The Seismic Data Denoising And Geological Object Recognition Based On Convolutional Neural Network

With the arrival of the big data era and the rapid development of data-based science,deep learning has made unprecedented progress.Honestly speaking,deep learning extracts high-level abstract features based on multiple low-level features by transforming them to multiple non-linear features.It also deeply excavates the data-based information for data science research.Deep learning is a new field of machine learning in artificial intelligence.It makes use of the learning mechanism of neural network to simulate the human brain for data interpretation.Deep learning theory based on data-driven can break through the constraints of model assumptions.Therefore,this paper focuses on introducing deep learning into the field of geophysical exploration.Specific contents are as follows:First of all,deep learning is introduced into seismic data denoising.Convolutional auto encoder belongs to the unsupervised model of deep learning.Through its specific end-to-end network structure,it can deeply excavate the high-level feature map involved in noisy data,and then recombine the high-level feature map so as to achieve data denoising.Secondly,deep learning is introduced into fault recognition of seismic data.Convolutional neural network belongs to the supervised model of deep learning.Convolutional neural network uses artificial interpretation fault data as training label,digging out the relationship between data characteristics and guidance label,and establishes a non-linear mapping based on deep characteristics for fault interpretation of seismic data.Finally,the convolution neural network is applied to three-dimensional model,Marmousi model and real data to verify the effectiveness of this method in fault identification.At the same time,the effects of data sets,hyper-parameters and network structure on the recognition accuracy are studied based on the model data.What's more,the convolution neural network and C3 coherence method are compared and analyzed,highlighting the advantage of convolution neural network based on data-driven,and its fault recognition effect in complex fault involved in high-angle layer and high-noise interference fault data is better than C3 coherence method.