Machine-learning-based data recovery and its contribution to seismic acquisition: Simultaneous application of deblending, trace reconstruction, and low-frequency extrapolation

Acquisition of incomplete data, i.e., blended, sparsely sampled, and narrowband allows for cost-effective efficient field seismic operations. This strategy becomes technically acceptable, provided that a satisfactory recovery the complete deblended, well-sampled, broadband is attainable. Hence, we explore machine-learning approach simultaneously performs suppression blending noise, reconstruction missing traces, extrapolation low frequencies. We have applied deep convolutional neural network in framework supervised learning which train using pairs incomplete-complete data sets. Incomplete are never used training use different subsurface properties acquisition scenarios, subsequently fed into trained to predict data. develop matrix representations indicating contributions strategies reducing operational effort. also determine simultaneous implementation source blending, sparse geometry, band limitation leads significant compression where size frequency-space domain much smaller than reduction indicative survey cost duration our can save. Synthetic examples demonstrate applicability proposed approach. Despite reduced amount information available results obtained from numerical cases clearly show scheme effectively deblending, trace reconstruction, low-frequency fashion. It noteworthy no discernible difference prediction errors between extrapolated frequencies preexisting observed. The potentially be acquired significantly compressed manner while recovering quality.