Unsupervised learning with a soft attention mechanism for 3-D erratic and random noise attenuation

Abstract

Summary Coherent and incoherent noise in seismic data inevitably reduces the quality of subsequent processing, e.g., migration and inversion. Different from random noise, erratic noise follows non-Gaussian distribution and has high amplitude, which is a challenge to the conventional denoising frameworks based on deep learning (DL). The computational cost of the commonly used supervised-learning-based framework is unaffordable when processing multi-dimensional data, such as seismic data in more than three-dimensional space. We propose an unsupervised-learning framework with a multi-branch attention mechanism to attenuate the erratic and random noise in 3-D seismic data. The propposed network can adaptively attenuate noise in multi-dimensional seismic data without the need to manually generate labels to train the network. In the encoding stage, several multi-branch attention blocks are used to extract significant features and reduce the data dimension. The proposed network integrates global features of waveforms extracted from multiple branches in a weighted way to enhance attention to significant features, thus obtaining a global and comprehensive representation of weights. In the decoding stage, we use several stacked multi-branch attention blocks to reconstruct features. To enhance the migration ability of shallow-level to deep-level features, we add some skip connections in the corresponding encoder and decoder. We apply the proposed network for 3-D synthetic and field data. The denoising results demonstrate that the proposed method has better signal preservation and noise attenuation abilities compared with the classic denoising methods.