The Suppression Of Random Noise And Separation Of Ground Roll In Seismic Signals Based On Sparse Representation

With the development of the technology in seismic exploration, geological exploration turns to areas with complex geological structure and environment, thus the quality of seimic data needs to be higher. While raw seismic data often appear to be with serious noise interference, low signal-to-noise ratio and the energy of effective signal is weak. The presence of noise seriously affect the subsequent processing and interpretation of seismic data, so the use of a suitable denoising method to remove noise and improve the signal-to-noise ratio of seismic data is becoming increasingly important.Sparse representation is a research focus in the field of signal processing. In recent years, the sparse representation theory is continually used for effective representation of seismic data in seismic exploration. Generally, effective signal and noise aliase together in space and time domain, so it is difficult to separate them directly. After transformation, the projecting coefficients of the reflected wave and the noise will have a better sorting. Noise coefficents can be filtered through setting appropriate threshold and then seismic signal can be reconstructed by inverse transform to achieve better denoisng effects. In the paper, the main research focus on sparse representation theory and its application in random noise denosing and ground-roll attenuation of seismic data. The main contents and innovations are as follows:1. We study sparse representation theoretical models and some common sparse decomposition algorithm. In addition, we describe the development of dictionary model in sparse representation theory. Furthermore, we study Morphological Component Analysis which is suitable for separation of different components in a complex signal.2. We construct the model of seismic signal random noise denoising based on sparse representation theory and introduce common methods of random noise suppression in seismic data. Then we describe random noise denoising in seismic data based on learning dictionaries method in detail. We firstly get an adaptive learning dictionary that can effectively represent seismic signal by using dictionary learning algorithm to seismic signal samples, then do sparse decomposition to the seismic signal under the learning dictionary to get sparse coefficients, finally we reconstruct seismic signals with the sparse coefficients to remove random noise. Pointing at the problem that training time of traditional learning dictionary is long and its structure is poor, we put double sparse learning dictionary into seismic signal denoising. This dictionary is composed of a based dictionary with a fixed structure and a sparse coefficient matrix obtained by training. The dictionary has a good structure and sparsity. It has a better sparse representation ability and denoising effect.3. We introduce the common methods of ground roll suppression in seismic signal. We describe dimensional wavelet transform thresholding algorithm to suppress surface waves. We propose to use morphological component analysis based on two dimensional dictionaries to separate ground roll. According to different waveform characteristics between reflected wave and ground roll, we selects different two-dimensional dictionary to sparsely represent them. We construct sparse representation model of seismic data under two-dimensional joint dictionary, use block coordinate relaxation algorithm to solve it and decompose seismic record into reflected wave part and ground roll part to achieve separation of gound roll. We use compared experiment to prove the validity of the method. The method takes the morphological structure of the different components in seismic signals into account and makes full use of the correlation between traces in seismic signals to achieve effective separation of ground roll.