Research On Identification Method Of Salt Dome In 3D Seismic Data

In seismic exploration,a very critical task is the interpretation of seismic data,in which the identification of geological anomalies in 3D seismic data is very important.In particular,the identification of salt domes plays an important role in understanding the salt structure and establishing the seismic migration velocity model.Currently,interpreting salt boundaries usually involves calculating one or more salt dome property maps and tracking salt boundaries in two steps.Although the current automated method has been applied to the calculation of salt dome properties and the extraction of salt dome boundaries,the interpretation of salt domes is still a time-consuming and labor-intensive task.Traditional methods derive seismic attributes based on physical principles,geometric structures,but these attributes may not fully represent real seismic data that contains noise and complex geological features.In order to reduce the impact of the above problems,realize the automatic interpretation of the salt dome and reduce the consumption of manpower and time cost,this thesis applies the convolutional neural network to the interpretation of the salt dome.The deep learning-based technology makes it possible to improve the efficiency and precision of salt dome recognition in large 3D seismic data sets,which is a very important research direction.Based on the original 3D seismic amplitude data volume,this thesis studies the method of salt dome recognition and segmentation with convolutional neural network and full convolutional neural network methods.The main work of this thesis is as follows:1.This thesis proposes an improved AlexNet network for segmentation of salt domes.By making some improvements to the AlexNet model for natural images,the patch is constructed centered on the sample points to make it suitable for the original 3D seismic amplitude data,which can effectively identify the geological features.Choosing a convolutional neural network model can automatically select the features that are most favorable for classification,and avoid the problem of manually selecting features or geological attributes.2.In order to solve the problem of inaccurate identification of the CNN-based method at the salt dome boundary,this thesis proposes a U-Net based salt dome segmentation method flow.Based on the original U-Net,the structure and parameters of the seismic data are adjusted to enable the network model to identify geological features more efficiently.U-Net extends from the classification of salt dome classification at the seismic amplitude image level to the classification at the pixel level.Through data expansion,the tag data can be used more effectively,and accurate salt dome classification can be realized while obtaining context information.The whole 3D seismic data volume is segmented to construct a three-dimensional salt dome model.In this thesis,the proposed method is validated by the actual 3D seismic data volume of F3 in the Netherlands.The results show that the proposed method can accurately classify the salt domes,and the obtained salt dome boundary is basically consistent with the artificial interpretation boundary.Computer vision can be successfully applied in the field of seismic exploration and deserves further study.