Pushing the limit of 5D interpolation using deep learning Available to Purchase

5D interpolation aims to recover the missing seismic traces in a 5D data volume, using all physical dimensions of seismic acquisition. Here, we develop a highly effective workflow for reconstructing highly incomplete data using deep learning (DL). The proposed method requires an ingenious data preconditioning scheme to provide a better sampled initial model for DL interpolation. The DL model works iteratively, passing the data between the unsupervised learning architecture for feature extraction and the iterative framework for reconstruction. An attention network highlights the important information within the extracted features, improving the denoising performance of the proposed DL model. Furthermore, we use several skip connections between the fully connected layers to enhance its learning capability. The proposed frameworkworks in an unsupervisedway where labeled data is not required. Aperformance comparison with benchmark methods using a challenging field data example shows that the proposed method outperforms the traditional methods.