A sparse representation method for seismic data: adaptive multilayered dictionary learning (AMDL)

The accuracy of seismic data reconstruction by compressed sensing (CS) largely depends on the performance of the dictionary used for sparse representation. The sparsity level of each training sample in the K-singular value decomposition (K-SVD) is fixed, which may lead to under-fitting or over-fitting of the original sample. Moreover, it only uses the features of the original samples as the training dictionary and cannot utilize the implicit features generated in the dictionary learning process, which affects the reconstruction accuracy. In this paper, we adopt the adaptive multilayered dictionary learning (AMDL) method for the sparse representation of seismic data to improve the K-SVD method. It not only makes full use of the features at different levels in the dictionary learning process but also adaptively determines the number of atoms chosen for each layer. The experimental results show that the method can provide a more accurate sparse representation for CS-based reconstruction of seismic data than the K-SVD method. © 2022, Editorial Department OIL GEOPHYSICAL PROSPECTING. All right reserved.