GNGTS 2018 - 37° Convegno Nazionale

588 GNGTS 2018 S essione 3.1 We report a couple of results computed on two different structural features: an area with the presence of a salt body and the water bottom interface (Fig. 2 top triplet) and an area of sediments only (Fig. 2 bottom triplet). As evaluation metric we compared the output and the target through the structural similarity index (SSIM). For each example we show: (a) the input migrated image; (b) the corresponding reference reflectivity; (c) the output of the generator trained with the additional L1 constraint for 200 iterations. In the first example, the average SSIM is around 0.66, as the salt makes the problem more challenging; in the second example, the average SSIM is 0.90, much closer to the optimal value of 1. Despite the use of L1 constraint in loss function does not improve the quality of the reconstructed image, the convergence was reached much faster. Fig. 3 is a 1D vertical profile extracted at the central horizontal location from the second example of Fig. 2. The blue dotted line represents the ideal reflectivity profile; the black lines are extracted from the migrated section (left) and the output of the GAN with regularization (right). The effect of deconvolution is quite evident, and the amplitude is well recovered. Conclusions. In this manuscript we proposed an alternative use of a GAN as seismic image processing operator. Our model relies on the architecture proposed by Isola (2017) with a modified loss function tailored to seismic image processing. We have proven the proposed pipeline as a useful tool for very different geophysical imaging applications. Future work will be addressed to real data processing, GAN generalization and 3D data. Acknowledgement. The authors would like to thank Nicola Bienati for his encouragement in focusing on this research topic. References Al-Regib G., Deriche M., Long Z., Di H., Wang Z., Alaudah Y., Shafiq M.A., Alfarraj M.; 2018: Subsurface Structure Analysis Using Computational Interpretation and Learning: A Visual Signal Processing Perspective . In: IEEE Signal Processing Magazine, 35 , 2 , pp. 82-98. Araya-Polo M., Jennings J., Adler A., Dahlke T.; 2018: Deep-learning Tomography . In: The Leading Edge, 37 , pp. 58-66. Bestagini P., Lipari V., Tubaro S.; 2017: A Machine Learning Approach to Facies Classification Using Well Logs . In: SEG Technical Program Expanded Abstracts 2017, pp. 2137-2142. Goodfellow I., Pouget-Abadie J., Mirza M, Xu B., Warde-Farley D., Ozair S., Courville A., Bengio Y.; 2014: Generative Adversarial Nets . In: Advances in Neural Information Processing Systems 27 , pp. 2672-2680. Hall B.; 2016: Facies Classification Using Machine Learning . In: The Leading Edge, 35 , 10 , pp. 906-909. Isola P., Zhu J., Zhou T., Efros A.A.; 2017: Image-to-Image Translation with Conditional Adversarial Networks . In: IEEE Conf. on Computer Vision and Pattern Recognition, pp. 5967-5976. Kingma D. P., Ba J.; 2015: Adam: A Method for Stochastic Optimization . In: Int. Conf. For Learning Representations. Lucas A., Iliadis M., Molina R., Katsaggelos A.K.; 2018: Using Deep Neural Networks for Inverse Problems in Imaging: Beyond Analytical Methods . In: IEEE Signal Processing Magazine, 35 , 1 , pp. 20-36. McCann M.T., Jin K. H., Unser M.; 2017: Convolutional Neural Networks for Inverse Problems in Imaging: A Review . In: IEEE Signal Processing Magazine, 34 , 6 , pp.85-95.