Missing Shots and Near-Offset Reconstruction of Marine Seismic Data With Towered Streamers via Self-Supervised Deep Learning

Marine seismic data with towered streamers have played an important role in marine exploration. However, the distance between adjacent sources and the distance between adjacent receivers/channels are inconsistent (i.e., like regularly missing shots) and near-offset information is unrecorded, which can decrease the performances of surface-related multiple elimination (SRME) and seismic migration. Traditional algorithms to provide prestack seismic data with consistent trace interval and to recover near-offset data have some drawbacks, including low efficiency of computation and super-parameter selection by trial and error. Thus, we propose a novel self-supervised deep learning (DL) algorithm to reconstruct regularly missing shots and recover near-offset information with an improved U-net by combining U-net and residual learning of ResNet. Via the spatial reciprocity of Green's function, common shot gathers (CSGs) have similar features as common receiver gathers (CRGs). The reconstruction performances of regularly missing shots in CRGs can be guaranteed by using the network that is trained and validated by adaptively extracted CSGs. To reconstruct near-offset information of CSGs, we first construct pseudo-seismic data with the dip approaching 0 at near-offset parts by a rotation-truncation strategy. Pseudo-seismic data can be regarded as seismic data with approximate near-offset information to train and validate the designed network, which is later used to reconstruct near-offset information for CSGs. Finally, field marine seismic data with towered streamers is used to demonstrate the validity and effectiveness of the proposed self-supervised algorithm in reconstructing regularly missing shots and recovering near-offset information, which are beneficial for subsequent processing of seismic data.