A three-dimensional linear structure intelligent identification method based on gravity and magnetic data

Abstract

The present application relates to geophysical exploration and data processing technical field, disclose a kind of 3D linear structure intelligent identification method based on gravity and magnetic data, comprising the following steps: obtaining the gravity and magnetic anomaly data of target area under original observation height, and the gravity and magnetic anomaly data is extended to calculate, obtain multiple different observation height gravity and magnetic anomaly plane grid data;The gravity and magnetic anomaly plane grid data under original observation height and the gravity and magnetic anomaly plane grid data of multiple different observation height are fused according to predetermined order, construct 3D gravity and magnetic data volume containing multi-height information, the present application is comprehensively utilized to different observation height gravity and magnetic anomaly data, can obtain the 3D distribution result of linear structure in underground space, can effectively make up the deficiency that linear structure identification result in prior art is limited to two-dimensional plane distribution, difficult to reflect the extension characteristics of structure in depth direction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geophysical exploration and data processing, and particularly relates to a three-dimensional linear structure intelligent identification method based on gravity and magnetic data. BACKGROUND

[0002] Traditional geological work methods mainly rely on surface geological observation to obtain structural information, and the underground structural morphology is inferred by analyzing the surface outcropped geological phenomena. However, in the area with serious surface coverage, it is often difficult to accurately identify the structural morphology due to the lack of effective outcropping conditions. Even in the bedrock outcrop area, the judgment of structural development trend usually depends on empirical inference, which has strong subjectivity and uncertainty.

[0003] In contrast, the geophysical method can reflect the physical structure and structural characteristics of the underground space by observing the physical field response information caused by the physical property difference of the underground geological body, and has obvious advantages in underground structure identification. Among them, the gravity anomaly data and the magnetic anomaly data have high sensitivity to the density difference and magnetic difference of the underground geological body, and have strong lateral resolution capability, which plays an important role in structural identification, especially in structural boundary interpretation.

[0004] The existing structure identification method of gravity and magnetic data mainly focuses on the extraction of the horizontal distribution characteristics of the structure. Common methods can be summarized as anomaly derivative method, numerical statistical method and image processing algorithm, etc. This kind of method is mostly based on two-dimensional plane gravity and magnetic anomaly data for analysis, which can reflect the distribution characteristics of the structure in the horizontal plane to a certain extent, but it is difficult to effectively reveal the extension of the structure in the depth direction.

[0005] In order to obtain structural information of different depths and different scales, some studies carry out potential field continuation or multi-scale decomposition on gravity and magnetic anomaly data, and use the characteristics of gravity and magnetic field under different height space distribution or different scale resolution to reflect the geological structural characteristics at different depths and different scales. On this basis, the multi-scale edge detection method based on different height gravity and magnetic anomaly data can further obtain the structural gradual change trend from shallow to deep according to the linear beam characteristics obtained by interpretation, which has certain application value in the capture of linear structure of large-scale gravity and magnetic data in ore concentration area.

[0006] However, although the multi-scale edge detection method and other methods realize the semi-quantitative description of the structural boundary at different depths to a certain extent by potential field continuation, the interpretation process still highly depends on manual intervention and empirical inference, and the automation and quantification degree is insufficient. In the area with serious surface coverage, the existing method is still difficult to accurately identify the structural morphology, and the judgment of structural development trend still has great uncertainty.

[0007] The linear structure result deduced by the existing gravity and magnetic data structure identification method is mostly a two-dimensional plane distribution result, which cannot provide three-dimensional distribution characteristics of the linear structure in the underground space, and is difficult to meet the needs of fine geological interpretation and ore prediction for three-dimensional structure information. SUMMARY

[0008] The application aims to solve the problems in the prior art and provides a three-dimensional linear structure intelligent identification method based on gravity and magnetic data.

[0009] In order to achieve the above-mentioned purpose, the application adopts the following technical scheme:

[0010] A three-dimensional linear structure intelligent identification method based on gravity and magnetic data comprises the following steps:

[0011] Obtain gravity and magnetic anomaly data of a target area at an original observation height, and perform continuation calculation on the gravity and magnetic anomaly data to obtain gravity and magnetic anomaly plane grid data at multiple different observation heights;

[0012] Fuse the gravity and magnetic anomaly plane grid data at the original observation height and the gravity and magnetic anomaly plane grid data at the multiple different observation heights in a predetermined order to construct a three-dimensional gravity and magnetic data body containing multi-height information;

[0013] Input the three-dimensional gravity and magnetic data body into a pre-constructed three-dimensional linear structure intelligent identification network, and use the intelligent identification network to perform feature extraction and linear structure prediction on the three-dimensional gravity and magnetic data body;

[0014] Output the three-dimensional distribution result of the predicted linear structure in the underground space by the intelligent identification network, wherein a spatial unit where the structure exists is marked as a structure unit, and a spatial unit where the structure does not exist is marked as a non-structure unit, and the spatial distribution of the structure unit represents three-dimensional spatial distribution information of the linear structure.

[0015] Preferably, the step of obtaining gravity and magnetic anomaly data of a target area at an original observation height and performing continuation calculation on the gravity and magnetic anomaly data to obtain gravity and magnetic anomaly plane grid data at multiple different observation heights comprises:

[0016] Obtain gravity and magnetic anomaly data of a target area at an original observation height;

[0017] When the gravity and magnetic anomaly data is gravity and magnetic anomaly data obtained by ground observation, perform upward continuation calculation on the gravity and magnetic anomaly data according to the distance from the measuring point to obtain gravity and magnetic anomaly plane grid data at multiple heights higher than the original observation height;

[0018] When the gravity and magnetic anomaly data is gravity and magnetic anomaly data obtained by aerial observation, the gravity and magnetic anomaly data is downward continued according to the distance from the measuring point, and a plurality of gravity and magnetic anomaly plane grid data at a lower height than the original observation height is obtained.

[0019] Preferably, the three-dimensional linear structure intelligent recognition network takes a U-Net network structure as a skeleton network, and includes a plurality of convolution layers, pooling layers, up-sampling layers, and skip connection structures.

[0020] Preferably, the convolution layers adopt a 3*3 convolution operator, the pooling layers adopt a 3*3 maximum pooling operator, and the up-sampling layers adopt a 3*3 up-sampling operator.

[0021] Preferably, a gated attention mechanism is introduced into the three-dimensional linear structure intelligent recognition network, and a continuous weight ranging from 0 to 1 is learned to weight the features obtained in the feature extraction process, so as to enhance the feature regions related to the linear structure.

[0022] Preferably, the gated attention mechanism is implemented based on a 1*1*1 convolution, and the spatial positions of the features obtained in the feature extraction process are selected by the weight generated by the gated attention mechanism.

[0023] Preferably, in the training process of the three-dimensional linear structure intelligent recognition network, a mean square error is used as a loss function, and the parameters of the intelligent recognition network are optimized by comparing the error between the three-dimensional distribution result of the linear structure output by the intelligent recognition network and a preset structure distribution result.

[0024] Preferably, in the three-dimensional distribution result, the linear structure is expressed in the form of a three-dimensional block, and the trend, tendency, and deep extension characteristics of the linear structure are reflected by the spatial arrangement of the three-dimensional block.

[0025] A three-dimensional linear structure intelligent recognition system based on gravity and magnetic data, the system comprising:

[0026] A data acquisition and continuation module is configured to acquire gravity and magnetic anomaly data of a target region at an original observation height, and to perform continuation calculation on the gravity and magnetic anomaly data to obtain a plurality of gravity and magnetic anomaly plane grid data at different observation heights.

[0027] A multi-height data fusion module is configured to fuse the gravity and magnetic anomaly plane grid data at the original observation height and the gravity and magnetic anomaly plane grid data at the different observation heights in a predetermined order to construct a three-dimensional gravity and magnetic data volume containing multi-height information.

[0028] The intelligent recognition module is configured to input the three-dimensional gravity and magnetic data volume into a pre-constructed three-dimensional linear structure intelligent recognition network, and utilize the intelligent recognition network to perform feature extraction and linear structure prediction on the three-dimensional gravity and magnetic data volume.

[0029] The result output module is configured to output a three-dimensional distribution result of the predicted linear structure in the underground space by the intelligent recognition network, wherein a spatial unit in which the structure exists is marked as a structure unit, and a spatial unit in which the structure does not exist is marked as a non-structure unit, and the spatial distribution of the structure unit represents three-dimensional spatial distribution information of the linear structure.

[0030] The present application has the following advantages:

[0031] The three-dimensional linear structure intelligent recognition method based on gravity and magnetic data can obtain a three-dimensional distribution result of the linear structure in the underground space by comprehensively utilizing gravity and magnetic anomaly data at different observation heights, and can effectively make up for the deficiency that the linear structure recognition result is limited to two-dimensional plane distribution and cannot reflect the extension characteristics of the structure in the depth direction.

[0032] The present application constructs a three-dimensional linear structure intelligent recognition network, and inputs multi-height gravity and magnetic anomaly data as network input, so as to provide different depth underground information for the network, and enable the network to extract feature information related to the linear structure on the basis of comprehensively utilizing gravity and magnetic field characteristics at different heights, thereby realizing automatic prediction of the three-dimensional spatial distribution characteristics of the linear structure and reducing the dependence of the structure interpretation process on artificial experience inference.

[0033] The present application introduces a gating attention mechanism to perform weighted processing on the features obtained in the feature extraction process, which is helpful to enhance the feature region related to the linear structure and improve the recognition ability of the structure boundary and the structure form. In the gravity and magnetic anomaly data processing process, by avoiding the use of batch normalization operation, the physical information of the underground geological body carried in the gravity and magnetic anomaly values is retained, which is beneficial to the network to learn the features related to the physical distribution of the underground structure and improve the effectiveness of the structure information extraction.

[0034] Through the above technical means, the method of the present application can realize automatic and quantitative three-dimensional recognition of the linear structure, and the obtained three-dimensional distribution result of the structure has good stability and consistency, which can provide reliable method and technical support for the underground structure recognition of the coverage area and the structure interpretation and modeling of the ore concentration area, and realize high-precision and high-resolution structure interpretation and modeling of the ore concentration area. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 A technical flowchart of a three-dimensional linear structure intelligent recognition method based on gravity and magnetic data is provided for the embodiments of the present application.

[0036] Figure 2 A network structure schematic diagram of a three-dimensional linear structure intelligent recognition method based on gravity and magnetic data is proposed for an embodiment of the present application.

[0037] Figure 3 Multi-height gravity anomaly data of a single linear structure model constructed for embodiment 1 of the present application, wherein (a) is the gravity anomaly at the original ground observation height, (b) is the gravity anomaly upwardly continued to 1 times the observation height of the measuring point distance, (c) is the gravity anomaly upwardly continued to 2 times the observation height of the measuring point distance, (d) is the gravity anomaly upwardly continued to 3 times the observation height of the measuring point distance, (e) is the gravity anomaly upwardly continued to 4 times the observation height of the measuring point distance, and (f) is the gravity anomaly upwardly continued to 5 times the observation height of the measuring point distance;

[0038] Figure 4 Prediction results of a single linear structure gravity model corresponding to embodiment 1 of the present application, wherein (a) is a true structure model, (b) is a network-predicted structure model, (c) is a plan view of the true structure model at z=0 m, (d) is a plan view of the network-predicted structure model at z=0 m, (e) is a profile view of the true structure model at x=2000 m, (f) is a profile view of the network-predicted structure model at x=2000 m, (g) is a profile view of the true structure model at x=5000 m, and (h) is a profile view of the network-predicted structure model at x=5000 m.

[0039] Figure 5 Multi-height magnetic anomaly data of a single linear structure model constructed for embodiment 2 of the present application, wherein (a) is the magnetic anomaly at the original ground observation height, (b) is the magnetic anomaly upwardly continued to 1 times the observation height of the measuring point distance, (c) is the magnetic anomaly upwardly continued to 2 times the observation height of the measuring point distance, (d) is the magnetic anomaly upwardly continued to 3 times the observation height of the measuring point distance, (e) is the magnetic anomaly upwardly continued to 4 times the observation height of the measuring point distance, and (f) is the magnetic anomaly upwardly continued to 5 times the observation height of the measuring point distance.

[0040] Figure 6 Prediction results of a single linear structure magnetic model corresponding to embodiment 2 of the present application, wherein (a) is a true structure model, (b) is a network-predicted structure model; (c) is a plan view of the true structure model at z=0 m; (d) is a plan view of the network-predicted structure model at z=0 m; (e) is a profile view of the true structure model at x=2000 m; (f) is a profile view of the network-predicted structure model at x=2000 m; (g) is a profile view of the true structure model at x=5000 m; and (h) is a profile view of the network-predicted structure model at x=5000 m. DETAILED DESCRIPTION

[0041] In order to have a clearer understanding of the technical features, objectives and benefits of the present application, the technical solutions of the present application are described in detail as follows, but it should not be understood as a limitation to the scope of the present application. Unless otherwise specified, the methods used in the present application are conventional methods in the technical field. In the present application, the materials, reagents or instruments used are not specified by the manufacturer, and are conventional products that can be obtained by commercial purchase.

[0042] The specific implementation of the present application is described in detail below in combination with specific examples.

[0043] With reference to Figure 1 The embodiment of the present application proposes a three-dimensional linear structure intelligent identification method based on gravity and magnetic data, which specifically includes the following steps:

[0044] Step S100, obtain gravity and magnetic anomaly data of a target area at an original observation height, and perform continuation calculation on the gravity and magnetic anomaly data to obtain gravity and magnetic anomaly plane grid data at multiple different observation heights.

[0045] The step S100 specifically includes the following steps:

[0046] Obtain gravity and magnetic anomaly data of a target area at an original observation height; when the gravity and magnetic anomaly data is gravity and magnetic anomaly data obtained by ground observation, perform upward continuation calculation on the gravity and magnetic anomaly data according to the measurement point distance to obtain gravity and magnetic anomaly plane grid data at multiple heights higher than the original observation height; when the gravity and magnetic anomaly data is gravity and magnetic anomaly data obtained by aerial observation, perform downward continuation calculation on the gravity and magnetic anomaly data according to the measurement point distance to obtain gravity and magnetic anomaly plane grid data at multiple heights lower than the original observation height.

[0047] In the embodiment of the present application, the measured gravity and magnetic anomaly data of the target area is subjected to continuation calculation to obtain gravity and magnetic anomaly plane grid data at multiple heights.

[0048] When the gravity and magnetic anomaly data is gravity and magnetic anomaly plane grid data obtained by ground observation, perform upward continuation calculation on the gravity and magnetic anomaly data according to the measurement point distance, and the continuation height is set to 1-5 times the measurement point distance of the original observation height, thereby obtaining gravity and magnetic anomaly plane grid data at multiple different observation heights except the original observation height.

[0049] When the gravity and magnetic anomaly data is gravity and magnetic anomaly data obtained by aerial observation, perform downward continuation calculation on the gravity and magnetic anomaly data according to the measurement point distance, and the continuation height is set to 1-5 times the measurement point distance of the original observation height, thereby obtaining gravity and magnetic anomaly plane grid data at multiple different observation heights except the original observation height.

[0050] Further, the three-dimensional linear structure intelligent recognition method based on gravity and magnetic data further comprises the following steps:

[0051] In step S200, the gravity and magnetic anomaly plane grid data at the original observation height and the gravity and magnetic anomaly plane grid data at the plurality of different observation heights are fused in a predetermined order to construct a three-dimensional gravity and magnetic data body containing multi-height information.

[0052] In the embodiment, the gravity and magnetic anomaly plane grid data at the original observation height and the gravity and magnetic anomaly plane grid data at the plurality of different observation heights obtained by continuation calculation are arranged in a predetermined order from low to high observation height and are fused in sequence to construct a three-dimensional gravity and magnetic data body containing multi-height information. In the embodiment, the gravity and magnetic anomaly plane grid data at five different observation heights obtained by continuation calculation.

[0053] Specifically, the three-dimensional gravity and magnetic data body takes the gravity and magnetic anomaly plane grid data corresponding to each observation height as different layers, each layer corresponds to the gravity and magnetic anomaly information at an observation height, and the three-dimensional gravity and magnetic data body contains the gravity and magnetic anomaly information at the original observation height and the gravity and magnetic anomaly information at the plurality of different observation heights by combining the layer data in the vertical direction.

[0054] The three-dimensional gravity and magnetic data body constructed by the above fusion method can comprehensively reflect the spatial variation characteristics of the gravity and magnetic anomaly at different observation heights, and provides input data containing multi-height information for subsequent feature extraction and linear structure prediction based on the three-dimensional linear structure intelligent recognition network.

[0055] Further, the three-dimensional linear structure intelligent recognition method based on gravity and magnetic data further comprises the following steps:

[0056] In step S300, the three-dimensional gravity and magnetic data body is input into the pre-constructed three-dimensional linear structure intelligent recognition network, and the intelligent recognition network is used to extract features and predict linear structures of the three-dimensional gravity and magnetic data body.

[0057] In the embodiment, the three-dimensional linear structure intelligent recognition network is constructed to extract features and predict linear structures of the three-dimensional gravity and magnetic data body. The three-dimensional linear structure intelligent recognition network takes the U-Net network structure as the skeleton network, and the overall structure is as shown in Figure 2 The specific channel number is represented in the manner of multiplying the numbers in the figure.

[0058] The three-dimensional linear structure intelligent recognition network is composed of multiple convolution layers, maximum pooling layers, up-sampling layers and skip connection structures, wherein the convolution operator size of the convolution layer is 3x3, the size of the maximum pooling layer is 3x3, and the size of the up-sampling layer is 3x3. The size of the input data is 6x128x128, and the size of the output data is 50x128x128.

[0059] In the down-sampling stage of the network, each convolution layer is composed of a convolution operator and a ReLU activation function. The convolution operator is used to extract feature information from the input data, and then the extracted features are down-sampled through a maximum pooling operation to reduce the data size and retain relatively large-scale features. Then, the feature information at this scale is further extracted in the next convolution layer, and the above process is repeated until the deepest layer of the network.

[0060] After the feature extraction is completed, the feature information from the lower layer is transmitted to the upper layer through the up-sampling layer, and the skip connection structure is used to fuse the feature information at different scales, so as to recover the spatial resolution while comprehensively utilizing the feature information at different scales, and to fully express the linear structure features.

[0061] In the three-dimensional linear structure intelligent recognition network, each convolution layer is composed of a convolution operator and a ReLU activation function, and its expression form is: wherein is an input vector.

[0062] The three-dimensional linear structure intelligent recognition network introduces a gated attention mechanism, which weights the features obtained in the feature extraction process by learning continuous weights The continuous weights are differentiable matrices, and the end-to-end training can be performed through the back propagation algorithm during the network training process, so that the network can adaptively focus on the feature area related to the linear structure recognition task.

[0063] In this embodiment, the gated attention mechanism is implemented based on a 1x1x1 convolution, and the weights generated thereby are used to select the spatial position of the features obtained in the feature extraction process. The input features and the weights generated by the gated attention mechanism are weighted in proportion, so as to highlight the feature area related to the linear structure. In order to realize the application of the weights in the feature space, a trilinear interpolation method is used to reconstruct the weights to complete the weighting processing of the features, which can significantly improve the segmentation accuracy of fuzzy boundaries or small-scale targets. The principle diagram is shown in Figure 2

[0064] ​In this embodiment of the invention, for the task of processing gravity and magnetic anomaly data, the absolute size of the anomaly value not only reflects the anomaly morphological characteristics but also relates to the physical property distribution of the underground geological body. To avoid the physical property information carried by the numerical value itself being weakened or eliminated during network training, the three-dimensional linear structural intelligent recognition network of this invention does not employ batch normalization, thus ensuring that the network can learn feature information related to the physical property distribution of the underground geological body and improve the structural information extraction capability.

[0065] In this embodiment of the invention, during the training process of the three-dimensional linear construction intelligent recognition network, mean squared error (MES) is used as the loss function. MES is a widely used evaluation index in regression tasks. By comparing the error between the linear construction three-dimensional distribution output by the intelligent recognition network and the preset construction distribution result, specifically by comparing the average of the squared differences between the linear construction three-dimensional distribution output by the intelligent recognition network and the preset construction distribution result, the accuracy of the model prediction is measured, the network parameter update process is constrained, and the network parameters of the intelligent recognition network are optimized.

[0066] The mean squared error (MES) is used as a loss function, and its mathematical expression is as follows:

[0067] ;

[0068] in, This is the mean squared error loss value. For the sample size, For the first One data point, For the true value, These are predicted values.

[0069] In this embodiment of the invention, the data used in the training process of the three-dimensional linear construction intelligent recognition network is divided according to a preset ratio. Specifically, a total of 50,000 datasets are used in the network training process, the ratio of training set to test set is 4:1, the training batch size is 4, and the learning rate is 0.0003.

[0070] Furthermore, the intelligent recognition method for three-dimensional linear structures based on gravity and magnetic data also includes the following steps:

[0071] Step S400: The intelligent recognition network outputs the predicted three-dimensional distribution result of the linear structure in the underground space, wherein the spatial unit with structure is marked as a structure unit and the spatial unit without structure is marked as a non-structure unit. The spatial distribution of the structure unit represents the three-dimensional spatial distribution information of the linear structure.

[0072] In this embodiment of the invention, the three-dimensional linear structure intelligent recognition network outputs the three-dimensional distribution results of linear structures in underground space. Underground subdivision units containing structures are marked as structural units, while those without structures are marked as non-structural units. The structural units are expressed in the form of three-dimensional blocks, and the arrangement of these three-dimensional blocks in space characterizes the three-dimensional spatial distribution information of the linear structures.

[0073] This invention proposes a three-dimensional linear structure intelligent recognition system based on gravity and magnetic data, the system comprising:

[0074] The data acquisition and extension module is used to acquire gravity and magnetic anomaly data of the target area at the original observation height, and to perform extension calculations on the gravity and magnetic anomaly data to obtain gravity and magnetic anomaly planar grid data at multiple observation heights.

[0075] The multi-height data fusion module is used to fuse the gravity and magnetic anomaly plane grid data at the original observation height with the gravity and magnetic anomaly plane grid data at multiple different observation heights in a predetermined order to construct a three-dimensional gravity and magnetic data volume containing multi-height information.

[0076] The intelligent recognition module is used to input the three-dimensional gravity and magnetic data volume into a pre-constructed three-dimensional linear structure intelligent recognition network, and use the intelligent recognition network to perform feature extraction and linear structure prediction on the three-dimensional gravity and magnetic data volume;

[0077] The result output module is used to output the predicted three-dimensional distribution results of linear structures in underground space by the intelligent recognition network. Spatial units with structures are marked as structural units, and spatial units without structures are marked as non-structural units. The spatial distribution of the structural units represents the three-dimensional spatial distribution information of the linear structures.

[0078] Example 1: Constructing a set of gravity field simulation observation data with a single linear structure (e.g.) Figure 3 The model was validated as shown in the figure. The model consists of a fault with a dip angle of 27° and randomly different residual densities on both sides of the fault. Its anomalous changes are gradual, and anomalous abrupt changes can only be clearly observed in the simulated ground gravity anomaly.

[0079] Based on the simulation model, corresponding gravity field observation data are obtained. Then, using the intelligent identification method for three-dimensional linear structures based on gravity and magnetic data described in this invention, the gravity anomaly data is processed to obtain the three-dimensional distribution results of linear structures in underground space. The obtained prediction results are as follows: Figure 4 As shown.

[0080] It can be seen from the prediction result that the structure predicted by the three-dimensional linear structure intelligent recognition network is basically consistent with the simulation model in spatial distribution and geometric shape, and presents as an inclined plate structure, and the strike, dip and other occurrence elements and the extension trend to the deep part are highly consistent with the simulation model.

[0081] wherein, Figure 4 (c) and Figure 4 (d) respectively show the structure projection results of the simulation model and the prediction model on the horizontal plane, and the two are consistent in surface traces, indicating that the method can predict the exposed position of the linear structure. Meanwhile, the prediction result can also reflect the extension depth of the linear structure and its distribution shape in the underground space, thereby providing a three-dimensional prediction basis for subsequent geological interpretation.

[0082] Example 2: A set of magnetic observation anomaly simulation data of single linear structure (as shown in Figure 5 The model is composed of a fault with a dip of 301° and an inclination of 17° and different random magnetic susceptibility on both sides of the fault.

[0083] Based on the model, the corresponding magnetic anomaly observation data are obtained, and the three-dimensional linear structure intelligent recognition method based on gravity and magnetic data is used to process the magnetic anomaly data, so as to obtain the three-dimensional distribution result of the linear structure in the underground space, and the prediction result is as shown in Figure 6

[0084] It can be seen from the prediction result that the structure predicted by the three-dimensional linear structure intelligent recognition network is basically consistent with the simulation model in spatial distribution and geometric shape, and presents as an inclined plate structure, and the strike, dip and other occurrence elements and the extension trend to the deep part are highly consistent with the simulation model, thereby indicating that the method is also applicable to the three-dimensional recognition of linear structure under the condition of magnetic anomaly data.

[0085] Compared with the commonly used boundary recognition method, the method can more clearly show the horizontal boundary of the linear structure, and can predict the three-dimensional spatial distribution characteristics of the linear structure in the underground space. By comprehensively utilizing the characteristics of gravity and magnetic field data at different observation heights, the automatic and quantitative prediction of the linear structure is realized, and the prediction result has good stability, thereby providing technical support in method for the structure interpretation and application of the actual ore concentration area.

[0086] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical solution and the inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.​

Claims

1. A three-dimensional linear structure intelligent recognition method based on gravity and magnetic data, characterized in that, Includes the following steps: Acquire gravity and magnetic anomaly data of the target area at the original observation height, and perform extension calculations on the gravity and magnetic anomaly data to obtain gravity and magnetic anomaly planar grid data at multiple observation heights. The gravity and magnetic anomaly plane grid data at the original observation height are fused with the gravity and magnetic anomaly plane grid data at multiple different observation heights in a predetermined order to construct a three-dimensional gravity and magnetic data volume containing multi-height information; The three-dimensional gravity and magnetic data volume is input into a pre-constructed three-dimensional linear construction intelligent recognition network. This network uses a U-Net network structure as its backbone and includes multiple convolutional layers, pooling layers, upsampling layers, and skip connections. The convolutional layers employ 3×3 convolution operators, the pooling layers employ 3×3 max pooling operators, and the upsampling layers employ 3×3 upsampling operators. A gated attention mechanism is introduced into the three-dimensional linear construction intelligent recognition network, which learns continuous weights with values ​​in the range [0,1] to weight the features obtained during feature extraction. To enhance the feature regions related to linear construction, the gated attention mechanism is implemented based on 1×1×1 convolution, and the spatial location of the features obtained during feature extraction is selected through the weights generated by the gated attention mechanism. During the training process of the three-dimensional linear construction intelligent recognition network, the mean square error is used as the loss function. By comparing the error between the linear construction three-dimensional distribution result output by the intelligent recognition network and the preset construction distribution result, the parameters of the intelligent recognition network are optimized. The intelligent recognition network is used to perform feature extraction and linear construction prediction on the three-dimensional gravity and magnetic data volume. The three-dimensional distribution of linear structures in underground space is predicted by the intelligent recognition network. Spatial units with structures are marked as structural units, and spatial units without structures are marked as non-structural units. The spatial distribution of structural units represents the three-dimensional spatial distribution information of linear structures.

2. The intelligent recognition method for three-dimensional linear structures based on gravity and magnetic data according to claim 1, characterized in that, The steps of acquiring gravity and magnetic anomaly data of the target area at the original observation height, and performing extension calculations on the gravity and magnetic anomaly data to obtain plane grid data of gravity and magnetic anomalies at multiple observation heights include: Acquire gravity and magnetic anomaly data for the target area at the original observation height; When the gravity and magnetic anomaly data is obtained from ground observation, the gravity and magnetic anomaly data is extended upward based on the distance between measurement points to obtain multiple gravity and magnetic anomaly plane grid data that are higher than the original observation height. When the gravity and magnetic anomaly data is obtained from aerial observation, the gravity and magnetic anomaly data is extended downward based on the distance between measurement points to obtain multiple gravity and magnetic anomaly planar grid data below the original observation altitude.

3. The intelligent recognition method for three-dimensional linear structures based on gravity and magnetic data according to claim 1, characterized in that, In the three-dimensional distribution results, the linear structure is expressed in the form of three-dimensional blocks.

4. A three-dimensional linear structure intelligent recognition system based on gravity and magnetic data, characterized in that, The system includes: The data acquisition and extension module is used to acquire gravity and magnetic anomaly data of the target area at the original observation height, and to perform extension calculations on the gravity and magnetic anomaly data to obtain gravity and magnetic anomaly planar grid data at multiple observation heights. The multi-height data fusion module is used to fuse the gravity and magnetic anomaly plane grid data at the original observation height with the gravity and magnetic anomaly plane grid data at multiple different observation heights in a predetermined order to construct a three-dimensional gravity and magnetic data volume containing multi-height information. The intelligent recognition module is used to input the three-dimensional gravity and magnetic data volume into a pre-constructed three-dimensional linear intelligent recognition network. The three-dimensional linear intelligent recognition network uses a U-Net network structure as its backbone, including multiple convolutional layers, pooling layers, upsampling layers, and skip connection structures. The convolutional layers use 3×3 convolution operators, the pooling layers use 3×3 max pooling operators, and the upsampling layers use 3×3 upsampling operators. The three-dimensional linear intelligent recognition network introduces a gated attention mechanism, which learns continuous weights with values ​​in the range [0,1] to apply to the features obtained during feature extraction. Weighting is applied to enhance the feature regions related to linear construction. The gated attention mechanism is based on 1×1×1 convolution, and the spatial location of the features obtained during feature extraction is selected through the weights generated by the gated attention mechanism. During the training of the three-dimensional linear construction intelligent recognition network, mean square error is used as the loss function. By comparing the error between the linear construction three-dimensional distribution result output by the intelligent recognition network and the preset construction distribution result, the parameters of the intelligent recognition network are optimized. The intelligent recognition network is used to perform feature extraction and linear construction prediction on the three-dimensional gravity and magnetic data volume. The result output module is used to output the predicted three-dimensional distribution results of linear structures in underground space by the intelligent recognition network. Spatial units with structures are marked as structural units, and spatial units without structures are marked as non-structural units. The spatial distribution of the structural units represents the three-dimensional spatial distribution information of the linear structures.

Images