Study On Signal And Noise Separation Based On Curvelet Transform

The study of seismic data denoising methods is a very important task in seismic data processing. With the increased difficulties of oil exploration and the enhancement of seismic imaging accuracy, the existence of ground roll, random noise and especially multiple seriously degrades the signal to noise ratio of seismic data, and interferes the identification of primary wave, meanwhile, may also cause some false geological features in seismic section.Therefore, the study of seismic data denoising methods is vital importance in seismic exploration. Common denoising methods have their own limitations, for the purpose of looking for a better method that can remove the noise in a greater degree and reserve more effective signal, meanwhile, also adapt to the seismic signal, the Curvelet theory which is widely used in recent years is introduced. Effectively combining sparse representation of the seismic signal and denoising methods, the signal and noise separation methods based on Curvelet transform is developed in the paper.Taking advantage of multi-scale and multi-directional property, a multi-stage ground roll removal method is devised. Making full use of the characteristic of non-overlapping of primary and ground roll in curvelet domain, the curvelt coefficient that only contains ground roll is processed. In order to avoid the loss of effective signal or incomplete suppression of random noise due to single threshold selection in the curvelet threshold denoising method, a new curvelet threshold denoising method based on empirical mode decomposition(EMD)was proposed in this paper. According to the distributing level of noise in each intrinsic mode function(IMF), different thresholds were chosen to process the IMFs with noise. Applying the methods to synthetical and real field data sets indicate that ground roll and random noise can be effectively removed.In the aspects of multiple suppression, the thesis improves the existing deficiencies of SRME. Making use of the multi-directional and multi-scale properties of curvelet theory, the wavefield domain multiple model prediction and separation method based curvelet transform is proposed. During the stage of multiple model prediction, the directional characteristic of curvelet transform allows for exploitation of Snell’s law at the free surface. In this way,multiple contribution gathers(MCG) in stationary zone can be automatically selected, thus reducing the number of non-constructive multiple contributions, which might cause artifacts in the stage of multiple model construction. Moreover, when the input original data exists space aliasing, taking advantage of multi-scale property of curvelet transform, an anti-aliasing weighting filter is designed using the low frequency data without spacial aliasing to reduce the adverse influence to multiple model.Standard matching methods can only correct the amplitude or phase errors between multiple model and actual multiple. As for the misalignment errors, these methods can not get better results. As a result, a new curvelet transform named complex curvelet transform(CCT)is introduced. Taking advantage of shift invariance property of CCT, we can use the phase and amplitude of the data’s and multiple model’s CCT coefficients to correct misalignment and amplitude errors between multiple model and actual multiple. In addition, for the purpose of protecting primary wave further, a non-linear masking filter is applied in advance, which can separate most of primary wave firstly, then recover the remaining primary using the CCT-based separation method. Finally, the new method can be demonstrated to be effective by testing synthetic and field data.