Z-Noise Attenuation Using Gaussian Mixtures Models and Unsupervised Learning
F. Perrone and S. Grion
Event name: 81st EAGE Conference and Exhibition 2019
Session: Poster: Full Waveform Inversion C
Publication date: 03 June 2019
Info: Extended abstract, PDF ( 1.13Mb )
Price: € 20
In Ocean Bottom Multicomponent acquisitions, the particle motion recordings can be heavily contaminated with shear-wave noise. The presence of this shear-wave noise, especially on the Z-component, hinders the pre-processing steps required to achieve an accurate calibration of the pressure and Z-component for up/down wave separation, also referred to as PZ summation. Failure to effectively attenuate this noise may reduce the quality of the separation and consequently the quality of the processing products that depend on it (up/down deconvolution and imaging). We implement a simple and flexible approach that operates in the Radon domain and uses clustering techniques borrowed from unsupervised machine learning to identify the coefficients that are likely noise and those that are likely signal. We use a set of local structural and similarity attributes to define the space of features for clustering and test the methodology on a real data example.