Noise attenuation of seismic data from simultaneous-source acquisition

Simultaneous shooting achieves a much faster seismic acquisition but poses a challenging problem for subsequent processing because of the interference from the neighbor crews. Separation of different sources, also called deblending, becomes important for the overall success of this acquisition technology. I propose a novel iterative estimation scheme for separating the blended simultaneous source seismic data to produce separate-source data as if they were acquired independently. I construct an augmented estimation problem, then use shaping regularization to constrain the characteristics of the model during the inversion and to obtain a suitable estimation result. The data reconstruction and source separation problems can be combined into one problem in order to make the future acquisition more flexible and efficient. In order to best utilize the capability of median filtering in attenuating spike-like noise, I also propose to use a new type of median filter (MF), termed as space-varying median filter (SVMF) to remove blending noise. SVMF can be regionally adaptive, instead of rigidly using a constant window length through the whole profile for MF. Simultaneous-source seismic data may also contain strong ambient random noise, so traditional denoising is still an important step. One of the most widely used approaches for removing random noise is using a sparse-transform thresholding strategy. I propose a double sparsity dictionary (DSD) for seismic data in order to combine the benefits of both analytic transform and learning-based dictionary. In the DSD framework, data-driven tight frame (DDTF) obtains an extra structure regularization when learning dictionaries, while the seislet transform obtains a compensation for the transformation error caused by slope dependency. DSD aims to provide a sparser representation than the individual transform and dictionary and therefore can help achieve better performance in denoising applications. Finally, considering that signal loss sometimes cannot be avoided in nearly all the existing denoising or deblending approaches. I propose a novel approach to retrieve the leakage energy from the initial noise section using local signal-and-noise orthogonalization. The proposed denoising approach corresponds to orthogonalizing the initially denoised signal and noise in a local manner. I evaluate denoising performance by using local similarity. The local signal-and-noise orthogonalization algorithm can also be used in the iterative deblending framework for obtaining better performance.