on the training sample data. A cross-entropy loss function is determined based on the actual mineralization labels corresponding to the mineralization prediction map and the training sample data. The model parameters of the initial prediction model are updated using the cross-entropy loss function to obtain an updated prediction model. The updated prediction model is validated using test sample data. When the updated prediction model passes validation, it is adopted as the mineralization prediction model. This approach enables precise model training, resulting in a more accurate mineralization prediction model.

[0113] Through the above embodiments, multimodal geological images of the target area are obtained. Using a pre-trained mineralization prediction model, one-dimensional multimodal data corresponding to the center point and two-dimensional multimodal data of the center point's neighborhood are determined from the multimodal geological images. The one-dimensional multimodal data can cover detailed local geological attributes corresponding to the center point, while the two-dimensional multimodal data can cover a wider range of spatial geological information in the center point's neighborhood. Both one-dimensional and two-dimensional multimodal data have rich data dimensions, enabling accurate mineralization prediction. Local attribute features are extracted from the one-dimensional multimodal data, and spatial context features are extracted from the two-dimensional multimodal data. Local attribute features accurately capture the geological attribute details of the center point location, while spatial context features fully consider the geological spatial relationships of the surrounding neighborhood. The local attribute features and spatial context features are fused to obtain target fusion features. Target fusion features contain detailed local information and consider overall spatial relationships, more comprehensively and accurately reflecting the relationship between geological data and mineralization. The mineralization probability value of the center point is determined based on the target fusion features, and the mineralization prediction result of the target area is determined based on the mineralization probability values ​​of all center points in the target area. This allows for a more accurate assessment of the mineralization potential of each central point, improving the accuracy of mineralization prediction results for the target area.

[0114] It should be noted that the embodiments of this disclosure can also be further described in the following ways:

[0115] Figure 4 This is a flowchart illustrating a regional mineralization prediction method based on an attribute-spatial convolutional neural network, as described in an embodiment of this disclosure. Figure 4 As shown, the regional mineralization prediction method based on attribute-spatial convolutional neural networks includes:

[0116] Step S1: Local attribute information extraction, extracting the geological-geophysical-geochemical-remote sensing spectral features of mineral deposits and non-mineral deposits.

[0117] Step S2: Extract neighborhood spatial information by extracting the multi-source spatial feature matrix of the neighborhood range of the center point.

[0118] Step S3: The one-dimensional convolutional layer (Attribute CNN) learns high-dimensional attribute feature representations.

[0119] Step S4: The two-dimensional convolutional layer (Spatial CNN) models the spatial context relationship to achieve multi-scale spatial feature extraction.

[0120] Step S5, Feature Fusion and Classification: Adaptive fusion of attribute features and spatial features is achieved through learnable weights, and the fully connected classification outputs the mineralization probability of the mineral point.

[0121] Step S6 involves sample cross-partitioning and multi-model ensemble based on the Bagging concept, with the multi-model prediction results using a pixel-level weighted average.

[0122] Step S7: Generate a regional mineralization prediction dataset, predict favorable areas for mineralization, and evaluate the prediction results.

[0123] It should be noted that the method of this disclosure embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this disclosure embodiment, and the multiple devices will interact with each other to complete the method described.

[0124] It should be noted that some embodiments of this disclosure have been described above. Other embodiments are within the scope of the appended embodiments. In some cases, the actions or steps described in the embodiments may be performed in a different order than that shown in the above embodiments and the desired results may still be achieved. In addition, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are also possible or may be advantageous.

[0125] Based on the same inventive concept, corresponding to any of the above embodiments, this disclosure also provides a ore-forming prediction device based on neural networks.

[0126] refer to Figure 5 The neural network-based mineralization prediction device includes:

[0127] The acquisition module 301 is configured to acquire multimodal geological images of the target area;

[0128] The geological data determination module 302 is configured to use a pre-trained mineralization prediction model to determine the one-dimensional multimodal data corresponding to the center point and the two-dimensional multimodal data of the neighborhood of the center point from the multimodal geological image.

[0129] The feature extraction module 303 is configured to extract local attribute features from the one-dimensional multimodal data and extract spatial context features from the two-dimensional multimodal data.

[0130] The feature fusion module 304 is configured to perform adaptive fusion processing on the local attribute features and the spatial context features to obtain the target fused features;

[0131] The mineralization prediction module 305 is configured to determine the mineralization probability value of the center point based on the target fusion features, and to determine the mineralization prediction result of the target area based on the mineralization probability values ​​of all center points in the target area combined with the integration strategy.

[0132] In some embodiments, the geological data determination module 302 includes:

[0133] A center point determination unit is configured to determine a center point for mineralization prediction from the target area;

[0134] A one-dimensional multimodal data determination unit is configured to use a first extraction module in a pre-trained mineralization prediction model to determine multiple first geological parameters corresponding to the center point from the multimodal geological image, and convert the multiple first geological parameters into one-dimensional multimodal data corresponding to the center point.

[0135] The two-dimensional multimodal data determination unit is configured to use a second extraction module in a pre-trained mineralization prediction model to determine multiple second geological parameters of the neighborhood of the center point from the multimodal geological image, and convert the multiple second geological parameters into two-dimensional multimodal data of the neighborhood of the center point.

[0136] In some embodiments, the feature extraction module 303 includes:

[0137] The local attribute feature extraction unit is configured to extract a first geological feature from the one-dimensional multimodal data using a one-dimensional convolutional layer in a pre-trained mineralization prediction model, and to obtain local attribute features by enhancing the first geological feature through a channel attention module.

[0138] The spatial context feature extraction unit is configured to extract a second geological feature from the two-dimensional multimodal data using a two-dimensional convolutional layer in a pre-trained mineralization prediction model, and to obtain spatial context features by enhancing the second geological feature through a spatial attention module.

[0139] In some embodiments, the feature fusion module 304 includes:

[0140] The first fusion feature determination unit is configured to adaptively fuse the local attribute features and the spatial context features based on a channel attention mechanism to obtain the first fusion feature.

[0141] The second fusion feature determination unit is configured to concatenate the local attribute features and the spatial context features to obtain concatenated features, and then process the concatenated features using a gating unit to obtain the second fusion feature.

[0142] The target fusion feature determination unit is configured to determine the target fusion feature based on the first fusion feature and the second fusion feature.

[0143] In some embodiments, the first fusion feature determination unit includes:

[0144] The channel descriptor determining subunit is configured to determine a first channel descriptor of the local attribute features and a second channel descriptor of the spatial context features;

[0145] The dynamic weight determination subunit is configured to determine a first dynamic weight corresponding to the local attribute feature based on the first channel descriptor, and to determine a second dynamic weight corresponding to the spatial context feature based on the second channel descriptor.

[0146] The first fusion feature determination subunit is configured to fuse the local attribute features and the spatial context features based on the first dynamic weight and the second dynamic weight to obtain the first fusion feature.

[0147] In some embodiments, the mineralization prediction module 305 includes:

[0148] The mineralization probability value determination unit is configured to predict whether the center point is a mineralization point based on the target fusion features to obtain a classification prediction result, and convert the classification prediction result into the mineralization probability value of the center point;

[0149] The mineralization probability map determination unit is configured to determine the mineralization probability value of all center points in the target area using a sliding window, and to determine the mineralization probability map of the target area based on the mineralization probability values ​​of all center points.

[0150] The mineralization prediction result determination unit is configured to perform weighted processing on multiple mineralization probability maps output by multiple mineralization prediction models to obtain the mineralization prediction result of the target area.

[0151] In some embodiments, the apparatus further includes a model training module, the model training module comprising:

[0152] The sample geological data acquisition unit is configured to acquire sample geological data of a target area and determine positive and negative samples from the sample geological data;

[0153] The sample partitioning unit is configured to partition the positive samples into first training data and first test data according to a preset ratio, and to partition the negative samples into second training data and second test data according to a preset ratio.

[0154] The sample data determination unit is configured to use the first training data and the second training data as training sample data, and the first test data and the second test data as test sample data;

[0155] The mineralization prediction map determination unit is configured to input the training sample data into an initial prediction model and use the initial prediction model to determine a mineralization prediction map based on the training sample data.

[0156] The model update unit is configured to determine the cross-entropy loss function based on the actual mineralization labels corresponding to the mineralization prediction map and the training sample data, and update the model parameters of the initial prediction model according to the cross-entropy loss function to obtain the updated prediction model.

[0157] The model validation unit is configured to validate the updated prediction model using the test sample data, and in response to determining that the updated prediction model has passed validation, to use the updated prediction model as a mineralization prediction model.

[0158] For ease of description, the above apparatus is described in terms of its functions, divided into various modules. Of course, in implementing this disclosure, the functions of each module can be implemented in one or more software and / or hardware.

[0159] The apparatus of the above embodiments is used to implement the corresponding neural network-based mineralization prediction method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0160] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the ore-forming prediction method based on neural networks described in any of the above embodiments.

[0161] Figure 6 This embodiment illustrates a more specific hardware structure of an electronic device. The device may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0162] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0163] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0164] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0165] The communication interface 1040 is used to connect the communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB (Universal Serial Bus), network cable, etc.) or wireless means (such as mobile network, WIFI (Wireless Fidelity), Bluetooth, etc.).

[0166] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0167] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0168] The electronic devices described above are used to implement the corresponding neural network-based mineralization prediction methods in any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0169] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this disclosure also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the neural network-based mineralization prediction method as described in any of the above embodiments.

[0170] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0171] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the ore-forming prediction method based on neural networks as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0172] Based on the same inventive concept, corresponding to any of the above embodiments, this application also provides a computer program product, including computer program instructions. When the computer program instructions are run on a computer, the computer causes the computer to execute the ore-forming prediction method based on neural networks as described in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0173] It is understood that before using the technical solutions of the various embodiments in this disclosure, users will be informed of the type, scope of use, and usage scenarios of the personal information involved in an appropriate manner, and user authorization will be obtained.

[0174] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose, based on the prompt message, whether to provide personal information to the software or hardware such as electronic devices, applications, servers, or storage media performing the operations of this disclosed technical solution.

[0175] As an optional but not limited implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0176] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0177] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this disclosure is limited to these examples; within the framework of this disclosure, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this disclosure as described above, which are not provided in detail for the sake of brevity.

[0178] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this disclosure, the provided drawings may or may not show well-known power / ground connections to integrated circuit (IC) chips and other components. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this disclosure, and this also takes into account the fact that the details of implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this disclosure will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this disclosure, it will be apparent to those skilled in the art that the embodiments of this disclosure can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0179] Although this disclosure has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0180] This disclosure is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this disclosure. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the protection scope of this disclosure.

Claims

1. A method for predicting mineral deposits based on neural networks, characterized in that, The method includes: Acquire multimodal geological images of the target area; Using a pre-trained mineralization prediction model, one-dimensional multimodal data corresponding to the center point and two-dimensional multimodal data of the neighborhood of the center point are determined from the multimodal geological image. Local attribute features are extracted from the one-dimensional multimodal data, and spatial context features are extracted from the two-dimensional multimodal data; The target fusion feature is obtained by adaptively fusing the local attribute features and the spatial context features; Based on the target fusion features, the mineralization probability value of the center point is determined, and the mineralization prediction result of the target area is determined by combining the mineralization probability values ​​of all center points in the target area with the integration strategy. The step of determining the one-dimensional multimodal data corresponding to the center point and the two-dimensional multimodal data of the center point's neighborhood from the multimodal geological image includes: Determine the center point for mineralization prediction from the target area; Using the first extraction module in the pre-trained mineralization prediction model, multiple first geological parameters corresponding to the center point are determined from the multimodal geological image, and the multiple first geological parameters are converted into one-dimensional multimodal data corresponding to the center point. Using the second extraction module in the pre-trained mineralization prediction model, multiple second geological parameters of the neighborhood of the center point are determined from the multimodal geological image, and the multiple second geological parameters are converted into two-dimensional multimodal data of the neighborhood of the center point.

2. The method according to claim 1, characterized in that, The step of extracting local attribute features from the one-dimensional multimodal data and extracting spatial context features from the two-dimensional multimodal data includes: Using a one-dimensional convolutional layer in a pre-trained mineralization prediction model, a first geological feature is extracted from the one-dimensional multimodal data. The first geological feature is then enhanced using a channel attention module to obtain local attribute features. Using a two-dimensional convolutional layer in a pre-trained mineralization prediction model, a second geological feature is extracted from the two-dimensional multimodal data. The second geological feature is then enhanced using a spatial attention module to obtain spatial context features.

3. The method according to claim 1, characterized in that, The adaptive fusion processing of the local attribute features and the spatial context features to obtain the target fusion features includes: The first fused feature is obtained by adaptively fusing the local attribute features and the spatial context features based on the channel attention mechanism; The local attribute features and the spatial context features are concatenated to obtain concatenated features, and the concatenated features are processed by a gating unit to obtain a second fused feature; The target fusion feature is determined based on the first fusion feature and the second fusion feature.

4. The method according to claim 3, characterized in that, The adaptive fusion of the local attribute features and the spatial context features based on the channel attention mechanism to obtain the first fused feature includes: Determine the first channel descriptor of the local attribute features and the second channel descriptor of the spatial context features; The first dynamic weight corresponding to the local attribute feature is determined based on the first channel descriptor, and the second dynamic weight corresponding to the spatial context feature is determined based on the second channel descriptor. Based on the first dynamic weight and the second dynamic weight, the local attribute features and the spatial context features are fused to obtain the first fused feature.

5. The method according to claim 1, characterized in that, The process of determining the mineralization probability value of the center point based on the target fusion features, and determining the mineralization prediction result of the target region based on the mineralization probability values ​​of all center points in the target region combined with the ensemble strategy, includes: Based on the target fusion features, a classification prediction result is obtained to predict whether the center point is a mineral deposit, and the classification prediction result is converted into the mineralization probability value of the center point. The mineralization probability value of all center points in the target area is determined using a sliding window, and a mineralization probability map of the target area is determined based on the mineralization probability values ​​of all center points. The mineralization prediction results for the target area are obtained by weighting multiple mineralization probability maps output by multiple mineralization prediction models.

6. The method according to claim 1, characterized in that, The pre-training process of the mineralization prediction model includes: Obtain sample geological data of the target area, and determine positive and negative samples from the sample geological data; The positive samples are divided into first training data and first test data according to a preset ratio, and the negative samples are divided into second training data and second test data according to a preset ratio. The first training data and the second training data are used as training sample data, and the first test data and the second test data are used as test sample data; The training sample data is input into the initial prediction model, and the initial prediction model is used to determine the mineralization prediction map based on the training sample data. Based on the mineralization prediction map and the actual mineralization labels corresponding to the training sample data, a cross-entropy loss function is determined, and the model parameters of the initial prediction model are updated according to the cross-entropy loss function to obtain an updated prediction model. The updated prediction model is validated using the test sample data. In response to determining that the updated prediction model has passed validation, the updated prediction model is adopted as the mineralization prediction model.

7. A neural network-based mineralization prediction device, characterized in that, include: The acquisition module is configured to acquire multimodal geological images of the target area; The geological data determination module is configured to use a pre-trained mineralization prediction model to determine the one-dimensional multimodal data corresponding to the center point and the two-dimensional multimodal data of the neighborhood of the center point from the multimodal geological image. The feature extraction module is configured to extract local attribute features from the one-dimensional multimodal data and spatial context features from the two-dimensional multimodal data. The feature fusion module is configured to perform adaptive fusion processing on the local attribute features and the spatial context features to obtain the target fused features; The mineralization prediction module is configured to determine the mineralization probability value of the center point based on the target fusion features, and to determine the mineralization prediction result of the target area based on the mineralization probability values ​​of all center points in the target area combined with the integration strategy. The geological data determination module includes: A center point determination unit is configured to determine a center point for mineralization prediction from the target area; A one-dimensional multimodal data determination unit is configured to use a first extraction module in a pre-trained mineralization prediction model to determine multiple first geological parameters corresponding to the center point from the multimodal geological image, and convert the multiple first geological parameters into one-dimensional multimodal data corresponding to the center point. The two-dimensional multimodal data determination unit is configured to use a second extraction module in a pre-trained mineralization prediction model to determine multiple second geological parameters of the neighborhood of the center point from the multimodal geological image, and convert the multiple second geological parameters into two-dimensional multimodal data of the neighborhood of the center point.

8. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the program, implements the method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions for causing a computer to perform the method according to any one of claims 1 to 6.

Images