Subsurface depths structure maps reconstruction with generative adversarial network

Aim. Show an approach to transferring knowledge about the structural forms of stratigraphic horizons from well-studied territories to little-studied ones using generative-adversarial networks.

Materials and methods. The work uses two algorithms based on the architecture of generative adversarial neural networks. The first StyleGAN2-ADA algorithm accumulates in the latent space of the neural network semantic images of geological forms, first of the mountainous terrain, and then, using transfer learning, the forms of structures of stratigraphic horizons. The second algorithm, the Pixel2Style2Pixel encoder, using the semantic level of generalization of the first algorithm, learns to reconstruct high-resolution original images from their discrete copies (super-resolution technology).

Results. Models for reconstruction were trained, with their help, detailed depth reconstructions were obtained from maps based on 2D seismic data, comparable to the quality of maps from 3D seismic. For two sites, an assessment of the quality of reconstruction was carried out. It is proposed to create a probabilistic space of depths of the study area, where each point is represented by the density of the probability distribution of depths of equally plausible reconstructed geological forms of structural constructions of the study area/

Conclusion. The proposed approach makes it possible to create a probabilistic representation of the possible forms of the buried relief within the framework of knowledge about the already studied territory. The correctness of the reconstruction depends on the representativeness of the high-resolution data for training and on the initial structural maps for reconstruction — how accurately they reflect the dynamics of changes in the absolute elevations of the studied horizon/

1. Goodfellow I., Pouget-Abadie J., Mirza M., Xu B., Warde-Farley D., Ozair S., Courville A., Bengio Y. Generative adversarial nets. NIPS, 2014, vol. 27.

2. Karras T., Laine S., Aittala M., Hellsten J., Lehtinen J., Aila T. Analyzing and improving the image quality of stylegan. arXiv:1912.04958, 2020.

3. Karras T., Aittala M., Hellsten J., Laine S., Lehtinen J., Aila T. Training Generative Adversarial Networks with Limited Data. arXiv:2006.06676, 2020.

4. Heusel M., Ramsauer H., Unterthiner T., Nessler B., Hochreiter S. GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium. arXiv:1706.08500, 2017.

5. Richardson E., Alaluf Y., Patashnik O., Nitzan Y., Azar Y., Shapiro S., Cohen-Or D. Encoding in Style: a StyleGAN Encoder for Image-to-Image Translation. arXiv:2008.00951, 2020.

6. Tov O., Alaluf Y., Nitzan Y., Patashnik O., Cohen-Or D. Designing an Encoder for StyleGAN Image Manipulation. arXiv:2102.02766, 2021.