Attenuation compensation based on U-Net

Abstract

Summary The effect of absorption attenuation on seismic wave is shown in amplitude attenuation and phase dispersion, which greatly reduces the resolution of the seismic profile. Attenuation compensation method is commonly used to eliminate these effects. Lots of researches about attenuation compensation have been carried out for many years. However, traditional attenuation compensation methods are usually not stable and have poor noise resistance. With the explosion of deep learning (DL), it is an inevitable trend to apply deep learning method to attenuation compensation. In this abstract, we first analyse the suitable network structure for the attenuation compensation problem, and we improve the structure of U-Net to make it suitable for processing single seismic trace, then we implement the attenuation compensation based on U-Net and Gabor transform, the comparison of the results shows that the compensation based on U-Net has better stability and noise resistance performance while effectively compensating the amplitude energy attenuation and the phase distortion.