Seismic random noise suppression by using MSRD-GAN

Abstract

Seismic exploration is a vital instrument for developing oil and gas reservoir resources, but the actual seismic data gathering process is usually plagued by noise interference. Therefore, reducing seismic data noise is critical for improving seismic data quality. This study proposed a seismic data noise suppression method based on a multi-scale residual density generative adversarial network (MSRD-GAN) to enhance seismic data quality. This network used local residual learning and density linking to establish multi-scale residual blocks (MSRBs) for multi-scale feature capture, avoiding the problem of the local perception of deep convolutional network features and the disappearance of hierarchically delivered features, which made it difficult to effectively recover signal details. It was also used with generative adversarial networks (GAN) to automatically learn the difference between noisy and legitimate seismic data signals, allowing it to suppress random noise while completely recovering valid signals. The MSRD-GAN was evaluated according to synthetic seismic data and field data to demonstrate its efficacy in suppressing seismic data noise. The experimental findings of both the synthetic and field data showed the benefits of the proposed MSRD-GAN in reducing complex random noise while preserving lower signal distortion.