High-Performance Deep Learning Models for Seismic Noise Detection and Quality Control in the Processing Workflow

Summary In this paper, we demonstrate deep neural network models’ ability to recognize noises with complex patterns in seismic images with high accuracy and generalizability. We designed a creative labeling strategy generating many high-quality labels for the supervised learning component. We built three deep learning models, predicting key quality metrics for the noise attenuation workflow in seismic processing projects: including a swell noise level model, a Seismic Interference (SI) noise level model, and a signal leakage model. These models have been successfully deployed to Shell exploration projects in the Gulf of Mexico.