Desert seismic data denoising based on energy spectrum analysis in empirical curvelet domain

• Published in: Studia geophysica et geodaetica 2020-07, Vol.64 (3), p.373-390
• Main Authors: Li, Mo, Li, Yue, Wu, Ning, Tian, Yanan
• Format: Article
• Language: English
• Subjects: Adaptability; Anisotropy; Atmospheric Sciences; Coefficients; Deserts; Domains; Earth and Environmental Science; Earth Sciences; Earthquakes; Electroconvulsive therapy; Empirical analysis; Energy; Energy spectra; Frequencies; Geophysics/Geodesy; Noise; Noise reduction; Random noise; Seismic activity; Seismic data; Seismic exploration; Seismograms; Seismological data; Spectrum analysis; Structural Geology; Transformations (mathematics); White noise
• ISSN: 0039-3169; 1573-1626
• DOI: 10.1007/s11200-019-0476-4

Desert seismic events are disturbed and contaminated by strong random noise, which complicates the subsequent processing, inversion, and interpretation of the data. Thus, noise suppression is an important task. The complex characteristics of random noise in desert seismic records differ completely from those of Gaussian white noise such that they are non-stationary, non-Gaussian, non-linear and low frequency. In addition, desert seismic signals and strong random noise generally share the same frequency bands. Such factors bring great difficulties in the processing and interpretation of desert seismic data. To obtain high-quality data in desert seismic exploration, we have developed an effective denoising method for desert seismic data, which performs energy spectrum analysis in the empirical curvelet transform (ECT) domain. The empirical curvelet coefficients are divided into two different groups according to their energy spectrum distributions. In the first group, which contains fewer effective signals, a large threshold is selected to remove lots of random noise; the second group, with more effective signals, a coherence-enhancing diffusion filter (CEDF) is used to eliminate the noise. Unlike traditional curvelet transforms, ECT not only has the multi-scale, multi-direction, and anisotropy properties of conventional curvelet transform, but also provides adaptability to separate the effective signals from the random noise. We examine synthetic and field desert seismic data. The denoising results demonstrate that the proposed method can be used for preserving effective signals and removing random noise.