Coherent noise suppression via a self-supervised deep learning scheme (Liu, S., and Alkhalifah, T., 2022)

Coherent noise attenuation is an essential step in seismic data processing to improve data quality and signal-to-noise ratio. We propose the use of self-supervised Structured Noise2Void. Through the inclusion of a noise mask, the coherency of noise is suppressed by randomizing the noise, allowing the network to learn how to predict only the signal component of a sample’s value. Unlike supervised methods that require noisy-clean data pairs, we directly use patches of raw noisy data as labels and corrupt these patches to use them as inputs to build the training sets that we employ via a blind-spot strategy to train a CNN network to effectively attenuate trace-wise coherent noise.

References

Liu, S., and Alkhalifah, T., 2022, "Coherent noise suppression via a self-supervised deep learning scheme", submitted to the 83rd EAGE Annual Conference and Exhibition.