Transfer learning seismic impedance inversion method based on temporal convolutional networks

Wang, Z.F., Cao, H., Yang, Z.F., Xu, H.Q., Yang, M.Q. and Zhao, Y., 2022. Transfer learning seismic impedance inversion method based on temporal convolutional networks. Journal of Seismic Exploration, 31: 391-405. The nonlinear mapping between seismic data and impedance can be established by Temporal Convolutional Networks (TCN), which has been proved by forward modeling data. However, whether the deep neural network can be used to train an inversion mapping model with good generalization ability under a small number of labeled samples remains to be explored. In view of this, the noise analysis of the TCN seismic impedance inversion method was firstly carried out, and the model test showed that the TCN seismic impedance inversion method had certain noise resistance. Secondly, an inversion mapping model was obtained based on the training of Marmousi-2 data set, and then five traces of Overthrust model samples were added for fine-tuning to obtain a new inversion mapping model. The inversion of Overthrust was performed based on the TCN transfer learning inversion mapping model. And the model test results showed that: with a small number of labels, the inversion results of the Overthrust dataset based on TCN transfer learning are higher than the Pearson sum determination results obtained by TCN inversion, and the error profile is relatively small compared to the true impedance. Furthermore, TCN transfer learning method ,which was effectively proved in adjacent blocks of the actual data, compared with the result of the TCN inversion. Therefore, the introduction of transfer learning in TCN seismic impedance inversion can improve the generalization ability of the inversion mapping model trained with a few labeled samples in practical application.
- Alfarraj, M. and Alregib, G., 2019. Semi-supervised learning for acoustic impedanceinversion. Expanded Abstr., 89th Ann. Internat. SEG Mtg., San Antonio.doi: 10.1190/segam2019-3215902.1.
- Bai, S.J., Kolter, J.Z. and Koltun, V., 2018. An empirical evaluation of genericconvolutional and recurrent networks for sequence modeling. arxiv:1803.01271.
- Cao, L.L., Li, H.T., Han, Y.S., Yu, F., Gu, H.Y., 2016. Application of convolutional neuralnetworks in classification of high resolution remote sensing imagery. Sci.
- Survey.mapp., 41(9): 170-175. doi:10.16251/j.cnki.1009-2307.2016.09.033.
- Das, V., Pollack, A., Wollner, U. and Mukerji, T., 2019. Convolutional neural network forseismic impedance inversion. Geophysics, 84(6): R869-R880.
- Guo, R., Zhang, J.J., Liu, D., Zhang, Y.B. and Zhang, D.W., 2019. Application of bi-directional long short-term memory recurrent neural network for seismic impedanceinversion. Extended Abstr., 81st EAGE Conf., London: 3-6.doi: 10.3997/2214-4609.201901386.
- He, Q. and Wang, Y., 2020. Reparameterized full waveform inversion using deep neuralnetworks. Geophysics, 86(1): 1-71.
- Hochreiter, S. and Schmidhuber, J., 1997. Long short-term memory. Neural Computat.,9(8). doi:10.1162/neco.1997.9.8.1735.
- Hou, Y.M., Zhou, H.M. and Wang, Z.Y., 2017. Overview of speech recognition based ondeep learning. Applict. Res. Comput., 34: 2241-2246.doi: 10.3969/j.issn.1001-3695.2017.08.001.
- Huang, X.R., Dai, Y., Xu, Y.G. and Tang, Y., 2020. Seismic inversion experiments basedon different datasets of deep learning algorithm. J. SW Petrol. Univ. (Nat. Sci. Ed.),42(6): 16-25.
- Lecun, Y. and Bottou, L., 1998. Gradient-based learning applied to document recognition.Proc. IEEE Conf., 86: 2278-2324.
- Li, S., Liu, B., Ren, Y., Chen, Y. and Jiang, P., 2020. Deep-learning inversion of seismic
- Data. IEEE Transactions on Geoscience and Remote Sensing, 57: 2135-2149.
- Liu, L.N,, Liu, H. and Li, Y.M., 2004.Wave-equation 3-D prestack depth migration for the
- SEG/EAGE salt and overthrust model. Chin. J. Geophys, 47(2): 312-320.
- Long, J., Shelhamer, E. and Darrell, T., 2015. Fully convolutional networks for semanticsegmentation. IEEE Transact. Patt. Analys. Mach. Intellig. 39: 640-651.
- Lu, H.T. and Zhang, Q.C., 2016. Applications of deep convolutional neural network incomputer vision. J. Acquis. Proces., 31: 1-17. doi:10.16337/j.1004-9037.2016.01.001.
- Martin, G.S., Wiley, R. and Marfurt, K.J., 2006. Marmousi-2: An elastic upgrade forMarmousi. The Leading Edge, 25:156-166.
- Mustafa, A., Alfarraj, M. and Alregib, G., 2019. Estimation of acoustic impedance fromseismic data using temporal convolutional network. Extended Abstr., 81st EAGEConf., London. arxiv:1906.02684.
- Pascanu, R., Gulcehre, C., Cho, K. and Bengio, Y., 2013. How to construct deep recurrentneural networks. Comput. Sci. arxiv:1312.6026v1.
- Richardson, A. and Feller, C., 2019. Seismic data denoising and deblending using deeplearning. arxiv:1907.01497v1.
- Shao, R.B., Xiao, L.Z., Liao, G.Z., Zhou, J. and Li, G.J., 2022. A reservoir parametersprediction method for geophysical logs based on transfer learning. Chin. J. Geophys.(in Chinese), 65(2):796-808. doi:10.6038/cjg2022P0057.
- Song, H., Mao, WJ. and Tang, H.H., 2021. Application of deep neural networks formultiples attenuation. Chin. J. Geophys. (in Chinese), 64: 2795-2808.
- Tzeng, E., Hoffman, J., Zhang, N., Saenko, K. and Darrell, T., 2014. Deep domainconfusion: Maximizing for domain invariance. Comput. Sci. arxiv:1412.3474v1.
- Tzeng, E., Hoffman, J., Saenko, K. and Darrell, T., 2017. Simultaneous deep transferacross domains and tasks, 2015. IEEE Internat. Conf. Comput. Vis. (ICCV). IEEE.arxiv.org/abs/1510.02192.
- Wu, B., Meng, D., Wang, L., Liu, N. and Wang, Y., 2020. Seismic impedance inversionusing fully convolutional residual network and transfer learning. IEEE Geosci.Remote Sens. Lett., 99:1-5.
- Wu, B., Meng, D. and Zhao, H., 2021. Semi-supervised learning for seismic impedance inversion using generative adversarial networks. Remote Sens., 13: 909.doi:10.3390/rs13050909.
- Yosinski, J., Clune, J., Bengio, Y. and Lipson, H., 2014. How transferable are features indeep neural networks? CoRR, arxiv:1411.1792v1.
- Zaremba, W., Sutskever, I. and Vinyals, O., 2014. Recurrent Neural Network Regularization. Eprint Arxiv, arxiv:1409.2329.
- Zhang, Y.L., Yu, Z.C., Hu, T.Y. and He, C., 2021. Multi-trace joint downhole microseismicphase detection and arrival picking method. Chin. J. Geophys. (in Chinese), 64: 2073-2085. doi:10.6038/cjg202 100379.
- Zhuang, F.Z., Luo, P., He, Q.and Shi, Z.Z., 2015. Survey on transfer learning research.
- Ruan Jian Xue Bao/Journal of Software,26(1):26-39.