EQUIVALENT Q ESTIMATION USING A DEEP-LEARNING-BASED DECOUPLING METHOD

Estimating Q model from post-stack seismic data plays an important role in seismic exploration. However, estimating Q is challenging because it is firmly established that the reflectivity and Q simultaneously affects the waveform of post-stack seismic data, leading to the fact that the Q model cannot be independently estimated without providing an accurate reflectivity model. The general approach for solving this problem is to simultaneously estimate these two parameters in an alternative iteration way. However, this problem is strongly ill-posed and the alternative iteration has no convergence guarantee. We propose a deep-learning-based decoupling method for estimating the equivalent Q model. Our basic idea is to use deep learning to decouple the effects of two parameters (reflectivity and Q) on seismic data, and establish two new single parameter inversion problems using the deep-learning-based decoupled seismic data to independently estimate reflectivity and equivalent Q. We propose a new objective function with regularization terms and minimize it using the Levenberg-Marquardt (LM) algorithm. Numerical results verified the effectiveness of the proposed method and demonstrated its advantages over common method.