
Full text loading...
Hyperbolic Radon transform being a time variant transform requires implementation in the time domain which is computationally expensive. To address this problem, we propose to use deep neural network for computing the velocity transform. The basic idea is to compute the inverse of Hessian using the Generative Adversarial network (GAN) for training data sets which can then be applied to the adjoint transform for test data sets to obtain a least-squares solution. This approach gives results similar to the least-squares inversion at a significantly reduced cost. Proposed algorithm is computationally efficient and cost effective as compared to the conventional time domain computation of the Hyperbolic Radon transform. Field data example verifies the effectiveness of the proposed approach.