CN-122020467-A - Method for identifying underground density abnormal body top interface based on convolutional neural network

CN122020467ACN 122020467 ACN122020467 ACN 122020467ACN-122020467-A

Abstract

The application discloses a method for identifying a top interface of an underground density abnormal body based on a convolutional neural network, which comprises the following steps of S1, obtaining measured gravity data, S2, performing gridding and normalization processing on the measured gravity data to obtain preprocessed gravity data, and S3, inputting the preprocessed gravity data into a convolutional neural network model trained in advance, and outputting predicted depth distribution of the top interface of the underground density abnormal body. The scheme realizes the rapid and high-precision identification of the top interface of the abnormal underground density body.

Inventors

Shan Xipeng
ZHAO TINGYAN
HAN SONG
LI SHIJUN
LIU WEI
YU YUNPENG

Assignees

中国自然资源航空物探遥感中心

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (10)

1. The method for identifying the underground density abnormal body top interface based on the convolutional neural network is characterized by comprising the following steps of: s1, acquiring actual measurement gravity data; S2, performing gridding and normalization processing on the actually measured gravity data to obtain preprocessed gravity data; And S3, inputting the preprocessed gravity data into a pre-trained convolutional neural network model, and outputting the predicted top interface depth distribution of the underground density abnormal body.
2. The method for identifying the top interface of the underground density anomaly body based on the convolutional neural network as set forth in claim 1, wherein the construction and training of the convolutional neural network model are carried out by the following steps: Determining a data set for training and testing, wherein the data set comprises a gravity anomaly data sample and a corresponding real underground density anomaly body top interface depth label thereof; constructing a convolutional neural network model for interface identification; and performing supervised training on the convolutional neural network model by using the data set to establish a nonlinear mapping relation between the gravity anomaly data and the top interface depth, so as to obtain a trained convolutional neural network model.
3. The method for identifying a subsurface density anomaly body top interface based on a convolutional neural network of claim 2, wherein the determining of the dataset comprises: Constructing an underground density abnormal body geologic model set, wherein the underground density abnormal body geologic model set comprises geologic models with various shapes, sizes and burial depths; For each geological model, calculating corresponding gravity anomaly data by utilizing a gravity forward formula; Determining tag data of the gravity anomaly data according to the top interface depth corresponding to each geological model; And preprocessing the gravity anomaly data and the tag data to obtain the data set.
4. The method for identifying an abnormal body top interface of an underground density based on a convolutional neural network according to claim 3, wherein the preprocessing the gravity anomaly data and the tag data to obtain the data set comprises: Normalizing the gravity anomaly data and the tag data to a preset interval by using a normalization method to obtain normalized data; And randomly adding Gaussian white noise into the normalized data to obtain the data set.
5. The method for identifying the underground density anomaly body top interface based on the convolutional neural network according to claim 2, wherein the convolutional neural network model comprises: an input layer for receiving input data; An encoder for extracting multi-scale spatial features of the input data through a multi-layer convolution and pooling operation; A bottleneck layer for processing the highest dimensional characteristics output by the encoder; the decoder is used for restoring the features to the original space size and fusing the shallow features of the corresponding level of the encoder by using jump connection; and the output layer is used for outputting a depth matrix consistent with the input data size.
6. The method for identifying the top interface of the underground density anomaly body based on the convolutional neural network of claim 5, wherein the encoder comprises three convolutional modules which are connected in sequence, and a pooling layer is connected behind each convolutional module; The first convolution module comprises two continuous first convolution layers, wherein each first convolution layer is connected with a first batch normalization layer and a first ReLU activation function, and then connected with a first pooling layer; The second convolution module comprises two continuous second convolution layers, each second convolution layer is connected with a second batch normalization layer and a second ReLU activation function, and then connected with a second pooling layer; The third convolution module comprises two continuous third convolution layers, each of which is followed by a third batch normalization layer and a third ReLU activation function, followed by a third pooling layer.
7. The method for identifying an abnormal body top interface of an underground density based on a convolutional neural network as set forth in claim 6, wherein the bottleneck layer comprises two continuous fourth convolutional layers, a dropoff layer is introduced between the two fourth convolutional layers, and the drop rate of the dropoff layer is 0.5.
8. The method for identifying the abnormal body top interface of the underground density based on the convolutional neural network according to claim 2, wherein when the convolutional neural network model is supervised and trained by using the data set, model parameter optimization is performed by using an Adam optimizer, and a mean square error is used as a loss function.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method for identifying a convolutional neural network-based subsurface density anomaly roof interface as defined in any one of claims 1-8 when the program is executed.
10. A readable storage medium having stored thereon a computer program, which when executed by a processor, implements a method for identifying a subsurface density anomaly roof interface based on a convolutional neural network as claimed in any one of claims 1-8.

Description

Method for identifying underground density abnormal body top interface based on convolutional neural network Technical Field The invention relates to the technical field of geophysical exploration, in particular to a method for identifying an underground density abnormal body top interface based on a convolutional neural network. Background Gravity exploration is an important geophysical method for studying subsurface geologic structures by observing surface gravity anomalies. The depth of the top interface of the underground density anomaly is determined, and the method has important significance for mineral resource evaluation and engineering geological disaster prevention. Traditional gravity inversion methods rely mainly on iterative linearization methods or global optimization algorithms (e.g., genetic algorithms, simulated annealing, etc.). These methods suffer from the major drawbacks of 1) the resulting multi-resolvability. Different subsurface density distributions may produce similar surface gravity anomalies, resulting in unstable inversion results. 2) Depending on the initial model. The linear inversion method extremely depends on the accuracy of an initial model, and if the initial guess deviation is large, local minima are easy to sink. 3) The resulting boundary is blurred. Conventional regularization methods tend to produce smooth physical distributions that make it difficult to clearly delineate sharp boundaries of abnormal volumes. 4) The calculation cost is high. Especially for large-scale three-dimensional data, the forward and backward iterative computation takes very long time. In recent years, deep learning has achieved great success in the field of image recognition. But systematically applying deep learning techniques to gravity exploration, particularly for end-to-end fast inversion of a specific target of subsurface density anomaly top interface depths, remains a challenge. Disclosure of Invention The invention aims to provide a method for identifying an underground density abnormal body top interface based on a convolutional neural network. To achieve the above object, an embodiment of the present invention is achieved by: in a first aspect, the invention provides a method for identifying an underground density abnormal body top interface based on a convolutional neural network, which comprises the following steps: s1, acquiring actual measurement gravity data; S2, performing gridding and normalization processing on the actually measured gravity data to obtain preprocessed gravity data; And S3, inputting the preprocessed gravity data into a pre-trained convolutional neural network model, and outputting the predicted top interface depth distribution of the underground density abnormal body. In one embodiment of the present invention, the convolutional neural network model is constructed and trained by the following steps: Determining a data set for training and testing, wherein the data set comprises a gravity anomaly data sample and a corresponding real underground density anomaly body top interface depth label thereof; constructing a convolutional neural network model for interface identification; and performing supervised training on the convolutional neural network model by using the data set to establish a nonlinear mapping relation between the gravity anomaly data and the top interface depth, so as to obtain a trained convolutional neural network model. In one embodiment of the invention, the determination of the data set comprises: Constructing an underground density abnormal body geologic model set, wherein the underground density abnormal body geologic model set comprises geologic models with various shapes, sizes and burial depths; For each geological model, calculating corresponding gravity anomaly data by utilizing a gravity forward formula; Determining tag data of the gravity anomaly data according to the top interface depth corresponding to each geological model; And preprocessing the gravity anomaly data and the tag data to obtain the data set. In one embodiment of the present invention, preprocessing the gravity anomaly data and the tag data to obtain the data set includes: Normalizing the gravity anomaly data and the tag data to a preset interval by using a normalization method to obtain normalized data; And randomly adding Gaussian white noise into the normalized data to obtain the data set. In one embodiment of the present invention, a convolutional neural network model includes: an input layer for receiving input data; An encoder for extracting multi-scale spatial features of the input data through a multi-layer convolution and pooling operation; A bottleneck layer for processing the highest dimensional characteristics output by the encoder; the decoder is used for restoring the features to the original space size and fusing the shallow features of the corresponding level of the encoder by using jump connection; and the output layer is used for outputting a dept