Deep Prior-Based Unsupervised Reconstruction of Irregularly Sampled Seismic Data

Irregularity and coarse spatial sampling of seismic data strongly affect the performances processing imaging algorithms. Therefore, interpolation is a usual preprocessing step in most workflows. In this work, we propose method based on deep prior paradigm: an ad hoc convolutional neural network used as to solve inverse problem, avoiding any costly prone-to-overfitting training stage. particular, proposed leverages multiresolution U-Net with 3-D convolution kernels exploiting correlations cubes data, at different scales all directions. Numerical examples corrupted synthetic field sets show effectiveness promising features approach.