Seismic Data Interpolation by Learning a Denoiser With Deep Implicit Prior

In seismic exploration, collected traces inevitably appear noise and irregular sampled along the spatial coordinates, which affects seismic inversion and imaging. Seismic data interpolation is modelled by solving an inverse problem with regularization terms in mathematics. But sparse or low-rank priors in model-based methods cannot capture complex information from seismic data. A denoiser learned by convolution neural network (CNN) can be regarded as an implicit prior which helps improving the accuracy of model. For this motivation, we choose an unbiased DnCNN as a blind denoiser, and use a stochastic gradient algorithm to iteratively train the interpolated model, thus using the a priori information implied in the denoiser to interpolate the seismic data. We demonstrate the feasibility of the proposed method through synthetic and field data with comparative experiments, where it has excellent performance with handling interpolation and spatial aliasing.