The task of picking the reflectors is one of the most time-consuming tasks in interpreting seismic data. The known analytical algorithms for arranging this task are characterized by an extremely unstable solution in areas with unclear wave field behavior, interference, etc. The article describes the existing experience with machine learning for picking the reflectors. An approach for automatic picking of reflectors using neural networks is proposed, which is based on providing a solution for the segmentation task. To expand the receptive field of the neural network, it is suggested to adopt the Feature Pyramid Network architecture and replace regular convolutions in the encoder with the extended ones. It is also suggested that additional layers with linear interpolation and convolutional layers have been added to the decoder to obtain masks with a resolution greater than that of the image applied to the neural network's input. The authors consider the methodology for preparing seismic data for training the neural network and post processing the resulting predictions. The results of testing the proposed approach are presented. The proposed approach was tested on a number of 2D and 3D seismic data of various quality, geographically located within the Volga-Ural, Timan-Pechera oil and gas provinces, Western and Eastern Siberia. Based on the results of testing, an acceptable ratio of training time to the quality of the resulting surface of the reflector, satisfying the requirements of production units, was obtained. The method proposed in the article allows to reduce the amount of time spent by specialists to pick the reflectors.
References
1. Khayrullin T.A., Avtomaticheskaya pikirovka pervykh vstupleniy otrazhennykh voln (Automatic picking of the first arrivals of reflected waves), Collected papers “Tsifrovye tekhnologii v dobyche uglevodorodov: ot modeley k praktike” (Digital technologies in hydrocarbon production: from models to practice), Proceedings of scientific and technical conference, Ufa: Publ. of RN-BashNIPIneft’, 2021, pp. 79–81.
2. Stark T.J., Relative geologic time (age) volumes – Relating every seismic sample to a geologically reasonable horizon, The Leading Edge, 2004, V. 23, pp. 928-932, DOI:10.1190/1.1803505
3. Wu X., Hale D., Horizon volumes with interpreted constraints, Geophysics, 2015, V. 80(2), pp. IM21– IM33, DOI: 10.1190/geo2014-0212.1
4. Stark T.J., Unwrapping instantaneous phase to generate a relative geologic time volume, Proceedings of 73rd Annual International Meeting, SEG – 2003, pp. 1707–1710, DOI:10.1190/1.1844072
5. Luo S., Hale D., Unfaulting and unfolding 3D seismic images, Geophysics, 2012, V. 78(4), pp. O45– O56, DOI:10.1190/segam2012-1356.1
6. Ronneberger O., Fischer P., Brox T., U-Net: Convolutional Networks for Biomedical Image Segmentation, URL: https://arxiv.org/pdf/1505.04597.pdf
7. Koryagin A., Mylzenova D., Khudorozhkov R., Tsimfer S., Seismic horizon detection with neural networks, URL: https://arxiv.org/pdf/2001.03390.pdf.
8. Tschannen V., Delescluse M., Ettrich N., Keuper J., Extracting horizon surfaces from 3D seismic data using deep learning, Geophysics, 2020, V. 85(3), pp. 1MJ– Z13, DOI: 10.1190/geo2019-0569.1
9. Lin T., Dollár P., Girshick R. et al., Feature pyramid networks for object detection URL: https://arxiv.org/pdf/1612.03144.pdf.
10. He K., Zhang X., Ren S., Sun J., Deep residual learning for image recognition, URL: https://arxiv.org/pdf/1512.03385.pdf.
11. Yu F., Koltun V., Multi-scale context aggregation by dilated convolutions, URL: https://arxiv.org/pdf/1511.07122v3.pdf.
12. Okonnoe preobrazovanie Fur’e. Okno Blekmana (Window Fourier transform. Blackman window), URL: https://ru.wikipedia.org/wiki/Оконное_преобразование_Фурье#Окно_Блэкмана
