A deep learning seismic processing framework based on pretraining: giving the dataset the attention it needs

Every seismic dataset has its particular characteristics guided mainly by the property of the subsurface it covers, the data acquisition parameters (the survey), and by the often unique noise condition for every dataset. Capturing such characteristics in a neural network model for efficient application of processing tasks offers a more effective approach to incorporating machine learning than training neural networks for specific tasks that may or may not transfer well to new data. We use a framework for seismic processing that allows us to pretrain a neural network to learn the features of a seismic dataset, and then fine tune the network for any downstream processing task. We take advantage of the fact that most processing tasks utilize the same features embedded in the seismic dataset, and thus, these features can be stored in a common pre-trained network in a self-supervised manner, we refer to as StorSeismic. We provide insights into the framework as it captures the seismic dataset features. Then, we use the labeled synthetic data to fine tune the pre-trained network in a supervised fashion to perform various seismic processing tasks, like denoising, low frequency extrapolation, first arrival picking, and velocity estimation, with satisfactory results.