Research On The Method Of Seismic Signal Enhancement In Generative Adversarial Networks

In today's society,human dependence on oil and gas resources is increasing.Exploration technology has also developed rapidly.Due to the influence of variable factors such as exploration technology and the natural environment,in the actual oil and gas exploration technology process,the normal sampling rate of the seismic data collected may cause a shift to imaging to produce spatial aliasing,and low-resolution seismic data may cause Inaccurate exploration.Therefore,in order to more easily and accurately carry out oil and gas exploration,while requiring high exploration technology,seismic signal enhancement methods also put forward higher requirements.In this paper,seismic signal enhancement is achieved through two aspects: seismic data trace interpolation and seismic data resolution enhancement.(1)The method of seismic trace interpolation based on generative adversarial networks is proposed and implementedTraditional seismic trace interpolation is generally based on complex mathematical transformations or certain assumptions,such as assuming that seismic data meets linearity or sparseness.In order to solve the problems existing in traditional seismic trace interpolation methods.This article will use the generative adversarial networks model in deep learning,while improving the network architecture and loss function in the model.The improved network architecture is to combine the generative adversarial networks and residual networks.The loss function is the sum of the Wasserstein distance and the interpolation error of the seismic trace as the loss function the analysis of the test results shows that the optimized generative adversarial networks can better realization of seismic trace interpolation,and suppress spatial aliasing.(2)The method of seismic data resolution enhancement based on generative adversarial networks is proposed and implementedIn order to better improve the resolution of seismic data,this paper will combine the generative adversarial networks with the U-Net network,and optimize the network architecture.The loss function consists of Wasserstein distance and L1 regularization.The optimized generative adversarial networks of the down-sampling portion can extract the characteristics of the seismic data,and the up-sampling portion increases the details of the seismic data characteristics while maintaining the data characteristics.The experimental results show that method proposed in this article can maintain the inherent characteristics of the seismic data and increase the high-frequency information in the seismic data to achieve the purpose of improving the resolution of the seismic data.In this paper,the proposed method and the traditional method are applied to the synthetic model data and the actual area seismic data respectively.Through experimental comparison and analysis,the proposed method can better realize the enhancement of seismic signal and provide theoretical support for accurate exploration.