Research On Segmented Time-variant Wavelet Estimation Based On The Improved Adaptive Molecular Decomposition

Most wavelet extraction methods existing are based on traditional convolution model and neglect the non-stationarity of seismic data, so the extracted wavelet is not accurate. It is a common way to split the non-stationary seismic data into segments and processing them as stationary seismic data approximately. It widens the conditions used for wavelet extraction and reduces the restrictions of non-stationarity by partitioning non-stationary seismic data properly. The thesis researches two time-variant wavelet estimation methods of which one is based on overlapping segment and the other is based on overlapping segment and spectral modeling. At last, a time-variant wavelet estimation method based on the improved adaptive molecular decomposition is presented as the two methods above both have defects. The proposed method improves wavelet estimation accuracy effectively.At first, following the wavelet estimation approach based on higher-cumulant ARMA model, an overlapping segment based method is used to estimate time-variant wavelet by discussing the influence rules of non-stationary seismic data length on wavelet parameters estimation. This method can make use of seismic signal information repeatedly, so wavelet estimation accuracy was raised effectively. However, the method is restricted by short data length.Then, in order to solve the above problem, a time-variant wavelet estimation method based on overlapping segment and spectral modeling is presented. This method used spectral modeling to fit time-variant wavelet amplitude. It had the advantages of rapid computing speed and high accuracy. Furthermore, it effectively resolved the restraint of short data length. At the same time we combined the method of overlapping segment to partition the non-stationary seismic data properly so as to achieve time-variant wavelet estimation. Simulation experiment and real seismic data processing results show that the method is applicable and effective in wavelet extraction.At last, the method of overlapping segment partitions time windows in a constant length without referencing the characteristics of non-stationary seismic data, so artificial error is brought in. This thesis presents a time-variant wavelet estimation method based on the improved adaptive molecular decomposition. According to the fundamental theory of adaptive molecular decomposition, we improved the way of fitting wavelet amplitude spectrum and added boundary conditions to the similarity coefficient function, so the optimizing accuracy of adaptive molecular windows was raised effectively. Meanwhile, the way of max similarity of time-variant wavelet amplitudes was used to determine optimum segment point so as to accurately partition adaptive molecular windows and obtain the time-variant wavelet via inverse Fourier Transform. The theoretical analysis shows that the method is able to fasten the solution speed and the result of adaptive molecular windows is more reasonable. Merits and defects between overlapping segment and adaptive segment are compared. Simulation experiment and real seismic data processing results show that adaptive segment is more suitable for time-variant wavelet estimation.