Seismic wave velocity modeling method for multi-receptive-field knowledge distillation

The invention discloses a seismic wave velocity modeling method of a multi-receptive-field knowledge distillation regression network. By implementing a distillation architecture, a seismic record is divided into two paths of "Yes-No" noise. Through the mutual learning among the seismic data, the network can capture and represent important contextual information as an internal supervision form, so that the influence of noise on seismic records is reduced. The multi-receptive-field knowledge distillation network proposed for the first time can effectively help geological exploration to solve the seismic velocity problem, excavate the correlation of seismic wave frequencies, and construct new seismic wave features based on the learned correlation, thereby realizing high-quality seismic wave velocity model construction. A large number of experiments carried out on SEG Salt and Sulled data sets show that the model obtains a good result in the aspect of speed modeling. In future work, we will continue to explore ho