Using Convolutional Neural Networks for Denoising and Deblending of Marine Seismic Data

Abstract

Processing marine seismic data is computationally demanding and consists of multiple timeconsuming steps. Neural network based processing can, in theory, significantly reduce processing time and has the potential to change the way seismic processing is done. In this paper we are using deep convolutional neural networks (CNNs) to remove seismic interference noise and to deblend seismic data. To train such networks, a significant amount of computational memory is needed since a single shot gather consists of more than 10 6 data samples. Preliminary results are promising both for denoising and deblending. However, we also observed that the results are affected by the signal-tonoise ratio (SnR). Moving to common channel domain is a way of breaking the coherency of the noise while also reducing the input volume size. This makes it easier for the network to distinguish between signal and noise. It also increases the efficiency of the GPU memory usage by enabling better utilization of multi core processing. Deblending in common channel domain with the use of a CNN yields relatively good results and is an improvement compared to shot domain.

A multi‐data training method for a deep neural network to improve the separation effect of simultaneous‐source data

“…In the training process with the GT method (Figure 1), the training data pair is selected as $$\{ {{\bf s},{{\bf s}}_1} \}$$ (Richardson & Feller, 2019; Slang et al., 2019; Sun et al., 2020; Zu et al., 2020). The network parameter trained by this training data set is $${{\bm{\theta }}}^A$$.…”

Attenuation of marine seismic interference noise employing a customized U‐Net

“…However, we note that once the network is trained, the actual denoising of a single shot gather is done in less than a second. The results obtained using a NDCNN were quite encouraging, but the network performed less optimal in cases with a low SNR (Slang et al 2019). In order to reduce the training time and possibly improve the denoising results further, a second class of CNN network architectures was investigated, namely the U-Net (Ronneberger et al 2015).…”

“…Slang et al . () employed marine seismic field data and demonstrated successful applications within deblending and denoising using CNNs.…”

“…Ma (2018) used a CNN to attenuate multiples, linear noise and random noise simultaneously on a controlled data set where both the training and test data sets were computed from the same model. Slang et al (2019) employed marine seismic field data and demonstrated successful applications within deblending and denoising using CNNs.…”

Time-lapse seismic cross-equalization using temporal convolutional networks

“…Recently, many geophysical processes incorporate deep learning tools, which provide more robust solutions than conventional approaches. In data processing: Ovcharenko et al (2019) extrapolated low frequencies from high ones using deep learning; Slang et al (2019) used deep learning for denoising and deblending. For time-lapse processing: Alali et al (2020a) correct for the time shift in the data using a fully-connected layer in the latent space of an autoencoder; Duan et al (2020) showed that a trained network can outperform a conventional cross-correlation method in estimating the time-shift; Alali et al (2020b) suggested to use recurrent neural networks to better account for time dependency in the data.…”