Research On The Method Of Improving The Resolution Of Seismic Data Based On Convolution Neural Network

Seismic exploration uses the difference of elasticity and density of underground media,and predicts the nature and shape of underground rock strata by observing and analyzing the response of earth to seismic wave,so as to achieve the purpose of improving the efficiency of petroleum resource development.However,seismic waves are easily affected by underground media,thin layer filtering,sampling rate and other factors,resulting in low resolution of seismic data collected.Therefore,improving the resolution of seismic data is the key step of seismic exploration,which is crucial to ensure energy independence and energy security.In recent years,due to the continuous development of artificial intelligence technology,convolution neural network has been widely used,and it has also received a lot of attention in the field of earthquake processing.In this paper,the method of improving seismic data resolution based on convolution neural network is studied.The main research contents include the following three aspects.Firstly,a method to improve seismic resolution based on dual attention U-Net network is proposed.First of all,the improved channel attention and spatial attention module can be added to the U-Net network,which can reasonably distribute the weight of different channels and spaces,and make full use of the correlation between data.Then,the combination of L₁loss and MS-SSIM loss is used as the loss function to train and debug the network,which improves the sensitivity of the model to local information changes and facilitates the recovery of detail information.The test results show that after network processing,the main frequency and band width of seismic data are improved,the phase axis is clearer,the detailed texture information is more abundant,and the resolution of seismic data is effectively improved.Secondly,a method of improving seismic resolution based on U-Net network of global context residuals in wavelet domain is proposed.Adding attention module to the network can improve the performance of the network,but it does not make good use of the component information and global context information between the sample and label data.To solve this problem,residual structure is added to the network to reduce gradient disappearance and network degradation;Adding wavelet transform and global context module can strengthen the learning of different levels of context and texture information in seismic data.At the same time,wavelet loss and data loss are used to train the network to achieve the purpose of efficient use of different information and improve the performance of the network.Thirdly,a method to improve seismic resolution based on dynamic convolution and shuffle convolution U-Net network is proposed.The improvement of U-Net network from the perspective of attention and wavelet domain has achieved good results,but the number of parameters and computation of the network is large,and it is difficult to deploy the model on edge devices or mobile devices.In order to solve this problem,in the down-sampling part,the double convolution operation is replaced by shuffle convolution,which aims to reduce the network parameters while ensuring the network performance;In the upper sampling part,the double convolution operation is replaced by a dynamic convolution separation,which reduces the amount of computation brought by the convolution layer,and the residual connection is added to enable the network to increase the learning of the input feature map.Finally,the goal of improving the speed of network reasoning is achieved.In this paper,the method of improving the resolution of seismic data is preliminarily explored.The research results show that the scheme of improving the resolution of seismic data through convolution neural network is feasible.This research not only enriches the theory of seismic data processing,but also expands the engineering practice of improving the resolution of seismic data.