Deep learning network structure and method for magnetic anomaly inversion

Through deep learning of network structures, including inversion network encoder and magnetic susceptibility structure decoder, the problems of slow calculation speed and low accuracy of traditional magnetic inversion methods are solved, and efficient and accurate three-dimensional magnetic susceptibility structure inversion is achieved. Improves the accuracy and efficiency of magnetic anomaly inversion.

CN119247489AInactive Publication Date: 2025-01-03CHINA UNIV OF GEOSCIENCES (BEIJING) +1
View PDF 6 Cites 3 Cited by

Patent Information

Application Number
CN202411483786.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-23
Publication Date
2025-01-03
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The traditional magnetic inversion method has slow calculation speed and low accuracy, making it difficult to deal with complex geological conditions, and there are local minimum values ​​problems, resulting in low calculation efficiency and poor stability, and it is difficult to obtain high-precision magnetic susceptibility physical inversion and three-dimensional magnetic susceptibility structure.

Method used

Adopting deep learning network structure, including inversion network encoder, 3D dimension converter and magnetic susceptibility structure decoder, two-dimensional features are extracted through residual blocks and deconvolution blocks and converted into three-dimensional data, and the three-dimensional magnetic susceptibility structure is reconstructed. Train the datasets of diversity and complexity to achieve end-to-end inversion prediction.

Benefits of technology

The efficiency and accuracy of magnetic anomaly inversion have been significantly improved. The deep learning inversion method has achieved accuracy of 97% in the entire region, the accuracy of magnetic structure area has reached 80%, and the calculation time has been reduced by 95%.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119247489A_ABST
    Figure CN119247489A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of magnetic exploration, in particular to a deep learning network structure and method for magnetic anomaly inversion, and solves the problems that a traditional linear inversion method is slow in calculation stable speed and relatively low in precision, and a traditional ground and aeromagnetic data inversion method is insufficient in vertical resolution and relatively low in precision. A traditional linear inversion method is always encountered with the problems of low calculation efficiency and poor stability due to the local minimum value problem, and comprises an inversion network encoder, and a residual block and down-sampling method is adopted to realize the characteristics of magnetic anomaly data to extract an encoding network unit in a two-dimensional form. According to the method, feature extraction of magnetic anomaly data is realized, a residual module is added in an encoder to fuse feature maps from different sources, a three-dimensional magnetic susceptibility structure is reconstructed from two-dimensional observation magnetic field data, and in order to solve the problem that a sample library of deep learning inversion needs enough diversity and complexity, the method has high robustness. The generalization ability of the magnetic method inversion neural network is fully improved, and the inversion precision is further improved.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Magnetic anomaly inversion method and system based on convolutional neural network, terminal and medium

    CN113484919A

  • Three-dimensional magnetic tensor gradient inversion method and device based on convolutional neural network

    CN114879263A

  • Tranform-based magnetotelluric inversion model establishment method

    CN116794736A

  • Multi-feature reconstruction three-dimensional gravity inversion method combining gravity and gradient anomalies

    CN117388940A

  • Large-data-volume gravity and gravity gradient combined rapid forward and reverse modeling method and system

    CN117725354A

Cited By

  • Data field elimination adaptive switching method based on autonomous learning

    CN119884832A

  • Gravity anomalous body inversion method based on PMU-Net deep learning network

    CN120910810A

  • Gravity anomaly body inversion method based on PMU-net deep learning network

    CN120910810B