卷积神经网络在频域电磁数据反演中的应用
Convolutional Neural Networks for Frequency-Domain Electromagnetic Data Inversion
摘要: 频域电磁感应法作为重要的地球物理勘探技术，在地下结构探测中发挥关键作用。然而，传统频域电磁反演方法存在解的非唯一性问题，容易陷入局部极小值，且计算效率低，难以满足现代勘探对高精度、快速反演的需求。为应对这些挑战，提出了基于双分支卷积神经网络架构的深度学习反演方法。该方法通过构建双分支结构同时处理同相位分量和正交分量，实现多维信息并行提取。通过设计自定义残差模块增强特征提取能力，有效解决深层网络梯度消失问题，并优化空间分辨率重建过程改善梯度传播。在含有80000个样本的合成电磁感应数据集上验证表明，相比传统反演方法，该方法反演相关系数从0.67提升至0.99，均方根误差降低83%，在含噪声环境下展现出良好的鲁棒性和泛化能力。研究结果为频域电磁勘探提供了高效可靠的反演技术。
Abstract: Frequency-domain electromagnetic induction serves as an important geophysical exploration technique, playing a crucial role in subsurface structure detection. However, traditional frequency-domain electromagnetic inversion methods suffer from solution non-uniqueness, tendency to fall into local minima, and low computational efficiency, failing to meet modern exploration demands for high-precision, rapid inversion. To address these challenges, a deep learning inversion method based on dual-branch convolutional neural network architecture is proposed. The method constructs dual-branch structures to simultaneously process in-phase and quadrature components, achieving parallel extraction of multi-dimensional information. Through designing custom residual modules, the network enhances feature extraction capabilities, effectively solving gradient vanishing problems in deep networks, and optimizes spatial resolution reconstruction processes to improve gradient propagation. Validation on a large-scale synthetic electromagnetic induction dataset containing 80,000 samples demonstrates that compared with traditional inversion methods, the proposed method improves inversion correlation coefficient from 0.67 to 0.99, reduces root mean square error by 83%, and exhibits excellent robustness and generalization capability under noisy conditions. The research results provide efficient and reliable inversion technology for frequency-domain electromagnetic exploration.
文章引用：方哲祯, 齐兴, 刘海兵. 卷积神经网络在频域电磁数据反演中的应用[J]. 计算机科学与应用, 2026, 16(1): 242-256. https://doi.org/10.12677/csa.2026.161020

参考文献

[1] Moghadas, D., Jadoon, K.Z. and McCabe, M.F. (2019) Spatiotemporal Monitoring of Soil Moisture from EMI Data Using DCT-Based Bayesian Inference and Neural Network. Journal of Applied Geophysics, 169, 226-238. [Google Scholar] [CrossRef] 
[2] Martinez, G., Huang, J., Vanderlinden, K., Giráldez, J.V. and Triantafilis, J. (2018) Potential to Predict Depth‐Specific Soil-Water Content beneath an Olive Tree Using Electromagnetic Conductivity Imaging. Soil Use and Management, 34, 236-248. [Google Scholar] [CrossRef] 
[3] Moghadas, D., Behroozmand, A.A. and Christiansen, A.V. (2020) Soil Electrical Conductivity Imaging Using a Neural Network-Based Forward Solver: Applied to Large-Scale Bayesian Electromagnetic Inversion. Journal of Applied Geophysics, 176, Article ID: 104012. [Google Scholar] [CrossRef] 
[4] Moghadas, D. and Vrugt, J.A. (2019) The Influence of Geostatistical Prior Modeling on the Solution of DCT-Based Bayesian Inversion: A Case Study from Chicken Creek Catchment. Remote Sensing, 11, Article 1549. [Google Scholar] [CrossRef] 
[5] 陈晗远, 陆佳炜, 王琪冰, 等. 基于短序列时间卷积网络的电梯故障诊断方法[J]. 现代电子技术, 2025, 48(12): 137-147.
[6] Ronneberger, O., Fischer, P. and Brox, T. (2015) U-Net: Convolutional Networks for Biomedical Image Segmentation. In: Navab, N., Hornegger, J., Wells, W. and Frangi, A., Eds., Medical Image Computing and Computer-Assisted Intervention—MICCAI 2015, Springer, 234-241. [Google Scholar] [CrossRef] 
[7] Roy, A.G., Navab, N. and Wachinger, C. (2018) Concurrent Spatial and Channel ‘Squeeze & Excitation’ in Fully Convolutional Networks. In: Frangi, A., Schnabel, J., Davatzikos, C., Alberola-López, C. and Fichtinger, G., Eds., Medical Image Computing and Computer Assisted Intervention—MICCAI 2018, Springer, 421-429. [Google Scholar] [CrossRef] 
[8] Chen, L.C., Papandreou, G., Schroff, F., et al. (2017) Rethinking Atrous Convolution for Semantic Image Segmentation. arXiv: 1706.05587.
[9] Kim, Y. and Nakata, N. (2018) Geophysical Inversion versus Machine Learning in Inverse Problems. The Leading Edge, 37, 894-901. [Google Scholar] [CrossRef] 
[10] Russell, B. (2019) Machine Learning and Geophysical Inversion—A Numerical Study. The Leading Edge, 38, 512-519. [Google Scholar] [CrossRef] 
[11] Das, V., Pollack, A., Wollner, U. and Mukerji, T. (2019) Convolutional Neural Network for Seismic Impedance Inversion. Geophysics, 84, R869-R880. [Google Scholar] [CrossRef] 
[12] Liu, W., Lu, Q., Yang, L., Lin, P. and Wang, Z. (2021) Application of Sample-Compressed Neural Network and Adaptive-Clustering Algorithm for Magnetotelluric Inverse Modeling. IEEE Geoscience and Remote Sensing Letters, 18, 1540-1544. [Google Scholar] [CrossRef] 
[13] Guo, R., Li, M., Yang, F., Xu, S. and Abubakar, A. (2020) Application of Supervised Descent Method for 2D Magnetotelluric Data Inversion. Geophysics, 85, WA53-WA65. [Google Scholar] [CrossRef] 
[14] Moghadas, D. (2020) One-dimensional Deep Learning Inversion of Electromagnetic Induction Data Using Convolutional Neural Network. Geophysical Journal International, 222, 247-259. [Google Scholar] [CrossRef] 
[15] Puzyrev, V. (2019) Deep Learning Electromagnetic Inversion with Convolutional Neural Networks. Geophysical Journal International, 218, 817-832. [Google Scholar] [CrossRef] 
[16] Puzyrev, V. and Swidinsky, A. (2021) Inversion of 1D Frequency-and Time-Domain Electromagnetic Data with Convolutional Neural Networks. Computers & Geosciences, 149, Article ID: 104681. [Google Scholar] [CrossRef] 
[17] Wait, J.R. (1955) Mutual Electromagnetic Coupling of Loops over a Homogeneous Ground. Geophysics, 20, 630-637. [Google Scholar] [CrossRef] 
[18] Frischknecht, F.C. (1967) Fields about an Oscillating Magnetic Dipole over a Two-Layer Earth, and Application to Ground and Air-Borne Electromagnetic Surveys. Quarterly of the Colorado School of Mines, 62, 1-326.
[19] Wait, J.R. (1962) A Note on the Electromagnetic Response of a Stratified Earth. Geophysics, 27, 382-385. [Google Scholar] [CrossRef] 
[20] Ryu, J., Morrison, H.F. and Ward, S.H. (1970) Electromagnetic Fields about a Loop Source of Current. Geophysics, 35, 862-896. [Google Scholar] [CrossRef] 
[21] Farquharson, C.G. and Oldenburg, D.W. (1996) Approximate Sensitivities for the Electromagnetic Inverse Problem. Geophysical Journal International, 126, 235-252. [Google Scholar] [CrossRef] 
[22] Guptasarma, D. and Singh, B. (1997) New Digital Linear Filters for Hankel J0 and J1 Transforms. Geophysical Prospecting, 45, 745-762. [Google Scholar] [CrossRef] 
[23] Anderson, W.L. (1982) Fast Hankel Transforms Using Related and Lagged Convolutions. ACM Transactions on Mathematical Software, 8, 344-368. [Google Scholar] [CrossRef] 
[24] Ward, S.H. and Hohmann, G.W. (1988) 4. Electromagnetic Theory for Geophysical Applications. In: Nabighian, M.N. and Corbett, J.D., Eds., Electromagnetic Methods in Applied Geophysics, Society of Exploration Geophysicists, 130-311. [Google Scholar] [CrossRef] 
[25] Li, G. and Li, Y. (2017) Joint Inversion for Transmitter Navigation and Seafloor Resistivity for Frequency-Domain Marine CSEM Data. Journal of Applied Geophysics, 136, 178-189. [Google Scholar] [CrossRef] 
[26] Hanssens, D., Delefortrie, S., De Pue, J., Van Meirvenne, M. and De Smedt, P. (2019) Frequency-domain Electromagnetic Forward and Sensitivity Modeling: Practical Aspects of Modeling a Magnetic Dipole in a Multilayered Half-space. IEEE Geoscience and Remote Sensing Magazine, 7, 74-85. [Google Scholar] [CrossRef] 
[27] Kingma, D.P. and Ba, J. (2014) Adam: A Method for Stochastic Optimization. arXiv: 1412.6980.
[28] Aster, R.C., Borchers, B. and Thurber, C.H. (2018) Parameter Estimation and Inverse Problems. Elsevier.
[29] Craven, J.A., McNeice, G., Powell, B., et al. (2007) Audio-Magnetotelluric Studies at the McArthur River Mining Camp and Shea Creek Area, Northern Saskatchewan. In: Jefferson, C.W. and Delaney, G., Eds., Geology and Uranium Exploration Technology of the Protoproterozoic Athabasca Basin, Saskatchewan and Alberta, Geological Survey of Canada, 413-424.
[30] Liu, X., Craven, J.A. and Tschirhart, V. (2023) Retrieval of Subsurface Resistivity from Magnetotelluric Data Using a Deep-Learning-Based Inversion Technique. Minerals, 13, Article 461. [Google Scholar] [CrossRef] 
[31] Simpson, J.M. and Heinson, G. (2020) Synthetic Modelling of Downhole Resistivity Data to Improve Interpretation of Basin Morphology from Magnetotelluric Inversion. Earth, Planets and Space, 72, Article No. 69. [Google Scholar] [CrossRef]