SEISMIC RANDOM NOISE ATTENUATION USING MULTI-SCALE SPARSE DICTIONARY LEARNING

摘要

Seismic data contain random noise, which affects data processing and interpretationand possibly limits the use of seismic data in parameter building and attribute prediction.To effectively remove such noise, we develop a denoising workflow based on multi-scalesparse dictionary learning. The multi-scale sparse dictionary learning method decreasesthe complexity of data by two approaches. One is seismic data-sparse representationfrom the data domain to the wavelet domain by wavelet bases. The other is denoising bysparse dictionary learning in a certain frequency band without the noise effect from otherfrequency bands. Meanwhile, the wavelet transform can also reduce the dimension of thedata, which ensures the computational efficiency of the proposed method. Afteranalyzing the effects of sparse dictionary learning parameters on seismic data denoising,we test the proposed method on two synthetic and two field datasets. We learn from theexamples that our approach can effectively recover signals from simulated and real noisydata, as well as time-variant noisy data. Compared with the sparse dictionary learning,K-singular value decomposition dictionary learning, and double-sparsity dictionary(DSD), our method can obtain the best-denoised result with less effective signals leakingin noise sections.