Active Deep Learning for Seismic Facies Classification

Seismic facies interpretation is a critical aspect of oil and gas exploration, yet it is not practical for human interpreters to thoroughly analyze every part of the data. Recent attention has been given to deep learning-based interpretation. However, to obtain a sufficiently large and accurately labeled training dataset within the project timeline can be challenging. To overcome this challenge, active learning methods have been proposed. They reduce the number of required training labels by creating an optimized labeled training set from unlabeled data. In this study, we developed an end-to-end encoding-decoding deep neural network for seismic facies classification and employed an active learning workflow with three different query strategies. The results showed that similar results to the baseline could be achieved using less than half of the labeled training dataset, even with rudimentary methods, such as random sampling. The uncertainty sampling method proved to be the most effective. These promising findings suggest that active learning can improve the interpretation of seismic data, particularly in emerging exploration frontiers where rapid interpretation is critical to success. The use of active learning can also make deep learning-based seismic interpretation more practical and efficient by reducing the dependence on large, labeled training datasets.