Deep Reinforcement and Machine Learning for Seismic Data Processing and Automated QC

In this paper we show, as examples of a move towards towards the large-scale automation of imaging and processing projects, two different applications of new machine-learning based technologies. Both of these applications have the potential to significantly accelerate the turnaround of imaging projects. We also discuss the barriers that exist towards the adoption of technologies and methods such as these in widespread industrial use. In all of these examples we are not replacing the physics in the solutions with ML based technology – rather we are using these new methods as complementary technologies to accelerate turnaround. Example 1 shows how the parameterisation of a noise attenuation algorithm may be automated by deep reinforcement learning. Example 2

Example 2 illustrates how automated quality-control using machine learning may be applied to noise attenuation algorithms.