«Automation and Informatization of the fuel and energy complex»

Segmentation of tectonic faults planes by self-supervised learning methods

UDC: 004.855.5
DOI: -

Authors:

¹ Tyumen petroleum research center, Tyumen, Russia
² Tyumen State University, Tyumen, Russia

Keywords: seismic exploration, faults, machine learning, self-supervised learning, transformers, segmentation, neural networks

Annotation:

The authors of the article present a new method of self-supervised learning adapted to the seismic data structure. Some specific features of the method include training of the model for solving two preliminary tasks: data recovery on each individual slice and determination of neighboring slices in a seismic cube. A set of 100224 seismic images with a size of 128×128 pixels out of which 76800 pixels were unmarked and used for self-supervised learning by the proposed method and methods such as DINOv2, SimMIM, SimCLR on DeiT-S and SwinV2-Tiny architectures were used for training. The remaining 23424 images with labels, divided into training and testing samplings in a ratio of 70 to 30 %, were used for models additional training by the supervised method of solving the problem of faults planes identification of tectonic disturbances, on which the quality of the learned representations was evaluated using the Dice measure. According to the test results, the proposed method achieved comparable quality with completely supervised learning, using 60 % less of the marked-up data and 15 % (DeiT-S) and 23 % (SwinV2-Tiny) and surpassed it in terms of segmentation quality on the entire volume of the data for the selected architectures, and the existing self-supervised learning methods: SimCLR, SimMIM, DINOv2 in the average of 6 % (DeiT-S) and 4 % (SwinV2-Tiny) relative to the best result. The proposed solution allows reducing the dependence on a large number of annotated examples subject to the interpreter subjective opinion, increasing the quality of solving the subsequent problem of faults detection and creating a universal model for solving many other problems of seismic data interpretation.

Bibliography:

1. Murtazin D.G., Belyaeva A.S. Razrabotka modulya klasterizatsii amplitudnykh spektrov na yazyke Python // Avtomatizatsiya i informatizatsiya TEK. – 2021. – № 2(571). – S. 6–11. – DOI: 10.33285/0132-2222-2021-2(571)-6-11
2. Zhao Tao, Mukhopadhyay P. A fault detection workflow using deep learning and image processing // SEG Technical Program Expanded Abstracts 2018, Anaheim, CA, Oct. 16, 2018. – P. 1966–1570. – DOI: 10.1190/segam2018-2997005.1
3. Convolutional neural networks for fault interpretation in seismic images / Wu Xinming, Shi Yunzhi, S. Fomel, Liang Luming // SEG Technical Program Expanded Abstracts 2018, Anaheim, CA, Oct. 16, 2018. – P. 1946–1950. – DOI: 10.1190/segam2018-2995341.1
4. FaultSeg3D: Using synthetic data sets to train an end-to-end convolutional neural network for 3D seismic fault segmentation / Wu Xinming, Liang Luming, Shi Yunzhi, S. Fomel // Geophysics. – 2019. – Vol. 84, Issue 3. – P. IM35–IM45. – DOI: 10.1190/geo2018-0646.1
5. Tiwari A. Supervised learning: From theory to applications // Artificial Intelligence and Machine Learning for EDGE Computing. – 2022. – Chapter 2. – P. 23–32. – DOI: 10.1016/B978-0-12-824054-0.00026-5
6. Automatic fault detection on seismic images using a multiscale attention convolutional neural network / Gao Kai, Huang Lianjie, Zheng Yingcai [et al.] // Geophysics. – 2022. – Vol. 87, Issue 1. – P. N13–N29. – DOI: 10.1190/geo2020-0945.1
7. MD Loss: Efficient Training of 3D Seismic Image Fault Segmentation Network Under Sparse Labels by Weakening Anomaly Annotation / Dou Yimin, Li Kewen, Zhu Jianbing [et al.] // IEEE Transactions on Geoscience and Remote Sensing. – 2022. – Vol. 60. – Article No. 5919014. – DOI: 10.1109/TGRS.2022.3196810
8. Automatic Geologic Fault Identification from Seismic Data Using 2.5D Channel Attention U-net / Lin Lei, Zhong Zhi, Cai Zhongxian [et al.] // Geophysics. – 2022. – Vol. 87, Issue 4. – P. 1–58. – DOI: 10.1190/geo2021-0805.1
9. Ziegler A., Asano Yu.M. Self-Supervised Learning of Object Parts for Semantic Segmentation // 2022 IEEE/CVF Conf. on Computer Vision and Pattern Recognition (CVPR), New Orleans, LA, USA, June 18–24, 2022. – P. 14502–14511. – DOI: 10.1109/CVPR52688.2022.01410
10. Monteiro B.A.A., Oliveira H., dos Santos J.A. Self-Supervised Learning for Seismic Image Segmentation from Few-Labeled Samples // IEEE Geoscience and Remote Sensing Letters. – 2022. – Vol. 19. – Article No. 8028805. – DOI: 10.1109/LGRS.2022.3193567
11. Zhang Zeren, Chen Ran, Ma Jinwen. Improving Seismic Fault Recognition with Self-Supervised Pre-Training: A Study of 3D Transformer-Based with Multi-Scale Decoding and Fusion // Remote Sensing. – 2024. – Vol. 16, Issue 5. – Article No. 922. – DOI: 10.3390/rs16050922