Research on Seismic Data Denoising Based on Dual Channel Residual Attention Network

Abstract

In recent years, seismic data denoising has attracted more and more scholars' attention and research, and the suppression of random noise is the key to improving the signal-to-noise ratio of seismic data. Aiming at the problem that traditional denoising methods are difficult to effectively remove a large amount of random noise and retain effective signals, we propose a neural network model based on dual channel residual attention network (DCRANet). Specifically, the model consists of a residual attention block (RAB), a dilated convolution sparse block (DCSB) and a feature enhancement block (FEB). The residual blocks in RAB can avoid some problems such as gradient vanishing and gradient exploding when the network is too deep, and the use of attention mechanism can guide the network to effectively extract complex noise information. The DCSB recovers the useful details from complex noise information by expanding the receptive field, fully acquiring important structural information and edge features of seismic data. The FEB integrates the noise features extracted by RAB and DCSB, it uses convolutional layers to extract the noise information of seismic data, and finally reconstructs clean seismic data image by the residual learning strategy. Compared with NL-Bayes, BM3D, DnCNN, CBDNet and DudeNet, DCRANet effectively suppresses random noise while retaining more local details and obtains a higher average peak signal-to-noise ratio (PSNR) and average structural similarity (SSIM).