Fast Matching Pursuit Decomposition And Instantaneous Spectral Analysis Of Seismic Signals

In the field of seismic data processing and interpretation, people pay more and moreattention to the sparse representation of the signal in recent years. The ideal signaldecomposition should select basis functions adaptively to decompose the signal based on thecharacteristics of the signal, which is especially important of the seismic signals with thelocalized characteristics both in time domain and frequency domain. Matching pursuitalgorithm (MP) proposed by Mallat et al. is just a signal decomposition method with theseadvantages. The algorithm decomposes the signal by creating redundant time-frequencydictionaries to achieve the adaptive decomposition.The huge computation of matching pursuit algorithm has restricted its wide application.Based on the three-parameter dynamic fast matching pursuit algorithm, the number of controlparameters of the time-frequency dictionaries is reduced to two by solving amplitude andphase of the time-frequency atoms in the complex field, which further improves the efficiencyof matching pursuit algorithm. According to the characteristics of the two-parameter dynamicfast matching pursuit algorithm, by means of parallel computers and parallel computinglanguage, In this paper, two parallel decomposition models based on CUDA and MPI arefurther designed, which makes a exploratory step of the matching pursuit parallel computing.Time-frequency analysis of the non-stationary signals is an important application ofmatching pursuit decomposition. In this paper, aiming at the shortcomings of the traditionalmatching pursuit Wigner-Ville distribution, a new calculation method of matching pursuittime-frequency representation is proposed. Under the premise of maintaining the accuracy, thespectrum of the amplitude and phase and other information can be obtained. By theintroduction of adjustable parameter in Morlet wavelet, the time resolution and frequency resolution of the matching pursuit time-frequency distribution are adjustable to meet the needof solving can transform the time-frequency distribution to the original signal inversely, whichopens up a broader space for the matching pursuit technology in seismic signal analysis.Finally, matching pursuit time-frequency analysis is applied to the seismic spectraldecomposition technique in this paper, and its advantage in the interpretation of thinreservoirs is demonstrated. Combining with the optimization methods of instantaneousspectral components, such as principal component analysis and highlight attributes, acomplete set of the instantaneous spectral interpretation techniques for thin reservoir areestablished, which have achieved good results in the delta fringe line recognition in actualwork area.