Seismic facies identification based on Res-Unet and transfer learning
Seismic facies is the direct reflection of sedimentary facies on seismic section.The effective identification of seismic facies is an important work in seismic interpretation,sedimentary facies research and oil and gas exploration and development.Traditional seismic facies identification mainly depends on the understanding and experience of interpreters.It not only has a large workload,but also has strong subjectivity and low efficiency.Automatic seismic facies recognition technology based on deep learning can effectively solve the above problems.A Res-Unet network is constructed to learn the nonlinear relationship between seismic data and seismic facies.The Res-Unet network can deepen the depth of the network,make full use of the spatial information of seismic profiles,and overcome the difficulties caused by the similarity of different seismic facies on seismic profiles.Aiming at the problem of less training samples,a random cutting curve method is proposed to cut the three-dimensional data volume to expand the samples.Aiming at the generalization problem,the transfer learning strategy is proposed to transfer the learned model to the data to be identified,which simplifies the recognition process and improves the recognition accuracy.Firstly,the proposed method is applied to the F3 work area of Beihai.The results of seismic facies identification show that the Res-Unet network has high accuracy in seismic facies identification.Finally,the above trained model is migrated to the seismic data of a work area in the South China Sea,and good seismic facies recognition results are obtained.
万方数据
维普
