[논문]심층 신경망을 이용한 탄성파 속도 모델 구축 사례 분석

조준현; 하완수

doi:10.7582/gge.2021.24.2.53

심층 신경망을 이용한 탄성파 속도 모델 구축 사례 분석
Case Analysis of Seismic Velocity Model Building using Deep Neural Networks 원문보기

지구물리와 물리탐사 = Geophysics and geophysical exploration, v.24 no.2, 2021년, pp.53 - 66

조준현 (부경대학교 에너지자원공학과) , 하완수 (부경대학교 에너지자원공학과)

초록
AI-Helper

속도 모델 구축은 탄성파 탐사 자료처리에서 필수적인 절차이다. 주시 토모그래피나 속도 분석과 같은 기존 기법들은 하나의 속도 모델을 예측하는 데 계산 시간이 오래 걸리며 역산 결과의 품질이 전문가의 판단에 크게 의존한다. 전파형 역산 또한 초기 속도 모델에 크게 의존한다는 문제가 있다. 최근 심층 신경망 기법이 복잡하고 비선형적인 문제를 푸는데 적용되는 사례가 많아지면서 널리 보급되고 있다. 이 논문에서는 심층 신경망 기법을 이용한 탄성파 속도 모델 구축 사례들을 각 연구에 사용한 신경망에 따라 분류하며 조사하였다. 또한 훈련용 인공 속도 모델 생성 사례도 포함하였다. 심층 신경망은 대량의 데이터로부터 신경망을 훈련함으로써 모델 매개변수를 자동으로 최적화한다. 따라서 기존 기법들에 비해 역산 결과에 사람의 판단이 개입될 여지가 적으며 훈련을 마친 후 하나의 속도 모델을 예측하는 비용은 무시할 수 있다. 또한, 심층 신경망은 전파형 역산과 달리 초기 속도 모델이 필요하지 않다. 여러 연구에서 계산 비용뿐만 아니라 역산 결과에서도 심층 신경망 기법이 뛰어난 성과를 달성하는 것을 보여주었다. 연구 결과들을 바탕으로 속도 모델 구축에 사용된 심층 신경망 기법의 특징에 대해 분석하고 논의하였다.

Abstract ▼ AI-Helper

Velocity model building is an essential procedure in seismic data processing. Conventional techniques, such as traveltime tomography or velocity analysis take longer computational time to predict a single velocity model and the quality of the inversion results is highly dependent on human expertise. Full-waveform inversions also depend on an accurate initial model. Recently, deep neural network techniques are gaining widespread acceptance due to an increase in their integration to solving complex and nonlinear problems. This study investigated cases of seismic velocity model building using deep neural network techniques by classifying items according to the neural networks used in each study. We also included cases of generating training synthetic velocity models. Deep neural networks automatically optimize model parameters by training neural networks from large amounts of data. Thus, less human interaction is involved in the quality of the inversion results compared to that of conventional techniques and the computational cost of predicting a single velocity model after training is negligible. Additionally, unlike full-waveform inversions, the initial velocity model is not required. Several studies have demonstrated that deep neural network techniques achieve outstanding performance not only in computational cost but also in inversion results. Based on the research results, we analyzed and discussed the characteristics of deep neural network techniques for building velocity models.

주제어

표/그림 (13)

그림 Fig. 1. The workflow of the DNN-based velocity model inversion algorithm. Through the velocity model provided, we obtain the corresponding seismogram by performing a wave equation simulation. In training the designed network, seismic data is employed as input data and the velocity model is used as a label. After training, given the new seismic data, the network can directly predict the velocity model (Liu et al., 2021).
표 Table 1. Classification of the deep neural network reviewed in this paper (continued).
그림 Fig. 2. Prediction results of velocity model with four- to seven-layered. Column (a) is a label, column (b) is GeoDNN output (prediction), column (c) is the segmented prediction from column (b) using k-means and the correct number of layers, and column (d) is the segmented prediction from column (b) using k-means and eight layers (Araya-Polo et al., 2018).
표 Table 1. Classification of the deep neural network reviewed in this paper (continued).
그림 Fig. 3. The workflow of label generation through GANs. The workflow comprises three main tasks: unsupervised label generation, data generation, convolutional network training (Araya-Polo et al., 2019).
그림 Fig. 4. The VMB-Net architecture with multi-shot seismogram as input data. The network consists of two modules: the first module is a 3-fully connected layer (black-boxed) that takes in one common-source gather and transform it to make the same dimension as the velocity models. The second module is the fully convolutional neural networks (red boxed) that predict velocity models from the transformed seismic data (modified from Wang and Ma, 2019).
그림 Fig. 5. Prediction results of velocity models from (a-c) the Marmousi model, (d-f) the Overthrust model, and (g-i) the Pluto model for testing, and in each subfigure, the label is on the left, and the prediction using VMB-Net is on the right (Wang and Ma, 2020).
그림 Fig. 6. An illustration of InversionNet architecture. This deep neural network consists of an encoder-decoder structure. The encoder (the top pipeline) extracts high-level features from the seismogram and compresses them into a single high-dimensional vector. The decoder (the bottom pipeline) then translates those features into velocity models (Wu et al., 2018).
그림 Fig. 7. Prediction results of the velocity models (a-d), and the label (e). The physics-driven methods (a-b) yield inversion results with a large number of artifacts. The CNN model (c) shows better predictions than physics-driven methods. Furthermore, the CNN-CRF model (d) outputs the most accurate predictions with some artifacts on boundaries (Wu and Lin, 2019).
그림 Fig. 8. An illustration of improved SeisInvNet architecture. The improved network extracts the characteristics of seismic data from two aspects: common-shot and common-receiver gathers (Liu et al., 2021).
그림 Fig. 9. Real marine seismic data: (a) predicted RMS velocity by the proposed network (CNN + RNN), (b) predicted interval velocity by the network, (c) stacked section using the predicted velocity, and (d) stacked section provided by USGS for seismic line 32 (Fabien-Ouellet and Sarkar, 2020).
그림 Fig. 10. An illustration of modified U-Net architecture. Each blue and green cube corresponds to a multi-channel feature map. The symbol is described in the black rectangular box (Yang and Ma, 2019).
그림 Fig. 11. Robustness of the modified U-Net to noise and increased amplitude: (a-c) show prediction results with noisy data and (d-f) prediction results with increased seismic amplitude (Yang and Ma, 2019).

참고문헌 (36)

Adler, A., Araya-Polo, M., and Poggio, T., 2019, Deep recurrent architectures for seismic tomography, Conference Proceedings, 81st EAGE Conference and Exhibition 2019, 1-5, doi: 10.3997/2214-4609.201901512.
Aminzadeh, F., Burkhard, N., Nicoletis, L., Rocca, F., and Wyatt, K., 1994, SEG/EAEG 3-D modeling project: 2nd update, Lead. edge., 13(9), 949-952, doi: 10.1190/1.1437054.

상세보기
Araya-Polo, M., Ferris, S., and Florez, M., 2019, Deep learning-driven velocity model building workflow, Lead. edge., 38(11), 872a1-872a9, doi: 10.1190/tle38110872a1.1.

상세보기
Araya-Polo, M., Jennings J., Adler A., and Dahlke T., 2018, Deep-learning tomography, Lead. edge., 37(1), 58-66, doi: 10.1190/tle37010058.1.

상세보기
Baysal, E., Kosloff, D., and Sherwood, W., 1983, Reverse time migration, Geophysics, 48(11), 1514-1524, doi: 10.1190/1.1441434.

상세보기
Biswas, R., Vassiliou, A., Stromberg, R., and Sen, M. K., 2018, Stacking velocity estimation using recurrent neural network, SEG Technical Program Expanded Abstracts, 2241-2245, doi: 10.1190/segam2018-2997208.1.
Bochen, W., and Zhenwei, G., 2020, Stochastic quasi-real seismic velocity models building for the offshore oil and gas exploration, SEG 2020 Workshop: Broadband and Wide-azimuth Deepwater Seismic Technology, Beijing, China, 88-94, doi: 10.1190/bwds2020_24.1.
Chen, L.-C., Papandreou, G., Kokkinos, I., Murphy, K., and Yuille, A. L., 2017, Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs, IEEE Trans. Pattern Anal. Mach. Intell., 40(4), 834-848, doi: 10.1109/TPAMI. 2017.2699184.

상세보기
Fabien-Ouellet, G., and Sarkar, R., 2020, Seismic velocity estimation; a deep recurrent neural-network approach, Geophysics, 85(1), U21-U29, doi: 10.1190/geo2018-0786.1.

상세보기
Geron, A., 2017, Hands-On Machine Learning with Scikit-Learn and Tensorflow, O'Reilly Media, 55-61.
Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y., 2014, Generative adversarial nets, Adv. Neural. Inf. Process. Syst., 2672-2680.
Hornik, K., Stinchcombe, M., and White, H., 1989, Multilayer feedforward networks are universal approximators, Neural Netw., 2(5), 359-366, doi: 10.1016/0893-6080(89)90020-8.

상세보기
Li, S., Liu, B., Ren, Y., Chen, Y., Yang, S., Wang, Y., and Jiang, Peng., 2020, Deep-learning inversion of seismic data, IEEE trans. Geosci. Remote Sensing, 58(3), 2135-2149, doi: 10.1109/TGRS.2019.2953473.

상세보기
Liu, B., Yang, S., Ren, Y., Xu, X., Jiang, P., and Chen, Y., 2021, Deep-learning seismic full-waveform inversion for realistic structural models, Geophysics, 86, R31-R44, doi: 10.1190/geo2019-0435.1.

상세보기
Liu, B., Yang, S., Xu, X., Ren, Y., and Jiang, P., 2020, Deep learning inversion of seismic data in tunnels, SEG 2019 Workshop: Mathematical Geophysics: Traditional vs Learning, Beijing, China, 5-7, doi: 10.1190/iwmg2019_13.1.
Liu, Z., and Bleistein, N., 1995, Migration velocity analysis: Theory and an iterative algorithm, Geophysics, 60(1), 142-153, doi: 10.1190/1.1443741.

상세보기
Long, J., Shelhamer, E., and Darrell, T., 2015, Fully convolutional networks for semantic segmentation, Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit., 3431-3440.
Nath, S. K., Chakroborty, S., Singh, S. K., and Ganguly, N., 1999, Velocity inversion in cross-hole seismic tomography by counterpropagation neural network, genetic algorithm and evolutionary programming techniques, Geophys. J. Int., 138(1), 108-124, doi: 10.1046/j.1365-246x.1999.00835.x.

상세보기
Ren, Y., Nie, L., Yang, S., Jiang, P., and Chen, Y., 2021, Building complex seismic velocity models for deep learning inversion, IEEE Access, 9, 63767-63778, doi: 10.1109/ACCESS.2021.3051159.

상세보기
Ronneberger, O., Fischer, P., and Brox, T., 2015, U-Net: Convolutional networks for biomedical image segmentation, International Conference on Medical Image Computing and Computer-Assisted Intervention, 234-241, doi: 10.1007/978-3-319-24574-4_28.
Roth, G., and Tarantola, A., 1994, Neural networks and inversion of seismic data, J. Geophys. Res., 99(B4), 6753-6768, doi: 10.1029/93JB01563.

상세보기
Stoughton, D., Stefani, J., and Michell, S., 2001, 2D elastic model for wavefield investigations of subsalt objectives, deep water gulf of Mexico, SEG Technical Program Expanded Abstracts, 1269-1272, doi: 10.1190/1.1816325.
Tarantola, A., 1984, Inversion of seismic reflection data in the acoustic approximation, Geophysics, 49, 1259-1266, doi: 10.1190/1.1441754.

상세보기
Versteeg, R., 1994, The Marmousi experience: Velocity model determination on a synthetic complex data set, Lead. edge., 13(9), 927-936, doi: 10.1190/1.1437051.

상세보기
Virieux, J., and Operto, S., 2009, An overview of full-waveform inversion in exploration geophysics, Geophysics, 74(6), WCC1-WCC26, doi: 10.1190/1.3238367.

상세보기
Wang, W., and Ma, J., 2019, VMB-net: A deep learning network for velocty model building in a cross-well acquisition geometry, SEG Technical Program Expanded Abstracts, 2569-2573, doi: 10.1190/segam2019-3216078.1.
Wang, W., and Ma, J., 2020, Velocity model building in a crosswell acquisition geometry with image-trained artificial neural networks, Geophysics, 85(2), U31-U46, doi: 10.1190/geo2018-0591.1.

상세보기
Wang, W., Yang, F., and Ma, J., 2018, Velocity model building with a modified fully convolutional network, SEG Technical Program Expanded Abstracts, 2086-2090, doi: 10.1190/segam2018-2997566.1.
Wu, X., Geng, Z., Shi, Y., Pham, N., Fomel, S., and Caumon, G., 2020, Building realistic structure models to train convolutional neural networks for seismic structural interpretation, Geophysics, 85(4), WA27-WA39, doi: 10.1190/geo2019-0375.1.

상세보기
Wu, X., Liang, L., Shi, Y., and Fomel, S., 2019, FaultSeg3D: Using synthetic datasets to train an end-to-end convolutional neural network for 3D seismic fault segmentation, Geophysics, 84(3), IM35-IM45, doi: 10.1190/geo2018-0646.1.

상세보기
Wu, Y., and Lin, Y., 2019, InversionNet: A real-time and accurate full waveform inversion with CNNs and continuous CRFs, IEEE Trans. Comput. Imaging, 6, 419-433, doi: 10.1109/TCI.2019.2956866.

상세보기
Wu, Y., Lin, Y., and Zhou, Z., 2018, Inversionet: Accurate and efficient seismic-waveform inversion with convolutional neural networks, SEG Technical Program Expanded Abstracts, 2096-2100, doi: 10.1190/segam2018-2998603.1.
Yang, F., and Ma, J., 2019, Deep-learning inversion: A next-generation seismic velocity model building method, Geophysics, 84(4), R583-R599, doi: 10.1190/geo2018-0249.1.

상세보기
Yilmaz, O., 2001, Seismic data analysis: processing, inversion and interpretation of seismic data, Vol. I, Soc. Expl. Geophys., 288-323.
Zhang, J., Brink, U. S., and Toksoz, M. N., 1998, Nonlinear refraction and reflection travel time tomography, J. Geophys. Res. Solid Earth, 103(B12), 29743-29757, doi: 10.1029/98JB01981.

상세보기
Zhang, Z., and Lin, Y., 2020, Data-driven seismic waveform inversion: A study on the robustness and generalization, IEEE Trans. Geosci. Remote Sens., 58(10), 6900-6913, doi: 10.1109/TGRS.2020.2977635.

상세보기

저자의 다른 논문 :

표제어: PCR

동의어: Packet Collision Rate

용어 설명 출처 목록 (6)

용어 설명: PCR은 세균 특이성이 있는 primer를 이용하여 적은 수의 세균이 있을지라도 쉽게 검출할 수 있는 유용한 방법이며, 이를 이용하여 구강 내 치면세균막이나 타액에서 직접 세균을 검출할 수 있게 되었다[8].

내보내기 구분	파일저장 인쇄 메일전송
구성항목	기본정보 상세정보 관리번호, 논문명, 저널/프로시딩명, 저자 , 발행년, 권, 호, 시작페이지, 끝페이지, 발행기관 관리번호, 논문명, 대등논문명, 저자 , 저널/프로시딩명, 발행기관, 발행년, 발행언어, 권, 호, 시작페이지, 끝페이지, ISBN, ISSN, 주제분야, 키워드, 초록(한글), 초록(영문), 저자(소속기관)
저장형식	Text(ASCII format) Excel format RefWorks Direct Export RIS format (for Reference Manager, ProCite, EndNote), Scholar's Aids, Mendeley
메일정보	받는사람 (필수) @ 보내는사람 (선택) @ 제목 내용 KISTI 검색결과 이메일 서비스
안내	총 건의 자료가 검색되었습니다. 다운받으실 자료의 인덱스를 입력하세요. (1-10,000) 검색결과의 순서대로 최대 10,000건 까지 다운로드가 가능합니다. 데이타가 많을 경우 속도가 느려질 수 있습니다.(최대 2~3분 소요) 다운로드 파일은 UTF-8 형태로 저장됩니다. 파일의 내용이 제대로 보이지 않을실 때는 웹브라우저 상단의 보기 -> 인코딩 -> 자동선택 여부를 확인하십시오. ~ Text(ASCII format) Excel format

연합인증