Accelerating full waveform inversion by transfer learning

被引:0
|
作者
Singh, Divya Shyam [1 ]
Herrmann, Leon [1 ]
Sun, Qing [2 ]
Buerchner, Tim [1 ]
Dietrich, Felix [2 ]
Kollmannsberger, Stefan [3 ]
机构
[1] Tech Univ Munich, Chair Computat Modeling & Simulat, Sch Engn & Design, Arcisstr 21, D-80333 Munich, Germany
[2] Tech Univ Munich, Chair Sci Comp Comp Sci SCCS, Sch Computat Informat & Technol, Boltzmannstr 3, D-85748 Garching, Germany
[3] Bauhaus Univ Weimar, Chair Data Sci Civil Engn, Coudraystr 13b, D-99423 Weimar, Germany
关键词
Transfer learning; Neural network; Deep learning; Adjoint optimization; Full waveform inversion;
D O I
10.1007/s00466-025-02600-w
中图分类号
O1 [数学];
学科分类号
0701 ; 070101 ;
摘要
Full waveform inversion (FWI) is a powerful tool for reconstructing material fields based on sparsely measured data obtained by wave propagation. For specific problems, discretizing the material field with a neural network (NN) improves the robustness and reconstruction quality of the corresponding optimization problem. We call this method NN-based FWI. Starting from an initial guess, the weights of the NN are iteratively updated to fit the simulated wave signals to the sparsely measured dataset. For gradient-based optimization, a suitable choice of the initial guess, i.e., a suitable NN weight initialization, is crucial for fast and robust convergence. In this paper, we introduce a novel transfer learning approach to further improve NN-based FWI. This approach leverages supervised pretraining to provide a better NN weight initialization, leading to faster convergence of the subsequent optimization problem. Moreover, the inversions yield physically more meaningful local minima. The network is pretrained to predict the unknown material field using the gradient information from the first iteration of conventional FWI. The training dataset consists of two-dimensional reference simulations with arbitrarily positioned elliptical voids of different shapes and orientations, mimicking experiments from phased array ultrasonic testing. We compare the performance of the proposed transfer learning NN-based FWI with three other methods: conventional FWI, NN-based FWI without pretraining and conventional FWI with an initial guess predicted from the pretrained NN. Our results show that transfer learning NN-based FWI outperforms the other methods in terms of convergence speed and reconstruction quality.
引用
收藏
页数:22
相关论文
共 50 条
  • [31] Variational full-waveform inversion
    Zhang X.
    Curtis A.
    Geophysical Journal International, 2021, 222 (01): : 406 - 411
  • [32] Full Waveform Inversion in Viscoelastic Media
    Cheverda, Vladimir
    Efimova, Ekaterina
    Reshetova, Galina
    COMPUTATIONAL SCIENCE AND ITS APPLICATIONS, ICCSA 2021, PT II, 2021, 12950 : 322 - 334
  • [33] Full-intensity waveform inversion
    Liu, Yike
    He, Bin
    Lu, Huiyi
    Zhang, Zhendong
    Xie, Xiao-Bi
    Zheng, Yingcai
    GEOPHYSICS, 2018, 83 (06) : R649 - R658
  • [34] Optimal Matching Full Waveform Inversion
    He, Weiguang
    Hu, Guanghui
    Zhang, Bing
    IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 2023, 61
  • [35] Full waveform inversion and Lagrange multipliers
    Gholami, Ali
    Aghazade, Kamal
    GEOPHYSICAL JOURNAL INTERNATIONAL, 2024, 238 (01) : 109 - 131
  • [36] Resolution analysis in full waveform inversion
    Department of Earth Sciences, Utrecht University, Utrecht, Netherlands
    Geophys. J. Int., 3 (1604-1624):
  • [37] Accelerating full-waveform inversion through adaptive gradient optimization methods and dynamic simultaneous sources
    Bernal-Romero, Marcos
    Iturraran-Viveros, Ursula
    GEOPHYSICAL JOURNAL INTERNATIONAL, 2021, 225 (01) : 97 - 126
  • [38] Accelerating 3-D Acoustic Full Waveform Inversion Using a Multi-GPU Cluster
    Chen, Yanling
    Zhu, Pei-Min
    Wen, Wudi
    Jiang, Jinpeng
    IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 2023, 61
  • [39] Simultaneous inversion of full data bandwidth by tomographic full-waveform inversion
    Biondi, Biondo
    Almomin, Ali
    GEOPHYSICS, 2014, 79 (03) : WA129 - WA140
  • [40] Efficient Progressive Transfer Learning for Full-Waveform Inversion With Extrapolated Low-Frequency Reflection Seismic Data
    Jin, Yuchen
    Hu, Wenyi
    Wang, Shirui
    Zi, Yuan
    Wu, Xuqing
    Chen, Jiefu
    IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 2022, 60