Dealiased Seismic Data Interpolation Using Time Dynamic Warping with Dictionary Learning

Abstract

In seismic data, interpolating regularly missing traces is generally regarded as more challenging than interpolating irregularly missing traces. To address regularly missing cases, an anti-aliasing strategy should be incorporated. In this paper, we employed dictionary learning approaches for seismic data anti-aliasing interpolation. In dictionary learning, it is crucial to pre-interpolate the sampled data to ensure that the learning dictionary captures the data structure. Currently, the nearest trace interpolation method is being used for pre-interpolation, which fails to utilize the spatial characteristics of data events. To overcome this limitation, we propose a pre-interpolation dictionary learning method based on time dynamic warping. The time dynamic warping technique calculates the similarity between two adjacent sampling traces and establishes the most similar path between the points. Subsequently, pre-interpolation data is obtained by linearly interpolating between these similar points. In the experimental comparison, we evaluate the performance of our proposed approach against the nearest pre-interpolation dictionary learning method. Synthetic and field data both demonstrate superior performance when using our proposed approach compared to the nearest pre-interpolation dictionary learning method.