Seismic data reconstruction via a hybrid convolutional neural network-low-rank method

ABSTRACT Convolutional neural networks (CNNs) have shown promising results in seismic data reconstruction due to their exceptional ability to extract local features. However, incomplete seismic data may impede CNNs’ ability to discern meaningful features, leading to a degradation of reconstruction performance. Therefore, we develop a hybrid CNN-low-rank (LR) method that leverages LR techniques to recover data structure information, followed by detail refinement using a plug-and-play (PnP) trained CNN within an iterative framework. This approach uses LR techniques to restore inherent structural information, compensating for the absence of local information, thereby enabling CNNs to infer missing traces more accurately. Specifically, we first train a network to learn the mapping from the LR recovered data patches with randomly selected rank parameters to complete the data patches. Next, we incorporate the PnP-trained CNN into an iterative framework, wherein the LR recovery and network refinement are interleaved within each iteration. This creates a continuous reconstruction process until a termination condition is met. Performance evaluations of our algorithm, conducted on synthetic and field data sets with a high missing trace ratio, demonstrate its superiority over the traditional LR method (multichannel singular spectrum analysis) and the CNN-based method (U-net). Moreover, our approach indicates strong performance in reconstructing large contiguous gaps and handling higher-dimensional seismic data (including three dimensions and five dimensions), broadening its applicability and robustness in practical seismic data reconstruction tasks.