Multiple attenuation is an essential step in seismic processing, as it significantly improves data inversion and interpretation. Deep learning (DL) has emerged as a promising alternative for various seismic processing tasks, including multiple attenuation. In particular, Convolutional Neural Networks (CNNs), originally designed for natural image processing, have been applied to seismic problems as image-to-image translation tasks. However, existing CNN-based deep learning approaches neglect the spatio-temporal characteristics of seismic recordings, which are critical for interpreting seismic data as signals rather than images. In this study, we introduce a novel methodology to integrate spatio-temporal information into CNNs for seismic data processing tasks. Specifically, we focus on seismic multiple discrimination based on moveout in CDP gathers. We evaluate the impact of incorporating spatio-temporal coordinates on multiple attenuation performance across different CNN architectures and validate our approach on realistic seismic scenarios. We believe this methodology can significantly enhance seismic processing with CNNs and extend its applicability to various tasks beyond multiple attenuation.
Fernandez et al. (Wed,) studied this question.