Broadband Reconstruction of Seismic Data with Generative Recurrent Adversarial Network

Wei et al. describe a new method to reconstruct broadband of seismic data by GRAN deep learning. The deep learning neural network is based on an improved network architecture of GAN and the discriminator is to build the loss function by introducing Wasserstein distance to ensure no gradient vanishing at any time. Although synthetic dataset was used in training, the method still gave satisfactory results in the real data processing study and it also showed the importance of multi disciplinary collaboration and data cleaning in deep learning.