Acoustic Impedance Inversion Based on Transfer Learning Combined with Convolutional Neural Network and Residual Network

Abstract

Summary Post-stack acoustic impedance inversion has definitive geophysical significance.Traditional acoustic impedance inversion is based on convolutional model and requires low frequency model. When the definition of the initial model is not accurate, the impedance inversion is not credible. In addition, due to the filtering effect of seismic response, impedance inversion is band-limited.In this abstract, a method combining deep learning and transfer learning is used to inverse acoustic impedance. Independent of the initial model, the nonlinear relationship between seismic data and acoustic impedance can be established by CRNN. The network architecture trained by simulated data is finetuned with little logging data. Transfer learning not only overcomes the problem of less label data in field inversion, but also solves the approximation problem of convolutional model. We used two typical models with different geological characteristics to prove the effectiveness of the inversion method. This provides a new method for seismic inversion in field area.