This chapter presents an innovative approach to generating high-granularity three-dimensional (3D) seismic data from low-granularity data using tensor sparse coding. The method achieves efficient seismic data reconstruction by simultaneously training both low-granularity and high-granularity seismic dictionaries, leveraging the advantages of both. First, we introduce how tensor sparse coding adapts to the high-dimensional characteristics of seismic data and serves as a solution for seismic data interpolation. Next, we propose a novel dictionary-sharing strategy, where dictionaries trained on low-granularity and high-granularity data share the same sparse representation, enabling effective recovery of high-granularity seismic data. Finally, experimental results demonstrate that the interpolation process based on this method significantly enhances the resolution of seismic data and achieves promising reconstruction performance on real-world seismic datasets.