摘要
Conventional SVD technology suppresses random noise in time domain,which is suitable for seismic data with flat events.But for dipping and curving events,SVD methods in time domain are greatly limited by the necessary requirement of automatic tracing slopes of events in practice.In this paper,local SVD filtering in frequency domain is described to avoid the above limitation.Seismic data within a local window sliding in space and time are first extracted and Fourier transformed.Then SVD filtering is applied to constant-frequency slices by forming the Hankel matrix.Once all frequencies within the signal bandwidth are noise reduced,the clean section is obtained by taking the inverse DFT of each trace.This method is also extended to remove random noise from stacked 3D seismic volumes by forming a Hankel matrix.Synthetic data and field data processing indicates that this method can suppress random noise more effectively and preserve signal simultaneously,and does much better than conventional prediction filtering methods in frequency domain.