Signal processing techniques for the enhancement of marine seismic data
ForfatterSanchis, Charlotte Juliette
This thesis presents several signal processing techniques applied to the enhancement of marine seismic data. Marine seismic exploration provides an image of the Earth's subsurface from reflected seismic waves. Because the recorded signals are contaminated by various sources of noise, minimizing their effects with new attenuation techniques is necessary. A statistical analysis of background noise is conducted using Thomson’s multitaper spectral estimator and Parzen's amplitude density estimator. The results provide a statistical characterization of the noise which we use for the derivation of signal enhancement algorithms. Firstly, we focus on single-azimuth stacking methodologies and propose novel stacking schemes using either enhanced weighted sums or a Kalman filter. It is demonstrated that the enhanced methods yield superior results by their ability to exhibit cleaner and better defined reflected events as well as a larger number of reflections in deep waters. A comparison of the proposed stacking methods with existing ones is also discussed. We then address the problem of random noise attenuation and present an innovative application of sparse code shrinkage and independent component analysis. Sparse code shrinkage is a valuable method when a noise-free realization of the data is generated to provide data-driven shrinkages. Several models of distribution are investigated, but the normal inverse Gaussian density yields the best results. Other acceptable choices of density are discussed as well. Finally, we consider the attenuation of flow-generated nonstationary coherent noise and seismic interference noise. We suggest a multiple-input adaptive noise canceller that utilizes a normalized least mean squares alg orithm with a variable normalized step size derived as a function of instantaneous frequency. This filter attenuates the coherent noise successfully when used either by itself or in combination with a time-frequency median filter, depending on the noise spectrum and repartition along the data. Its application to seismic interference attenuation is also discussed.
ForlagUniversitetet i Tromsø
University of Tromsø
Følgende lisensfil er knyttet til denne innførselen: