Random Noise Attenuation Using Convolutional Neural Networks

Abstract

Summary We propose a new method to attenuate random noise from seismic data using convolutional neural networks. Only synthetic data from Pluto dataset are used for training. The proposed method fits the noise rather than the signal to avoid overfitting and damaging the signals. Experiments on Sigsbee dataset shows that the propose method removes more random noise and preserves signal better than f–x deconvolution method.