A Method for Data Denoising and Smoothing Based on DAS VSP

Distributed acoustic sensing (DAS) systems with weak fiber Bragg gratings (wFBGs) array are widely used in oil and gas exploration to acquire vertical seismic profiles (VSPs) due to its high sensitivity. However, the quality of seismic data based on DAS VSP is affected by the fading noise and the staircase signals. A method based on a multiscale residual attention network (MRANet) is proposed in this article to effectively suppress fading noise and smooth staircase signals. MRANet employs a multiscale pyramid feature extraction (MSPFE) module to capture global features of the continuity of seismic signals. Full-scale skip connections and SimAM are used to improve the ability of network to extract important local features regarding the details of seismic signals. Then, MRANet maps the noise and staircase data to the clean data to suppress noise and smooth staircase. MRANet achieved a signal-to-noise ratio (SNR) improvement of 13.125 dB and a fitting smoothness (FS) of 0.999 in simulated data, exceeding the best result of other methods by 1.815 dB in SNR and 0.001 in FS. For field data, MRANet produced an SNR enhancement of 16.916 dB and an FS of 0.999, outperforming the best result of other methods by 1.044 dB in SNR and 0.001 in FS. This method enhances data utilization and supports the wider application of DAS in resource exploration.