Post-stack seismic impedance inversion method based on TransUNet neural network
Convolutional networks are the mainstream framework for deep learning seismic impedance inversion method;however,convolutional networks have the deficiency of capturing the long-term data dependence ability,which leads to the inversion results for seismic with long-term task characteristics will be greatly affected.Given the advantages of Transformer's ability to focus on global features of seismic,it can not only provide the location information of seismic data features,but also compensate the deficiency of convolutional for global information characterization of seismic data.Therefore,the author constructs a network structure based on Transformer and UNet networks that can not only portray the local details of seismic data but also characterize the global features of seismic data,that is,a network formed by embedding Transformer within the UNet framework as an inverse mapping network for seismic impedance inversion(TransUNet).The advantage of TransUNet is that it utilizes both the function of UNet that can extract the features of seismic data and the role of Transformer to encode the location of the above features,thus making TransUNet have the ability to capture the global information of seismic data and provide a new method and idea for the mapping relationship between seismic data and impedance,and it has been effectively verified in model and actual It provides a new method and idea for the mapping relationship between seismic data and impedance,and is effectively verified in models and real data.
NETL
NSTL
万方数据
