Leveraging a supervised machine learning toolkit for better seismic processing quality control

Geophysicists are facing great challenges to process increasingly large volumes of seismic data in a faster, better and more reliable way. In order to achieve this, one key element is to be able to perform quality control of processing steps in an effective and efficient manners. In this paper we present two cases where machine learning techniques are used to help speed up the quality control process. The first case describes an application of supervised learning with K-nearest neighbor to identify areas that may potentially cause cycle-skipping in full waveform inversion. The second case uses a logistic regression machine learning approach to detect and classify the presence of rig noise on shot points. Furthermore, we also tested an automated data reduction method using a long short-term memory auto-encoder, that could speed up the algorithm significantly without losing key information in the original data.