Deep convolutional neural network automatic horizon tracking method based on constant fast mapping

Abstract

The invention discloses a deep convolutional neural network automatic horizon tracking method based on constant fast mapping. The method comprises the following steps: S1, preprocessing a seismic profile image; S2, extracting, compressing and refining seismic reflection features by using a constancy fast connection convolutional neural network; S3, further collecting local features and global features of the information extracted in the step S2; and S4, performing classification calculation on the acquired features. According to the method, the deep convolutional neural network is used as a backbone to build a model, constant quick connection is used as a core, gradient flow of a traditional network is changed, features in seismic reflection are extracted more effectively, deep cascade aggregation of seismic phase features is achieved, and the accuracy of horizon resolution is improved.

Description

technical field [0001] The invention relates to an automatic layer tracking method of a deep convolutional neural network based on identity shortcut mapping. Background technique [0002] There is a very important correlation between seismic horizon interpretation and sequence stratigraphy division. Accurate horizon information is the basic work of seismic interpretation, and horizon identification and tracking is an important part of seismic horizon interpretation. Researchers have been focusing on Solve automatic tracking of events. Seismic data is generated by explosives or seismic vehicles on the ground through explosion or heavy impact. When it encounters the subsurface strata, it will emit reflected waves and be received by the surface geophones. Finally, reflection data such as seismic amplitudes are obtained. . Seismic interpreters typically consider peaks or troughs in a section of seismic data to be locations where events pass. Conventional artificial horizon ma...

AI Technical Summary

Problems solved by technology

[0009] Huang, Lu et al. published an unsupervised automatic layer tracking method using a self-organizing map network, but the success of the scheme depends on the initialization of the self-organizing map network, which is directly related to the quality of the results

Alberts, Huang et al. used artificial neural networks to track layers. This method requires more training data and has requirements for data representation, but the results are good.

Images