Convergence Improvement and Noise Attenuation Considerations for POCS Reconstruction

摘要

In recent years, a variety of algorithms for multidimensional seismic survey reconstruction have been proposed. Projection onto Convex Sets (POCS) seismic trace reconstruction appears to be an affective, uncomplicated and robust algorithm for the recovery of unrecorded seismic traces. Its slow convergence, however, could jeopardize its computational appeal. For this reason, we investigate the performance of POCS in terms of different threshold schedules. We show that a data driven threshold schedule leads to an efficient implementation of POCS where high-quality solutions can be obtained in a modest number of iterations (>30-40). In addition, we present a variant of POCS that permits concurrent reconstruction and denoising of seismic volumes. The latter is achieved by defining a weighted reinsertion trace strategy that alleviates the influence of the noisy traces in the final reconstruction of the seismic volume. We illustrate our findings with synthetic and real data examples.