KG-Unet: a knowledge-guided deep learning approach for seismic facies segmentation

The accurate segmentation of seismic facies is of great significance to the study of sedimentary facies and the exploration and development of oil and gas resources. In recent years, scholars both domestically and abroad have proposed a variety of methods for the intelligent identification of seismic facies based on the DL (Deep Learning) model. However, because the DL model is purely data-driven, the one used for seismic facies segmentation frequently has issues like unclear interpretation and rough edges when seismic data reflection characteristics are not visible or little sample data is employed. To address the aforementioned issues, we suggest the KG-Unet model, a deep learning seismic facies segmentation method based on knowledge constraints. We evaluated KG-UNet model on the F3 three-dimensional seismic dataset in the North Sea. The experimental results indicate that seismic facies should be determined not only based on their own properties but also on their relationships with the surrounding facies. The supplementary information provided by non-target facies improves the segmentation performance of seismic facies. With the adoption of this model, the seismic phase segmentation results are good, and issues such as interpretation confusion and rough edges are well resolved.