Paper Summary
Swell noise attenuation is an important part of a seismic processing flow and is often subject to extensive testing. The optimal parameters will often not only vary between surveys, but also within a survey. An automatic classification process based on deep learning can be used with a traditional noise suppression algorithm to pick the optimal noise attenuation result even if the best parameterization varies throughout the survey. We show how extending the classification from purely differentiating between noise and signal to also include an additional mixed class helps to identify regions of visible signal with residual noise. Similarly, we show the same mixed class approach helps to identify areas in the attenuated energy with traces of signal leakage. The improved classification will make the automated QC procedure more robust.