Deep learning for high-resolution multichannel seismic impedance inversion

Seismic impedance inversion can obtain subsurface physical properties and plays an important role in hydrocarbon and mineral exploration. Due to the inaccurate and insufficient seismic data, the inverse problem is ill-posed as characterized by unreliability and non-uniqueness of solutions. Regularization techniques relying on certain prior information are often introduced to force the inverse problem to obtain stable results with predetermined characteristics. However, for complex geologic conditions, these methods are usually difficult to achieve satisfactory accuracy and resolution. We propose a deep-learning-based multichannel impedance inversion method, which flexibly incorporates prior information by training with numerous realistic structural 2D impedance models on the basis of features of field data. Our deep learning framework is supplemented by the attention mechanism and residual block to automatically learn more features and details from training data. We also introduce a new hybrid loss function that combines the ℓ 1 loss and Multi-scale Structural Similarity (MS-SSIM) loss to better enable the network to learn structural features. Synthetic and field examples demonstrate that the proposed method can effectively produce inversion results with high resolution, good lateral continuity, and enhanced structural features compared with traditional methods.