Machine Learning SIG Meeting – February (Online)

PESGB members can register free of charge via the membership platform

If you are interested in becoming a member of PESGB please visit: https://www.ges-gb.org.uk/membership/

Speaker: Ross 0’Driscoll – ION Geophysical

Topic: Deep reinforcement learning for random noise attenuation

Abstract
Random noise in marine seismic data can come from a variety of sources including swell and other seismic sources. Attenuating random noise is a complex yet essential task as it improves the signal to noise ratio in the dataset. Noise character varies within a survey and between surveys due to wave height, acquisition parameters, wind direction and many other factors. As such, parameterisation for one sail-line may not be applicable for other lines/surveys

We implemented a reinforcement learning algorithm to train an AI agent to parameterise a simple denoisealgorithm. There were two drivers for this test. Firstly, if the parameterisation can be learned, denoise testing can be sped up. Secondly, if the parameterisation can adapt to variations in swell noise, a more targeted workflow can be developed.

We found this method did adapt to noise levels in the data. Meaning the geophysicist could now move from parameterising the algorithm directly to designing the reward system for the AI agent.

Biography
Ross O’Driscoll works for ION Geophysical’s R&D department. He is currently working on signal processing algorithms and integrating machine learning into the geophysicist’s toolkit. He has worked as a Geophysical Advisor and Geophysicist over his ten years at ION.