An attention-guided convolution neural network for denoising of distributed acoustic sensing-vertical seismic profile data

Distributed acoustic sensing technology is a new type of signal acquisition technology, and this technology has been widely used in obtaining vertical seismic profile data in recent years. Distributed acoustic sensing technology has the advantages of high sampling density and strong tolerance to a harsh environment. However, in the real distributed acoustic sensing-vertical seismic profile data, the effective signal will be annihilated by various noises, which significantly complicates data analysis and interpretation. Deep learning approaches have developed rapidly in the noise suppression field in recent years. In order to eliminate the noise in distributed acoustic sensing-vertical seismic profile data, based on traditional convolution neural network, we add channel attention and spatial attention modules to the network to enhance the feature extraction ability of the network and use extended convolution to increase the receptive field to build a more efficient denoising model. In addition, we use different indicators to evaluate the quality of denoising, including signal-to-noise ratio, mean absolute error, kurtosis and skewness. The experimental results show that our method can recover the uplink wave field and downlink wave field, remove horizontal noise, optical system noise, random noise and other noises and improve the overall signal-to-noise ratio before and after denoising by 22 dB, reflecting a good ability of denoising and recovering effective signals.